Regulations (Preambles to Final Rules) - Table of Contents Regulations (Preambles to Final Rules) - Table of Contents
• Record Type: Occupational Exposure to Cadmium
• Section: 6
• Title: Section 6 - VI. Quantitative Risk Assessment

VI. Quantitative Risk Assessment

In February, 1990, OSHA proposed two alternative permissible exposure limits (PELs) of 1 ug/m(3) and 5 ug/m(3) as an 8-hour time-weighted average (TWA) and a 5 ug/m(3) and a 25 ug/m(3) ceiling limit for cadmium. These new exposure limits were based on an evaluation of carcinogenic and renal effects observed in workers or experimental animals following exposure to cadmium.

Summary of Quantitative Risk Assessment for Lung Cancer from Proposed Rule

To evaluate the potential carcinogenicity of cadmium, quantitative estimates of risk were made by OSHA based on the Takenaka et al. (Ex. 4-67) study in rats and the Thun et al. (Ex. 4-68) study in humans. These studies provide the strongest evidence of carcinogenicity of cadmium in animals and humans, as well as the most appropriate data for a quantitative assessment.

Experimental Data

Cadmium has been shown to be a carcinogen in animals following administration via inhalation. At the time of the proposed rule, the strongest evidence of carcinogenicity in animals came from a rat bioassay conducted by Takenaka et al. (Ex. 4-67). In this bioassay, Takenaka exposed three groups of 40 male rats 23 hours/day for 18 months to cadmium chloride aerosol at nominal cadmium concentrations of 12.5, 25, and 50 ug/m(3). An additional group of 41 male rats served as controls. The animals were observed for 13 months following exposure, at which time all surviving rats were sacrificed. A statistically significant increase in the incidence of malignant lung tumors was observed in treated animals and a statistically significant dose response relationship was observed. The incidence of lung carcinomas was 0/38 (0%) in the controls, 6/39 (15.4%) in the low-dose group, 20/38 (52.6%) in the mid-dose group, and 25/35 (71.4%) in the high-dose group. Table VI-1 contains a summary of the lung tumor data from this study.

OSHA concluded that the Takenaka study was particularly suitable for quantitative risk assessment for several reasons. First, the exposures were well documented. The study was run with concurrent controls, and a statistically significant excess of malignant neoplasms in the exposed rats and a statistically significant dose-response relationship were observed. Finally, the route of exposure used in this study (inhalation) is the primary exposure route in most occupational settings.


       Table VI-1 - Incidence of Lung Carcinomas in Male Wistar Rats
                    Exposed to Cadmium Chloride Aerosols(a)
____________________________________________________________________________
       Tumor Type            | Controls|12.5 ug/m(3)| 25 ug/m(3)|50 ug/m(3)
                             |(percent)| (percent)  |  (percent)|(percent)
_____________________________|_________|____________|___________|___________
Adenocarcinoma ............. | 0/38 (0)|   4/39 (10)| 15/38 (39)| 14/35 (40)
Epidermoid carcinoma ....... | 0/38 (0)|   2/39 (5) |  4/38 (11)|  7/35 (20)
Mucoepidermoid carcinoma ... | 0/38 (0)|   0/39 (0) |  0/38 (0) |  3/35 (9)
Combined epidermoid          |         |            |           |
carcinoma and adenocarcinoma | 0/38 (0)|   0/39 (0) |  1/38 (3) |  1/35 (3)
  Total carcinomas ......... | 0/38 (0)|   6/39 (15)| 20/38(53) | 25/35 (71)
_____________________________|_________|____________|___________|___________
  SOURCE: Ex. 18.
  Footnote(a) Number of animals with lung carcinoma from Takenaka et al.
(Ex. 4-67)

Quantitative Risk Assessment Using Animal Data

The extrapolation of carcinogenic risk across species rests on the assumption that when dose is measured in equivalent units for both species, then the risk associated with lifetime exposure to a substance, such as cadmium, is the same for each species at each dose. Exposure levels in rats were scaled to equivalent doses for rats and humans by expressing dose in units of micrograms per kilogram of body weight per day (ug/kg/day). No adjustments were made for particle size due to a lack of precise data available on the size of cadmium particles to which workers are exposed. Without specific human absorption data on various cadmium compounds, no adjustment was made for the solubility of cadmium chloride used in the experiment.

The probit, logit, Weibull, multistage, and one-hit models were all fit to the data. As indicated in the OSHA proposal, the preliminary Maximum Likelihood Estimates (MLEs) of excess cancer deaths following 45 years of occupational exposure to 5 ug Cd/m(3) were estimated to be 11 (multistage), 15 (one-hit), 0 (probit), 0.7 (logit), and 4 (Weibull) per 1,000 workers (Table VI-2). MLEs of excess cancer deaths following 45 years of occupational exposure to 100 ug Cd/m(3), the current time-weighted average PEL for cadmium fume, were estimated to be 221 (multistage), 266 (one-hit), 186 (probit), 190 (logit), and 210 (Weibull) per 1,000 workers. OSHA concluded that regardless of which model would be considered the "best," each model indicates that a reduction of the PEL to 5 ug/m(3) will lead to a significant reduction in risk. However, OSHA relied primarily upon the multistage model of carcinogenesis. OSHA agreed with the position of the Office of Science and Technology Policy (OSTP) (Ex. 8-693) in that when data and information are limited, and when much uncertainty exists regarding the mechanisms of carcinogenic action, models or procedures which incorporate low-dose linearity are preferred when compatible with limited information. The multistage model and the one-hit model, a special case of the multistage model, are linear at low doses. The multistage model incorporates the biological assumption that there are a number of stages that cell lines go through to produce a tumor, and a carcinogen exerts its effect by increasing the rate at which cell lines pass through one or more of these stages. In the form in which it was used by OSHA, the multistage model assumes there is no threshold of exposure below which a carcinogen cannot induce cancer. Upper confidence limits on the excess risk from low exposures determined using the multistage model are proportional to the amount of exposure, i.e., the dose response relationship is linear at low doses.


Table VI-2 - OSHA Preliminary Estimates of Excess Cancer Deaths per 1,000
             Workers With 45 Years Occupational Exposure to Cadmium(a),(b)
__________________________________________________________________________
    Dose (ug/m(3)) |Multistage|  One-hit |  Probit  |  Logit   | Weibull
                   |  Model   |   Model  |   Model  |  Model   |  Model
___________________|__________|__________|__________|__________|__________
                   |          |          |          |          |
200 .............. | 485 (528)| 461 (541)| 468 (552)| 469 (557)| 445 (528)
100 .............. | 221 (312)| 266 (322)| 186 (278)| 190 (280)| 210 (299)
50 ............... | 109 (171)| 143 (177)|   44 (99)|  58 (114)|  90 (157)
40 ............... |  87 (139)| 116 (144)|   25 (64)|   39 (82)|  68 (126)
20 ............... |   43 (72)|   60 (75)|    2 (11)|   11 (28)|   28 (60)
10 ............... |   21 (37)|   30 (38)|   0.2 (1)|     3 (9)|   11 (28)
5 ................ |   11 (18)|   15 (19)|     0 (0)|   0.7 (3)|    4 (13)
1 ................ |   2 (3.7)|     3 (4)|     0 (0)|   0 (0.2)|   0.5 (2)
X(b) ............. |      3.00|      3.63|      1.52|      1.59|      2.58
Degrees of Freedom |         2|         3|         1|         1|         1
P-value .......... |     >0.25|     < 0.50|      0.22|      0.21|      0.11
___________________|__________|__________|__________|__________|__________
  SOURCE: Ex. 18
  Footnote(a) Estimates derived using data from the Takenaka rat bioassay.
  Footnote(b) Numbers in parentheses are the 95% upper confidence limits.

Epidemiological Data

Human data for quantifying lung cancer risk associated with cadmium exposure was found in a mortality study conducted by Thun et al. (Ex. 4-68). This study is a historical prospective study of 602 white men employed in a production area of a cadmium smelter for at least six months between 1940 and 1969. Follow up was through 1978. Sufficient exposure data existed to determine exposure levels for workers and a dose-response relationship between cadmium exposure and lung cancer mortality was reported.

Prior to 1926, the cadmium smelter functioned as an arsenic smelter. Because arsenic is a known carcinogen resulting in lung cancer, the cohort was divided into two subcohorts for analysis: a sub-cohort of 26 men hired prior to 1926 and a sub-cohort or 576 workers hired after 1926. A statistically significantly elevated incidence of death due to lung cancer (4 observed versus 0.56 expected) was found in the sub-cohort of workers hired prior to 1926. Among the workers hired after 1926, 16 lung cancers were observed (versus 10.87 expected). Table VI-3 contains a summary of the lung cancer data from Thun et al. (Ex. 4-68).


TABLE VI-3. - DATA USED FOR ESTIMATING RISKS FROM A MORTALITY STUDY OF
               CADMIUM SMELTER WORKERS BY THUN ET AL.
_________________________________________________________________________
                    |              | No. lung  | No. lung  |
Cumulative Exposure | Person years | cancers   |  cancers  |   SMR
  (mg/m(3)-days)    |  at risk     | observed  |expected(a)|
___________________________________|___________|___________|_____________
                    |              |           |           |
<584................|       7005   |       2   |    3.77   |      53
585-2920............|       5825   |       7   |    4.61   |     152
>2921...............|       2214   |       7   |    2.50   |     280
____________________|______________|___________|___________|_____________
                    | TWA Equivalent (ug/m(3)) |  Median   | Continuous
Cumulative Exposure |__________________________|  dose (d) |  dose (e)
 (mg/m(3)-days      |              |           | (mg/m(3)  | (ug/m(3)
                    | 40-year(b)   | 45-year(c)|  -days)   |  -years)
____________________|______________|___________|___________|_____________
                    |              |           |           |
<584................|         >40  |       >36 |     280   |     168
585-2920............|      40-200  |    36-178 |    1210   |     727
>2921...............|        >200  |      >178 |    4200   |    2522
____________________|______________|___________|___________|_____________
  Source: Ex. 18.
  Footnote(a) Expected incidence based on calendar time, age-specific
death rates for U.S. white males.
  Footnote(b) Calculated as (cumulative dose X 1000)/(365 X 40).
  Footnote(c) Calculated as (cumulative dose X 1000)/(365 X 45).
  Footnote(d) As provided by Thun to EPA.
  Footnote(e) Calculated as median dose X 1000 X (8/24) X (1/365) X
(240/365).

Airborne cadmium concentrations were measured by Smith et al. (Ex. 4-64) as 8-hour TWA for nine departments in the smelter and for office and laboratories combined (non-production work areas). Exposure for each of these areas was classified as high exposure or low exposure. Estimates of individual cumulative cadmium exposure were based on the period of time a worker was employed in a high or low exposure dose area.

Using this exposure information, Thun et al. divided the post-1926 cohort into three exposure groups: a low-dose group with a cumulative exposure of less than 584 mg/m(3)-days, a mid-dose group with cumulative exposures of 585-2920 mg/m(3)-days, and a high-dose group with cumulative exposures of greater than 2921 mg/m(3)-days. Based on this division, a dose-response relationship between cadmium and death from lung cancer was observed. For the low-dose group, two deaths due to lung cancer were observed while 3.77 were expected (RR=0.53). For the mid-dose group, seven deaths due to lung cancer were observed while 4.61 were expected (RR=1.52). For the high-dose group, seven deaths due to lung cancer were observed while 2.50 were expected (RR=2.80).

Quantitative Risk Assessment Based on Epidemiological Data

OSHA quantified risks from the Thun data using an absolute risk model and a relative risk model. Both are linear models; however, the two models are based on different assumptions which lead to different estimates of risk. The absolute risk model is based on the assumption that the increased lung cancer mortality rate from cadmium exposure depends only upon cadmium exposure and not upon age. The relative risk model rests on the assumption that the ratio of the lung cancer mortality risk in an individual exposed to cadmium to what his mortality risk would be if he had not been exposed to cadmium depends only upon cadmium exposure and not upon age. Thus, an absolute risk model would predict the same lung cancer mortality rate from cadmium exposure in 20-year olds as in 50-year olds, given equal cadmium exposures, whereas the relative risk model would predict a higher rate from cadmium exposure in 50-year olds because 50-year olds have a higher background rate of lung cancer than 20-year olds.

The estimates of excess lung cancer death from the relative risk model were approximately twice as large as those from the absolute risk model, but both models predicted significant risk at the current OSHA PEL of 100 ug/m(3). As published in the proposal, at 100 ug/m(3), these models predicted between 16 and 30 excess lung cancer deaths per 1000 exposed workers. At exposure levels as low as 5 ug/m(3), the excess risk of lung cancer death estimated by these models was 0.8 per 1000 exposed workers for the absolute risk model and 1.6 per 1000 exposed workers for the relative risk model. At exposure levels of 1 ug/m(3) the excess risk of lung cancer death estimated by these models was 0.2 per 1000 exposed workers for the absolute risk model and 0.3 per 1000 exposed workers for the relative risk model. (See Table VI-4.)


TABLE VI-4. - OSHA PRELIMINARY ESTIMATES OF EXCESS LUNG CANCER DEATHS
  PER 1,000 WORKERS WITH 45 YEARS OCCUPATIONAL EXPOSURE TO CADMIUM (a,b)
___________________________________________________________________________
        Dose (ug/m(3))                      |  Number of excess deaths
____________________________________________|______________________________
                             | Continuous(c)| Absolute risk | Relative risk
            TWA              |              |    model      |    model
_____________________________|______________|_______________|______________
                             |              |               |
200..........................|        4.384 |  32.3 (4,60)  |  60.2 (8,109)
100..........................|        2.192 |  16.3 (2,30)  |   30.7 (4,57)
50...........................|        1.096 |   8.2 (1,15)  |   15.5 (2,29)
40...........................|        0.877 | 6.6 (0.7,12)  |   12.5 (2,23)
20...........................|        0.438 |  3.3 (0.4,6)  |  6.3 (0.8,12)
10...........................|        0.219 |  1.6 (0.2,3)  |   3.1 (0.4,6)
5............................|        0.110 |  0.8 (0.1,2)  |   1.6 (0.2,3)
1............................|        0.022 |  0.2 (0.0,3)  |   0.3 (0.0,6)
_____________________________|______________|_______________|_______________
  SOURCE: Ex. 18.
  Footnote(a) Estimates derived using data from the Thun mortality study of
cadmium smelter workers.
  Footnote(b) Numbers in parentheses are 5% lower and 95% upper confidence
limits.
  Footnote(c) Assumes exposure occurs for 8/24 hours and 240/365 days.

OSHA proposed a PEL of 1 ug/m(3) based on the risk assessment using the Takenaka rat bioassay. However, there is support for the use of the Thun et al. data for establishing an exposure level because no extrapolation across species is required. Estimates of risk based on the Thun data are lower than those from the Takenaka data, and support a PEL of 5 ug/m(3). OSHA therefore proposed alternated PELs of 1 ug/m(3) and 5 ug/m(3) based in part on the estimates of risk from Takenaka et al. (Ex. 4-67) and Thun et al. (Ex. 4-68) and in part upon the concerns for the technological feasibility of achieving a PEL of 1 ug/m(3).

Summary of Quantitative Risk Assessment for Kidney Dysfunction from Proposed Rule

In its proposal, OSHA quantified the risk of kidney dysfunction associated with cadmium exposure using two studies of cadmium workers that contained adequate data for such an assessment. One of the studies was a study of cadmium smelter workers conducted by Ellis et al. (Ex. 4-27). The other study was of workers at a refrigeration compressor production plant conducted by Falck et al. (Ex. 4-28). Kidney dysfunction was defined in both studies as an excess of urine protein, specifically B(2)-microglobulin, a low molecular weight protein that, when found in the urine, indicates that damage has occurred to the proximal tubules and/or glomerulus. Because this damage may be irreversible and can lead to more serious health effects, OSHA considers this dysfunction to represent material impairment of health.

Ellis et al. studied 82 workers at the same smelter that was investigated by Thun et al. (Ex. 4-68; see discussion above in summary of cancer QRA). Air concentrations of cadmium in a given work area were estimated using industrial hygiene data provided by Smith (Ex. 4-64). A cumulative exposure was estimated for each worker based on these air concentrations and duration of work in a given work area. Twenty-four hour urine samples were obtained from each worker and used to determine whether a worker had abnormal kidney function. Kidney function was judged to be abnormal if urinary levels of B(2)-microglobulin exceeded 200 ug/g creatinine or if total urinary protein levels exceeded 250 mg/g creatinine. Eighteen out of 51 active workers and 23 out of 31 retired workers were classified as having abnormal kidney function. The mean time-weighted inhalation exposure estimate for active workers with abnormal kidney function was 1690 ug/m(3)-years and for retired workers with abnormal kidney function was 3143 ug/m(3)-years.

Falck et al. studied 33 male workers at a plant which produces refrigeration compressors with silver brazed copper fittings. The silver brazing consisted of 18-24% cadmium. Estimates of cumulative exposure were made for each worker based on data from air monitoring done by the Michigan Department of Industrial Health. The mean estimated cadmium exposures were 39 ug/m(3) for 11 years of operation for the automated brazing line and 110 ug/m(3) for 21 years of operation for the manual line. A time-weighted exposure was estimated for each worker based on the length of time in each brazing line. Twenty-four hour urine samples were obtained for eight of the 33 workers who were found to have elevated protein levels in spot urine samples. Seven of these eight workers were found to have urinary protein levels in excess of the 95% tolerance limit, based on urinary protein levels in 41 unexposed workers who served as controls. Based on this increase in urinary protein in exposed workers compared to unexposed controls, these seven workers were judged to have abnormal kidney function.

Using the logistic regression model and data provided by Ellis et al. and Falck et al., OSHA estimated the risk of kidney dysfunction from 45 years of exposure to a variety of occupational doses of cadmium. Estimates of kidney dysfunction per 1,000 workers with 45 years of occupational exposure to 5 ug/m(3) were 164.6 using the Ellis et al. data and 9.0 using the Falck et al. data. (See Table VI-5.)


TABLE VI-5. - OSHA PRELIMINARY ESTIMATES OF KIDNEY DYSFUNCTRION PER
  1,000 WORKERS WITH 45 YEARS OF OCCUPATIONAL EXPOSURE TO CADMIUM
_________________________________________________________________________
  8-Hour      | Cumulative |       Incidence of kidney dysfunction
 TWA Dose     | dose (ug/  |_____________________________________________
 (ug/m(3))    |  m(3)-yrs) |    Ellis model    |     Falck model
______________|____________|___________________|_________________________
              |            |                   |
1.............|        45  |            26.1   |         0.1
5.............|       225  |           164.6   |         9.0
10............|       450  |           317.7   |        58.9
20............|       900  |           523.7   |       300.5
40............|      1800  |           722.0   |       746.7
50............|      2250  |           744.0   |       845.7
100...........|      4500  |           890.0   |       974.1
______________|____________|___________________|_________________________
  Source: Ex. 18.

New Evidence and Issues Arising Since Publication to OSHA'S Proposed Rule

Since the quantitative risk assessment was conducted for the proposed rule for cadmium, new information has become available which prompted a re-evaluation of the risk assessment conducted by OSHA. Additional animal studies have been conducted that demonstrate an increased risk of lung cancer in animals following exposure to cadmium (Exs. 8-694B, 8-694D, L-140-29F). The only animal study available for the risk assessment in the proposed rule which provided quantitative information concerning the carcinogenicity of cadmium was the study conducted by Takenaka et al. (Ex. 4-67). This study was conducted in male rats only and exposure was to cadmium chloride. Preliminary results from a study conducted by Oldiges et al. (Exs. 12-10i, 12-10h, and 12-35) were available and provided qualitative evidence that supported the carcinogenicity of cadmium. However, because these were preliminary results, they were not evaluated quantitatively. Since the proposed rule, the final results of the Oldiges et al. study (Ex. 8-694D) have been published, as well as a later report of this same study (Glaser et al. Ex. 8-694B). In Oldiges et al. and Glaser et al., groups of male and female Wistar rats were exposed to various cadmium compounds. These studies demonstrated that various cadmium compounds caused lung tumors in male and female rats.

The followup of the major epidemiological study available at the time of the proposed rule has been extended (Ex. 4-68). The Thun et al. cohort reported in the proposed rule was followed up through 1978. Since that time the cohort has been followed through 1984.

Since the proposed rule was issued, Stayner et al. (Ex. L-140-20) have completed a risk assessment based on the Thun cohort that differs in several respects from the one reported in the proposed rule. The Stayner et al. risk assessment is based on a followup of the Thun cohort through 1984, whereas the risk assessment reported in the proposed rule was based on followup only through 1978. Stayner et al. had access to the unprocessed data from the Thun cohort and consequently was able to conduct a wider range of analyses in its risk assessment than was possible based on just the data reported in the published Thun report (Ex. 4-68). In addition, Lamm et al. (Ex. 144-7-b) have conducted a case control analysis of lung cancer cases from the same cohort as that studied by Thun et al.

Several studies concerning the kidney effects observed in workers following exposure to cadmium were not evaluated quantitatively in the proposed rule, and these studies provide quantitative information that would be useful in evaluating the risk of kidney effects following exposure to cadmium. In its proposed rule, OSHA quantified the risk of kidney dysfunction due to cadmium exposure using only the studies by Falck et al. (Ex. 4-28) and Ellis et al. (Ex. 4-27). OSHA has since identified four additional studies that may contain useful quantitative information. These include a study by Elinder et al. (Ex. L-140-45) of a cohort of 60 workers who had previously been exposed to cadmium through welder fume and dust associated with the use of cadmium soldiers, a study by Jarup et al. (Ex. 8-661) of 440 workers exposed to cadmium at a Swedish battery plant, a study by Mason et al. (Ex. 8-669A) of 75 workers exposed to copper-cadmium alloy in a factory in the United Kingdom, and a study by Thun et al., which is based on the same population of smelter workers studied earlier by Ellis et al. (Ex. 4-27).

Several issues concerning the quantitative risk assessment for cancer in the proposed rule have been raised that also prompted a reevaluation of OSHA's risk assessment. Concerns were raised over the appropriateness of the Takenaka et al. study for quantitative risk assessment by the Office of Management and Budget (OMB) (Ex. 17), Dr. Oberdorster (Ex. 31), Richard Bidstrup, Counsel for SCM Chemicals, Inc. (Ex. 19-42A), as well as others. The main issue was that in the Takenaka et al. study the rats were exposed continuously to cadmium chloride, while in the workplace exposure is not continuous and is mostly to cadmium oxide (Ex. 19-43). Therefore, the Takenaka et al. study was reevaluated to determine its relevance for the quantitative risk assessment of cadmium.

Mr. Leonard Ulicny (Ex. 144-17) and the Dry Color Manufacturers Association (Ex. 144-20) have requested that cadmium sulfide should have a separate PEL from other cadmium compounds. This request is based on the view that cadmium sulfide is less soluble in the human body, and therefore less toxic, than other cadmium compounds. Both commenters noted that the preparation of cadmium sulfide in the studies conducted by Glaser et al. (Ex. 8-694B) and Oldiges et al. (Ex. 8-694D) consisted of cadmium sulfide particles in suspension and subjected to light. The commenters concluded that this reaction could be responsible for the effects observed in animals following exposure to cadmium sulfide which they believe should be attributed to another cadmium compound, namely cadmium sulfate. Analyses have been conducted and reported by Dr. Oberdorster (Ex. 141) and Dr. Heinrich (Ex. 142) which need to be evaluated to determine if cadmium sulfide should have a separate PEL.

OMB (Ex. 17) questioned the impact of the Heinrich et al. (Ex. L-140-29F) study on the relevance of the quantitative risk assessment. The Heinrich et al. study was conducted in mice and hamsters. OMB thought that this study provided negative data and could possibly show that the carcinogenic effect of cadmium may be species-specific, because this study provided the only relevant data in species other than rat. This study was reevaluated to determine if the study could actually be considered negative and its impact on the quantitative risk assessment of cadmium.

In the quantitative risk assessment for cancer using animal bioassay data, OSHA performed its interspecies extrapolation assuming the risks were equal across species when dose across species was equal on a body weight basis. Dr. Oberdorster (Ex. 31) has recommended the use of lung dosimetry instead of equivalency based on body weight. The methodology recommended by Dr. Oberdorster will be considered in the reevaluation of OSHA's risk assessment.

The appropriateness of the Thun et al. cohort for quantitative risk assessment was questioned by many commenters (Exs. 38; 19-43; L-140-23; 144-8a; 144-8b; 114-8c; 114-17). The Globe plant from which the cohort was taken was formerly an arsenic smelter and many commenters, including George M. Obelodobel, Vice-President and General Manager of Big River Zinc Corporation (Ex. 19-30), Richard Bidstrup, Counsel for SCM Chemicals, Inc. (Ex. 19-42A), OMB (Ex. 17) and The Cadmium Council (Ex. 119), thought that arsenic, a known human carcinogen, may have been a contributing factor to the lung cancer observed in this cohort. Thun et al. had conducted ananalysis of the contribution of arsenic to the risk of lung cancer observed in their study. However, this analysis was questioned (Ex. 17). Cigarette smoking was also mentioned by several of these reviewers as a possible confounding factor. These issues were considered in the reevaluation of the Thun et al. cohort.

The appropriateness of the mathematical dose-response models used in the quantitative risk assessments was raised as an issue by several commenters. In particular, the use of the linearized multistage model as the most appropriate model for the animal cancer quantitative risk assessment was raised by OMB (Ex.17), as well as the use of the absolute and relative risk models for the quantitative risk assessment based on human epidemiological data (Exs. 17, 38). The risks from the animal and human quantitative risk assessments for cancer were also compared when these risk assessments were reevaluated. For the quantitative risk assessment of the kidney effects of cadmium, several commentors, including The Cadmium Council (Exs. 119, 19-43) recommended the use of the probit model for evaluating the kidney effects of cadmium rather than the logistic model used by OSHA.

In view of the new evidence that has become available since the proposed rule and the comments that have been received concerning the proposed rule, OSHA has decided to reevaluate the quantitative risk assessments for cadmium. The reevaluation of these quantitative risk assessments, as well as a discussion of the major issues are included below.

Risk Assessment for Lung Cancer

Assessment of Lung Cancer Risk Using Animal Data

The inhalation study conducted by Takenaka et al. (Ex. 4-67) exposed male rats to cadmium chloride, while Oldiges et al. (Ex. 8-694D) compared the results following exposure to cadmium chloride, cadmium oxide (dust or fume), cadmium sulfate, and cadmium sulfide in both male and female rats. In the Takenaka study, groups of 40 (41 in the control) male Wistar rats were exposed to a cadmium chloride aerosol continuously (23 hr/d, 7 d/wk) for 18 months at nominal concentrations of 0, 12.5, 25, or 50 ug/m(3) of cadmium. Rats were observed up to 31 months; necropsy was performed only on animals that had survived at least 18 months. In the Oldiges et al. study (Ex. 8-694D), groups of 20 male and female Wistar rats were exposed to cadmium chloride (30 or 90 ug/m(3)), cadmium oxide dust (30 or 90 ug/m(3)), cadmium oxide fumes (10 or 30 ug/m(3)), cadmium sulfate (90 ug/m(3)), or cadmium sulfide (90, 270, 810, or 2430 ug/m(3)) for approximately 18 months and followed for up to approximately 31 months.

In its reassessment of cancer risk based on animal data contained in the Takenaka et al. (Ex. 4-67) and Oldiges et al. (Ex. 4-694D) studies, OSHA has utilized three dose response models, all of which are different versions of the multistage model of cancer: the Armitage-Doll multistage model, the multistage model, and the multistage-Weibull model. The Armitage-Doll multistage model of cancer assumes that individual cell lines go through a series of stages in order to initiate a tumor, and different cell lines compete independently to be the first to produce a tumor. The rate at which cell lines go through a particular stage is assumed to be increased by exposure to the carcinogen by an amount that is proportional to the instantaneous dose of the carcinogen. This implementation of the Armitage-Doll theory was proposed for risk assessment by Crump and Howe (1984). This model is fit to time-to-tumor data. It does not require a TWA measure of dose as input, but instead utilizes the full time-varying pattern of dose in assessing risk.

The multistage model (Crump, 1984) is a generalization of the Armitage-Doll model that provides a dose-response that is flexible enough to model both linear and non-linear responses. However, upper statistical confidence limits on risk computed using this model will vary linearly with dose (hence, this model is sometimes called the linearized multistage model). This model is applied to quantal data on the number of animals with tumors; it does not use information on the time required for tumors to appear.

The multistage-Weibull model (Krewski et al., 1983) is an extension of the linearized multistage model that is applicable to time-to-tumor data. This model assumes that the probability of a tumor as a function of dose, for a fixed age, has the same form as the (linearized) multistage model, and the probability of tumor as a function of age, for a fixed dose, has a Weibull distribution.

Both the Armitage-Doll model and the multistage-Weigull model require information on the time at which tumors were discovered (time-to-tumor data), whereas the multistage model does not require this information. Since time-to-tumor data were not available from Takenaka et al., the Armitage-Doll and multistage-Weibull models were applied only to data from Oldiges et al. These time to tumor data were obtained by OSHA from an unpublished report from the Oldiges et al. study (Ex. 73). Both of the models that utilize time-to-tumor data require information on whether or not a tumor was fatal or incidental (i.e., observed incidentally at death from a different cause). OSHA did not have information regarding whether the lung tumors in the Takenaka et al. or Oldiges et al. studies were fatal or incidental, and consequently applied the models using both assumptions. These two approaches were found to give comparable results, and only results based on the assumption that all lung tumors were incidental are reported herein. These two approaches estimate different end points. When all tumors are assumed to be incidental, these models estimate the probability of having a tumor large enough to be observed in a histological examination by a given age; however, if all tumors are assumed to be fatal, the models estimate the probability of dying from a tumor by a given age.

Both the Armitage-Doll and multistage-Weibull require that an animal's age be specified to estimate risk. (These models estimate the probability of an animal acquiring a tumor by a particular age.) The age used to estimate risk from each study of a particular cadmium compound by Oldiges et al. is the duration of the study (days on test of the last animal to die).

The Oldiges et al. study involved groups in which animals were dosed for varying periods of time and the lifespans of the animals in some of the groups were reduced, apparently as a result of cadmium toxicity. Since the multistage model does not take into account reduced lifespans, this model was only fit to data from groups in which exposures lasted for at least 14 months (Reduced survival appeared to be less of a problem in such groups.) With this method of fitting, the multistage model gave an acceptable fit (based on a chi-square goodness-of-fit test) to all of the data sets.

The Armitage-Doll and multistage-Weibull models were fit to all of the data on each cadmium compound. However, if these models did not adequately fit the data, data at the highest dose level was removed from the model fit. Data related to the highest dose level may be the least relevant to dose response modelling when there is a response at lower dose levels because of high or similar tumor responses at lower dose levels, reduced survival, or the possibility of altered metabolism including saturation of major or primary metabolic pathways at the highest dose levels. Removing data from the highest dose level resulted in reducing the number of dose groups for male rats exposed to CdO dust and for both male and female rats exposed to CdS when applying the multistage-Weibull model. However, the Armitage-Doll model gave an adequate fit to all of the data sets without any dose groups omitted.

The Armitage-Doll model makes use of the exact time-varying pattern of exposure. The dose of cadmium (in units of ug/kg/day) applied to this model during a period in which exposure was occurring was calculated as


DOSE = CC*IR*FD/W,

where

CC [ug/m(3)]  = airborne cadmium concentration;
IR [m(3)/day] = volume of air inhaled per day (assumed to be
                0.254 m(3)/day for male rats and 0.223 m(3)/day
                for female rats);
FD = fraction of day exposed;
W [kg] = average weight of rats at 18 months (assumed to be 0.43 kg for
         males and 0.35 kg for females).

A dose of zero was applied during a period in which the animals were not exposed.

The multistage and multistage-Weibull models require that a single summary dose be applied that represents the average daily dose, including periods in which animals were not exposed. This adjusted dose was calculated from the unadjusted dose as follows:


ADJDOSE = DOSE*ME/MO,

where

ME = number of months rats were exposed;
MO = number of months rats were observed (estimated as the number of days
     on test of the last rat to die in an experiment).

To estimate human risk from these models, a human exposure is calculated in units of ug/kg/day and applied to the dose response model estimated from the animal data. Human exposures were assumed to be occupational and to last from age 20 to age 65. For the Armitage-Doll model, the corresponding exposure period, expressed in terms of the life of a rat, was from [20/74*MO] to [65/74*MO] days, where MO is defined above. The average daily dose (in units of ug/kg/day) during the exposure period was calculated as


ADD = HCC*HIR/HW*[DW/365],

where

HCC [ug/m(3)] = assumed human airborne cadmium concentration;
HIR = volume of air inhaled per 8 hour shift (assumed to be 10 m(3)/day);
HW = human body weight (assumed to be 70 kg);
DW = days worker per year (assumed to be 250).

To calculate the corresponding average daily human exposure, averaged over the entire lifespan, which is required to estimate human risk from the multistage and multistage-Weibull models, ADD was adjusted as follows:


ADJADD = ADD*45/74,

where 45 is the number of years of work and 74 is the assumed human
lifespan.

!ht!/per!ht!

  Table VI-6 contains the results of applying these three models to ten
data sets from the Takenaka et al. and Oldiges et al. studies involving
male or female rats exposed to five different types of cadmium.  This
table presents estimates of excess lung cancer deaths per 1000 workers
having 45 years of occupational exposure to TWA cadmium exposures of 1, 5,
10, or 100 ug/m(3).  Upper and lower 95% confidence limits for the
expected number of excess deaths are presented for the multistage model,
but only upper 95% confidence limits are presented for the other two
models because of the lack of a computer program to calculate lower limits
for these models.

     Table VI-6 - Estimates Derived From Annual Data of Excess
                  Cancer Deaths Per 1000 Workers With 45 Years
                  of Occupational Exposure to Cadmium
!ht!fig PRcm6_t6 Table VI-6!/ht!
  (For Table VI-6, see printed copy)

The maximum likelihood estimates (MLEs) based on the Takenaka et al. (Ex. 4-67) from OSHA's reassessment are slightly different than the estimates reported in the proposed rule. The estimates of excess cancer risk based on the Takenaka et al. data were 10.6 per 1,000 workers following 45 years of occupational exposure to 5 ug/m(3) in the proposed rule, but in the new assessment, this risk has changed slightly to 15 per 1,000 workers. This difference is due to a change in the method used to convert the animal dose in ug/m(3) to ug/kg/day in the Takenaka et al. rat bioassay. In the proposed rule, it was assumed that the average survival for the animals in the study was two years; therefore, rats were exposed 23 hours/day for 75% of their lifespan (18 months). To adjust the experimental dose for less than lifetime exposure, the experimental dose was multiplied by 0.75 to produce an equivalent lifetime dose. However, in the Takenaka et al. study, animals were observed for up to 31 months, with 50% of the animals in the control group surviving 30 months. Normally, to adjust for less than lifetime exposure, the average daily exposure is prorated over the lifetime of the animal (51 FR 33992; Sept. 24, 1986). This is done by taking a ratio of the number of months an animal is exposed to the number of months in the lifespan of the animal. For the Takenaka et al. study the lifespan is assumed to be 31 months. Therefore, in the reassessment based on this study a factor of 18 months/31 months (58%) rather than 18 months/24 months (75%) was used, resulting in slightly different estimates of dose and of risk.

The multistage model and Armitage-Doll models gave an acceptable fit to all of the data sets (e.g., p > 0.05, based on a chi-square goodness-of-fit test). (However, recall that the multistage model was only fit to data in which exposures lasted 14 months or longer, whereas the other two models were fit to all of the data.) The multistage-Weibull model also gave an acceptable fit to all of the data sets; however, to achieve an acceptable fit with this model, the data from the highest dose was deleted in the analyses involving exposure of males and females to CdS and exposure of males to CdO dust. (Also, in the multistage model to the data on exposure of male rats to CdO dust, the average exposures in the two experimental groups differed by only about 5%, and these two groups were therefore combined into a single exposure group for model fitting.) The estimates of excess risk obtained from these models were consistently lower based on exposure to CdO fume than to other forms of cadmium. Dr. Oberdorster also noted the lower response of lung tumors in the Oldiges et al. study after exposure to CdO fume and concluded that this observation could most likely be explained by a lower lung burden of cadmium that resulted from a lower deposition fraction of the inhaled fume particles (Ex. 31). Estimates of risk resulting from other types of exposure agree much more closely. With the exception of CdO fume exposures, the ranges of estimates for the excess lung cancer risk from 45 years of occupational exposure to 5 ug/m(3) lifetime are as follows: multistage model, 5.5-28 excess deaths per 1000 workers; Armitage-Doll model, 2.5-63 excess deaths per 1000 workers; multistage-Weibull model, 1.2-35 excess deaths per 1000 workers. If exposure is to a TWA of 100 ug/m(3), the corresponding ranges are: multistage model, 104-433 excess deaths per 1000 workers; Armitage-Doll model, 49-726 excess deaths per 1000 workers; multistage-Weibull model, 104-512 excess deaths per 1000 workers.

Discussion of Issues Related to the Risk Assessment for Lung Cancer using Animal Data

Weight-of-Evidence Provided by the Takenaka Study

A weight-of-evidence evaluation is the first step in determining the likelihood that a chemical is a human carcinogen. According to EPA's methodology, the evidence is characterized separately for human studies and animal studies as sufficient, limited, inadequate, no data, or evidence of no effect. The characterizations of the animal and human data are combined, and based on the extent to which the chemical has been shown to be a carcinogen, the chemical is given a provisional weight-of-evidence classification. This classification can then be adjusted up or down based on other supporting evidence such as mutagenicity data (USEPA, 1989).

In EPA's Guidelines for Carcinogen Risk Assessment (51 FR 33992; Sept. 24, 1986), it is stated:

"The weight of evidence that an agent is potentially carcinogenic for humans increases (1) with the increase in number of tissue sites affected by the agent; (2) with the increase in number of animal species, strains, sexes, and number of experiments and doses showing a carcinogenic response; (3) with the occurrence of clear-cut dose-response relationships as well as a high level of statistical significance of the increased tumor incidence in treated compared to control groups; (4) when there is a dose-related shortening of the time-to-tumor occurrence or time to death from tumor; and (5) when there is a dose-related increase in the proportion of tumors that are malignant."

OSHA believes that these guidelines for a weight-of-evidence are not meant to be used in a pass-fail approach, since EPA refers to increases in the weight of evidence. Thus, all five conditions or a majority of conditions need not be satisfied for a chemical to be considered carcinogenic. Most especially all five do not have to be satisfied in any single study. Rather, these guidelines, and other sets of evaluation criteria, such as the EPA or IARC classification schemes, are meant to apply in a weight-of-evidence evaluation using the overall experimental and epidemiological data.

When the data base for cadmium is evaluated, the overall weight-of-evidence for the carcinogenicity in animals is strong, based on the above mentioned conditions. Statistically significant increases in the incidence of lung tumors were noted in male rats (Ex. 4-67, 8-694D) and female rats (Ex. 8-694D), exposed to cadmium compounds by the inhalation route, and when an adjustment for survival was made by life-table analysis a significant increase in lung tumors was also observed in female mice (Ex. L-140-29F) (see discussion below). Significant increases were noted in mammary fibroadenomas in rats given cadmium intratracheal instillation (Sanders and Mahaffey, Ex. 4-61) and in sarcomas (injection site) in rats injected with cadmium (Levy et al., Ex. 8-194; Kazantis, Ex. 8-576; Haddow et al., Ex. 4-34; Levy et al., Ex. 8-117). In the Takenaka et al. and Oldiges et al. studies, statistically significant dose-response relationships were evident, and the proportion of malignant tumors increased and the latency time decreased with increased dose. (The controls did not have any lung tumors in either the Takenaka et al. or Oldiges et al. studies.) The National Toxicology Program (NTP) (1984) in its Guidelines on Chemical Carcinogenesis Testing and Evaluation, has reported:

"Clear Evidence of Carcinogenicity is demonstrated in studies that are interpreted as showing a chemically related increased incidence of malignant neoplasms, studies that exhibit a substantially increased incidence of benign neoplasms, or studies that exhibit an increased incidence of a combination of malignant and benign neoplasms where each increases with dose."

Based on NTP's definition of clear evidence of carcinogenicity, the Takenaka et al. and Oldiges et al. studies provide clear evidence of the carcinogenicity of cadmium in animals.

Evidence of Carcinogenicity of Cadmium in Other Species

Cadmium, when administered by intratracheal instillation or injection, produced statistically significant increases in certain tumors but not lung tumors (Exs. 4-34, 8-576, and 8-117). Cadmium when given by the oral route, either in drinking water (Exs. 8-308, 8-121, 8-196), by gastric instillation (Levy et al., Exs. 8-034, 8-117), or in the diet (Loser, Ex. 8-643) did not produce an increase in either the total number of tumors or an increase in any specific type of tumor. The data show that cadmium, which is carcinogenic by the inhalation route in rats, has not been demonstrated to be carcinogenic by the oral route. This is consistent with the route-specific patterns for other heavy metals, such as chromium or nickel, that are established human carcinogens by the inhalation route but not by the oral route. However, lung tumors in workers occupationally exposed by the inhalation route is the major concern. Therefore, for the proposed rule, the inhalation studies are the most relevant when assessing the risk due to cadmium exposure.

According to the Risk Assessment Guidelines published by the Office of Science and Technology Policy (OSTP, 1985), negative data as well as positive data should be considered in a weight of evidence determination of the carcinogenicity of a compound. As importantly, the selection of data for analysis should maximize any correlations between animals and humans with regard to pharmacokinetic considerations and mechanism of action. Inhalation studies have been conducted in male and female rats, mice, and hamsters. The studies in rats have convincingly demonstrated inhalation exposure to cadmium results in the production of lung tumors. The study in mice was less convincing, while that conducted in hamsters was labelled as negative by some. However, the mouse and hamster studies conducted by Heinrich et al. (Ex. L-140-29F) had several limitations.

In the mouse study, many of the animals were treated for a short duration of time with treatment duration ranging from 6 to 69 weeks. Survival problems were also reported in treated animals versus controls. Nine out of fourteen of the cadmium oxide exposed groups had significant shortening of mean survival time, based on life-table analysis. This shortening of lifespan was attributed to toxic effects in the respiratory tract. Of the remaining five treatment groups, three had significantly increased incidences of lung tumors. Due to the shortened survival of many of the treated mice, animals may not have survived long enough for some tumors to be observed. When a life-table analysis was conducted, which adjusts for survival, the probability of an animal dying with a lung tumor was statistically significantly greater in treated groups versus controls in most of the CdO treated groups and one of the CdS groups (90 ug/m(3)). No information was reported concerning the time to first tumor or the type of lung tumors observed. Therefore, the latency period for tumor development in mice following exposure to cadmium cannot be determined.

As in the mouse study, exposure of hamsters to cadmium compounds were for short durations, ranging from 13 to 65 weeks. Shortened survival was also observed in some treated groups resulting from toxic effects to the respiratory tract. Survival problems may have precluded the development of lung tumors. However, dose-dependent statistically significant increases in the incidence of bronchiolar-alveolar hyperplasia and proliferation of connective tissue, which are considered preneoplastic lesions and may indicate progression to cancer, were found with all cadmium compounds tested. Heinrich et al. reported that a few of the animals developed respiratory tract tumors; however, no tumor incidence data were reported. Although the results appear to indicate a progression to carcinogenicity, no definitive conclusions can be drawn as to the carcinogenicity of cadmium in hamsters.

These studies conducted by Heinrich et al. (Ex. L-140-29F) should not be considered as negative. Histopathological changes in the respiratory tract of hamsters, as well as the observation of some lung tumors, indicate the progression to possible carcinogenic effects. In mice, cadmium compounds appeared to be carcinogenic when adjustments were made for decreased survival. Despite the flaws in the Heinrich et al. study, this study provides some evidence of the carcinogenic potential of cadmium in species other than rats. Therefore, when all of the data are considered in a weight-of-evidence evaluation, the conclusions remain unchanged as to the relevance of the Takenaka data for use in risk assessment, and the potential carcinogenicity of cadmium in the occupational setting.

Continuous Exposure in Animal Studies Versus Intermittent in Occupational Settings

When actual exposure situations of concern differ from continuous, constant lifetime exposure, the EPA's Guidelines for Carcinogen Risk Assessment (51 FR 33992; Sept. 24, 1986) recommend that unless there is evidence to the contrary, the appropriate measure of exposure is the total or cumulative dose of the chemical of concern averaged or prorated over a lifetime, resulting in an average lifetime daily exposure. This assumes that a high dose of a carcinogen received over a short period of time is equivalent to a corresponding low dose averaged over a lifetime in terms of extra cancer risk. The supporting rationale for this relies on the underlying assumptions of the carcinogenic process: risk is linearly related to dose, particularly in the low-dose region (51 FR 33992, Sept. 24, 1986; 50 FR 10372, Mar. 14, 1985; NAS, 1983). Currently when conducting a risk assessment, the total amount of chemical exposure (intake or dose) resulting from less-than-lifetime or intermittent exposure patterns is adjusted (prorated) over the expected lifetime of the individual. The result is an average lifetime daily exposure that corresponds to the same cumulative or total amount of chemical.

The Guidelines state that as the exposures in question become more intense but less frequent, this approach becomes problematic, especially when the agent has demonstrated dose-rate effects. Dose-rate effects are defined as a different degree or type of response that may occur with different dose patterns even when the total dose is the same for these dosing patterns.

Criticisms have arisen that the continuous exposures of animals to cadmium in Takenaka et al. (Ex. 4-67) (23 hours/day; 7 days/week) do not reflect occupational exposures of humans to cadmium (8 hours/day; 5 days/week). However, the available pharmacokinetic information for cadmium does not provide supporting evidence that dose-rate effects would be observed following intermittent versus continuous exposure to cadmium. Cadmium has a long retention time in the rat lung of 60-80 days and it is ten times longer in the human lung; therefore, although exposure in the workplace may be intermittent, cadmium remains in the lung during periods when exposure is discontinued, such as at the end of an 8-hour shift or over a two day weekend (Ex. 31).

Animal studies conducted by Glaser et al. (Ex. 8-694B) also provide evidence that dose-rate effects may not be an issue with cadmium. To evaluate this, the cancer potency estimated from results in rats exposed continuously was compared to that estimated from rats exposed using an intermittent pattern to simulate a work week. In this study, one group of male rats was exposed to 30 ug/m(3) CdO dust for 22 hours/day, 7 days/week for 18 months, followed by 13 months of observation; another group was exposed to 90 ug/m(3) CdO dust for 40 hours/week for 6 months, followed by 24 months of observation. The doses, expressed as an average daily intake, were approximately 9.9 ug/kg/day for the male rats exposed to 30 ug/m(3) for 18 months and 2.7 ug/kg/day for male rats exposed to 90 ug/m(3) for 6 months. The doses were prorated using the following equation:


                0.254 m(3)/day     number of hours     number of months
                                  per week exposed         exposed
 ug/m(3)  X     ______________ X  ________________  X  ________________
                0.43.kg body       160 hours/week       total months on
                   weight                                   study

The incidence of lung tumors in these two groups was 28/39 for the 30 ug/m(3) group and 4/20 for the 90 ug/m(3) group. No lung tumors were reported in controls.

When these two data sets were evaluated using the multistage model, the potency factors obtained for these two data sets were similar, with 0.176 (ug/kg/day)-1 obtained for the 30 ug/m(3) data set and 0.170 (ug/kg/day)-1 obtained for the 90 ug/m(3) data set. If a dose-rate effect existed, i.e., if longer, continuous exposure resulted in higher estimates of risk, then the potency factors for these two data sets would be different rather than comparable. Therefore, the empirical data reported by Glaser et al. (Ex. 8-694B) do not show evidence of a dose rate effect for the lung carcinogenicity of cadmium.

Extrapolation From the Animals to Humans

Dr. Oberdorster has recommended that for the extrapolation of results of animal studies to humans, OSHA should use a lung dosimetric approach, rather than equivalency based on body weight. A lung dosimetric approach assumes that equal accumulated doses of cadmium per gram of lung tissues have the same carcinogenic potential in the peripheral lung of the rat and the human. Using a lung dosimetry methodology, the dose delivered to the lung can be expressed based on lung specific parameters, such as lung weight or lung airway surface area, rather than on body weight or body surface area.

Because the histopathology is not available to identify a specific cell line or area of the lung from which the tumors resulting from cadmium exposure arise in humans, if lung dosimetry is to be used, with surface area of the lung as a lung specific parameter, the total surface area of the lung must be used. If the histopathology were available to identify these areas, ideally the surface area of a specific region of the lung would be used.

Dr. Oberdorster compared the body weight equivalency approach used by OSHA and other regulatory agencies and the lung dosimetric approach and it appears that these two approaches give very similar risk estimates for cadmium. He concluded that the similarity in risk assessment in the case of cadmium is coincidental and does not mean that the choice of a dosimetric method is unimportant. However, it has recently been determined that total lung surface area, which would have to be used in this case, scales allometrically as body weight raised to the 0.96 power (BW0.96) (USEPA, 1991). Therefore, the body weight equivalency method used by OSHA would be a close approximation of the lung dosimetric method, based on total surface area of the lung, and the similarity in the risk estimates derived by Dr. Oberdorster may not be a coincidence.

Regulation of Cadmium Sulfide

Several commenters (Exs. 31; 144-20; 19-42b) recommended that a higher occupational standard should be developed for cadmium sulfide (CdS) than OSHA's proposed standards of 1 or 5 ug/m(3) based primarily on the difference in solubility between CdS and CdCl(2) or CdO. To consider the validity of distinguishing CdS from other cadmium compounds several factors must be considered. There is sufficient evidence in animals for the carcinogenicity of cadmium. Studies conducted by Glaser et al. (Ex. 8-694B) and Oldiges et al. (Ex. 8-694D) appear to provide evidence that CdS is carcinogenic to animals exposed by the inhalation route. However, SCM Chemicals (Ex. 19-42b) suggested that the carcinogenic response observed in Glaser et al. (Ex. 8-694B) may be a result of exposure to CdSO(4) formed during the preparation of CdS. Two studies investigated the aerosol preparation technique of CdS used in the Glasser et al. (Ex. 694B) inhalation study (Glaser et al., 1991; Konig et al. 1991, refs. in Ex. 141). Both studies demonstrated that under the preparation technique used CdS may be solubilized under the influence of light and that, at low aerosol concentrations equivalent to 90 ug/m(3), 50% to 63% of the CdS solubilized. This indicates that the aerosol to which the rats were exposed contained both CdS and the more soluble CdSO(4). If it is assumed that only the more soluble CdSO4 contributed to the carcinogenicity observed, then in the study by Glasser et al. (Ex. 8-694B), it would be expected that the tumor incidence of the CdS/CdSO(4) exposed rats to be approximately half that for the rats exposed to CdSO(4) since approximately 50% of the inhaled Cd was in the form of CdSO(4) (Ex. 142). However the tumor response in animals exposed to 90 ug/m(3) CdS was greater than 50% of that in animals exposed to 90 ug/m(3) CdSO4, and in fact, the response was comparable in animals exposed to the two compounds. In males the lung tumor incidence among CdS exposed rats was 17/20 and only 11/20 among CdSO(4) exposed rats (although the CdS group was exposed for 18 months versus 14 months for the CdSO(4) group). In females the incidence was 15/20 in the CdS group and 18/20 in the CdSO(4) group. Thus, even if it is assumed that the 50% of the CdS had been converted to CdSO(4), these data suggest that the carcinogenic response was not due to the CdSO(4) alone and the data are, in fact, consistent with the CdS being just as potent a carcinogen as CdSO(4).

Intratracheal instillation studies provide further support that CdS is carcinogenic. Intratracheal instillation studies conducted by Pott et al. (Ex. 8-757) were aimed at investigating the pulmonary carcinogenicity of different cadmium compounds, including CdS. A statistically significant increase in the incidence of lung tumors was observed in rats exposed to 10 weekly instillations of 250 ug cadmium via CdS. Although it is possible that a small portion of the CdS could have dissociated, based on information from Konig et al. (Ex. L-140 27b), this could only have resulted in the formation of approximately 3% CdSO(4) from CdS in the group administered 250 ug cadmium. Therefore, the formation of CdSO(4) cannot account for the excess tumors observed following exposure to CdS. According to Oberdorster (Ex. 141) and Heinrich (Ex. 142) the results of the instillation study along with the inhalation study indicate that CdS is a pulmonary carcinogen; however, inhaled CdS may have a lower potency than other cadmium compounds.

If, as is generally assumed, only the solubilized Cd ion is responsible for the observed carcinogenicity, the potential carcinogenicity of any cadmium compound theoretically should be related to the cumulative amount of cadmium ion released in close proximity to target lung cells, averaged over a specific period of time. The release of cadmium is governed by the rate of dissolution of the cadmium ion from the cadmium compound, the biological half-time in the lung, and the mechanism of clearance from the lung. These are interdependent and each contributes to the estimate of lung burden. Since lung dosimetry is extremely complex, the relative solubility of CdS compared to that of CdO or CdCl(2) is only one of the determining factors. While the dissolution rate of the Cd ion from CdS or other cadmium compound may be a function of that specific compound in a physiological environment, biological half-time and mechanism of clearance are dependent on a number of factors that include the nature of the inhaled material, i.e., gas, vapor, aerosol, particle; the characteristics of the respiratory tract; and breathing pattern. The mechanism of clearance is directly related to the deposition within the respiratory tract, which is a function of the particle size. In the upper respiratory tract, clearance by the mucociliary escalator is operative, while in the lower part of the respiratory tract or in the alveoli, clearance can occur by dissolution and direct uptake by macrophages, which are also cleared by the mucociliary escalator. Biological half-time is influenced by the competence and efficiency of these clearance processes. Removal of particles from the lung by mucociliary action, rather than by dissolution and diffusion, may allow for a longer retention time. However, any cytotoxicity that slows ciliary movement or creates an overburden on macrophage capability would increase that retention time, thus allowing for more dissolution and formation of the free cadmium ion, resulting in a higher carcinogenic potency than may be expected based on solubility alone. This would also be of concern because the biological halftime of particles with low solubility is about ten time longer in the human lung than in the rat lung (Ex. 142).

To differentiate risk associated with CdS from that for other cadmium compounds is not feasible at this time. In the workplace, exposure typically does not occur to CdS alone, but rather to a mixture of cadmium compounds. CdS appears to have carcinogen potential, and the contribution to risk from CdS would be additive to that of the other cadmium compounds (Ex. 142). Factors other than the solubility of CdS affect the retention in the lung, and hence the availability of cadmium ions. Factors in the work place that influence lung dosimetry, such as the size of cadmium particles or the total lung burden from all particulates, may also be important. A better understanding of the molecular and cellular mechanisms of cadmium carcinogenicity is required to determine if the carcinogenic potential of CdS is different from that of other cadmium compounds (Ex. 142).

Assessment of Lung Cancer Risk Using Human Data

The Risk Assessment by NIOSH (Stayner et al., Ex. L-140-20)

Since the publication of OSHA's proposed rule, Stayner et al. have completed a risk assessment based on the Thun cohort. There are several differences between this risk assessment and the one reported in the proposed rule (Ex. 18). The Stayner et al. risk assessment is based on recent additional followup of the Thun cohort through 1984, whereas the risk assessment reported in the proposed rule was based on followup only though 1978. Stayner et al. had access to the unprocessed data from the Thun cohort and, consequently, was able to conduct a wider range of analyses in their risk assessment than was possible based on just the data reported in the published Thun report (Ex. 4-68).

The cohort studied by Stayner et al. contained 606 white males and was defined the same way as in the earlier study of Thun et al. (Ex. 4-68). As in the earlier study, workers with employment prior to January 1, 1926 were excluded from the analysis in order to minimize the potential for confounding by arsenic exposure.

A life-table analysis was used to study the lung cancer mortality of the cohort. Person-years were accumulated beginning with the time an individual had been employed for six months at the facility or with January 1, 1940, whichever came later. Because Hispanics are reported to have lower lung cancer rates than non-Hispanics, the cohort was separated into Hispanic and non-Hispanic based on surname, and results were reported separately for these two groups; however, U.S. white males were still used as the comparison population for each group. Stayner et al. categorized the person-years in four ways: cumulative exposure, latency (elapsed years since first exposure), year of observation, and age. Results from this analysis are reported in Table VI-7.


TABLE VI-7. - RESULTS OF LIFE TABLE ANALYSIS OF DATA FROM NIOSH UPDATE
               THROUGH 1984 OF CADMIUM SMELTER COHORT
_____________________________________________________________________________
                   |    Non-Hispanic   |   Hispanic(a)    |   Combined
  Category         |___________________|__________________|__________________
                   | OBS | EXP  | SMR  | OBS |  EXP | SMR | OBS |  EXP  | SMR
___________________|_____|______|______|_____|______|_____|_____|_______|____
                   |     |      |      |     |      |     |     |       |
Overall............|  21 | 9.95 |**211 |   3 | 6.12 |  49 |  24 | 16.0  | 149
                   |     |      |      |     |      |     |     |       |
Exposure:(b)       |     |      |      |     |      |     |     |       |
   < 584............|   1 | 3.35 |   29 |   1 | 2.38 |  42 |   2 |  5.73 |  34
   585-1460........|   7 | 2.64 | *265 |   0 | 1.64 |   0 |   7 |  4.28 | 163
   1461-2920.......|   6 | 1.55 | *386 |   0 | 1.20 |   0 |   6 |  2.75 | 217
   >2921...........|   7 | 2.41 | *290 |   2 | 0.90 | 223 |   9 |  3.30 |*272
Latency(c) (Years):|     |      |      |     |      |     |     |       |
   < 10.............|   0 | 0.41 |    0 |   1 | 0.28 | 363 |   1 |  0.69 | 145
   10-19...........|   2 | 1.41 |  142 |   0 | 1.00 |   0 |   2 |  2.41 |  83
   >20.............|  19 | 8.13 |**233 |   2 | 4.84 |  41 |  21 | 12.9  |*161
                   |     |      |      |     |      |     |     |       |
Year:              |     |      |      |     |      |     |     |       |
   1940-1959.......|   2 | 0.89 |  225 |   0 | 0.38 |   0 |   2 |  1.26 | 158
   1960-1969.......|   5 | 2.24 |  223 |   1 | 1.27 |  78 |   6 |  3.51 | 171
   1970-1979.......|  10 | 4.43 | *228 |   2 | 2.88 |  69 |  12 |  7.30 | 164
   >1980...........|   4 | 2.39 |  167 |   0 | 1.59 |   0 |   4 |  3.98 | 101
Age (Years):       |     |      |      |     |      |     |     |       |
   <  50............|   0 | 0.78 |    0 |   1 | 0.50 | 201 |   1 |  1.28 |  78
   50-54...........|   2 | 1.01 |  198 |   0 | 0.67 |   0 |   2 |  1.68 | 118
   55-59...........|   1 | 1.61 |   62 |   2 | 1.00 | 200 |   3 |  2.60 | 115
   60-64...........|   5 | 2.20 |  227 |   0 | 1.20 |   0 |   5 |  3.40 | 146
   65-69...........|   4 | 2.12 |  188 |   0 | 1.13 |   0 |   4 |  3.25 | 123
   70-74...........|   5 | 1.37 | *366 |   0 | 0.84 |   0 |   5 |  2.20 | 227
   >75.............|   4 | 0.87 | *547 |   0 | 0.79 |   0 |   4 |  1.66 | 241
___________________|_____|______|______|_____|______|_____|_____|_______|____
  Footnote(a) U.S. rates for white males were used as the referent group
for hispanic and non-hispanic males in this analysis.
  Footnote(b) Milligrams cadmium per cubic meter of air-days.
  Footnote(c) Time since first exposure.
  *p <  0.05 (two-tails)
  **p <  0.01 (two-tails)

Overall, there was an excess of lung cancer (OBS = 24; EXP = 16.07;

SMR = 149) that is statistically significant (p=0.035) by a one-tailed test for higher lung cancer rates in the cohort. Moreover, there is a clear dose-response trend of higher SMRs for lung cancer in groups with higher cumulative cadmium exposures. Lung cancer was significantly elevated among non-Hispanics [SMR = 221, 95% Confidence Intervals (CI) = 131, 323], but reduced among Hispanics (SMR = 49, 95% CI = 10, 143). This latter finding is consistent with the fact that Hispanics are reported to have lower lung cancer rates in general than non-Hispanics (Ex. 33) and that reference rates used were for U.S. white males.

There is a deficit in lung cancer deaths in the lowest exposure group (584 mg-days/m(3)) relative to the control population, among both non-Hispanics (OBS=1, EXP=3.35) and Hispanics (OBS=1, EXP=2.38). However, neither of these deficits is statistically significant (p=0.15 among non-Hispanics and p=0.31 among Hispanics, one-tailed tests). Neither is the deficit statistically significant in the combined cohort (OBS=2, EXP=5.73, p=0.075) despite the fact that a deficit among Hispanics is expected because control rates were for white males. Thus, these deficits are not inconsistent with ordinary random fluctuation.

Excesses of lung cancer are observed in the remaining three exposure groups for both non-Hispanic and combined Hispanic and non-Hispanic. These excesses are statistically significant (using U.S. white males as the referent population) in all three groups for non-Hispanics and in the highest exposure group for combined Hispanics and non-Hispanics.

When person-years are categorized according to latency, significant responses occur only in the group with a latency of = or > 20 years, which is consistent with the latency of other agents that cause lung cancer. Table VI-7 also indicates that the excess of lung cancer was greatest among older members of the cohort (+ or - 70 years of age).

Stayner et al. used both Poisson regression and Cox regression to model the relationship between cumulative cadmium exposure and risk. Both Poisson regression and Cox regression involve a regression model that expresses the lung cancer rate per person-year (i.e., per person per year) in the cohort in terms of various potential explanatory variables for lung cancer such as age, calendar year, and cadmium exposure.

In Poisson regression, the person-years of observation are categorized according to values of the explanatory variables. The model is fit to the data using the assumption that the number of observed cases in each cell determined by the categorization has approximately a Poisson distribution with expected value equal to the number of cases predicted by the regression model. Poisson regression can either utilize background lung cancer rates from an external control population in defining the regression model, or else can estimate all of the parameters necessary to define the lung cancer rate directly from the cohort data without resort to an external control population. OSHA used Poisson regression of the former type in its risk assessment that was presented in the proposed rule (Ex. 18), whereas Stayner et al. utilized Poisson regression of the latter type. In addition to a measure of prior cadmium exposure, Stayner et al. included age, calendar year, and Hispanic ethnicity as covariates (explanatory variables) in the Poisson regression analyses.

Stayner et al. utilized the following functional forms for the lung cancer mortality rate per person-year in Poisson regression analyses:


Exponential (log-linear):
     h = exp(alpha + E(j)(theta(j)W(j)) + gamma(chi) + deltaY + betaX)

Linear:
     h = alpha + E(j)(theta(j)W(j)) + gamma(chi) + deltaY + betaX

Power:
     h = exp(alpha + E(j)(theta(j)W(j)) + gamma(chi) + deltaY) * ([X + 1]
beta)

Additive relative rate:
     h = exp(alpha + E(j)(theta(j)W(j)) + gamma(chi) + deltaY) * [1 +
betaX]

where

h is the lung cancer mortality rate per person-year hazard rate,

Alpha is the intercept,

W(j) represents the calendar-year groups (W(j) is a category variable that
equals 1 if the observation is from the j-th calendar year group and equals
0 otherwise)

Theta(j) is the regression coefficient for the j-th calendar-year group,

E(theta(j)W(j)) represents the effect of the calendar-year group

[E(j); (theta(j)W(j)) = theta(j') where j' is the particular calendar
year group associated with the observation]

Chi represents Hispanic ethnicity (chi = 1 if the observation is from
a person of Hispanic ethnicity and zero otherwise)

Gamma is the regression coefficient for Hispanic ethnicity,

Y is age,

Delta is the regression coefficient for age,

X is a measure of prior cadmium exposure,

Beta is the regression coefficient for cadmium exposure.

Each of these functional forms, except the linear form, was also fit using Cox regression. Both cumulative exposure and cumulative exposure lagged 5 years were used as the measure of exposure in the Poisson regression analyses.

Stayner et al. utilized the following functional forms for the lung cancer mortality rate per person-year in Cox regression analyses:


Exponential (log-linear):
     h = h(0)(t) * exp(E(j)(theta(j)W(j)) + gamma(chi) + betaX)

Power:
     h = h(0)(t) * exp(E(j))theta(j)W(j)) + gamma(chi)) * ([X + 1](beta))

Additive relative rate:
     h = h(0)(t) * exp(E(j)(theta(j)W(j)) + gamma(chi)) * [1 + betaX]

where

t is age,
h(0)(t) is a base-line mortality rate.

Otherwise, the variables have the same meanings as in the Poisson
regression models.

In Cox regression the baseline mortality rate, h(0)(t), is not estimated, but is left unspecified. Consequently, the method can only estimate the mortality rate relative to this baseline rate. Since the linear model applied in Poisson regression cannot be represented as the product of a baseline mortality rate and a function of the explanatory variables (as is required in Cox regression) no counterpart to this model could be applied in Cox regression. Cox regression does not involve categorization and grouping of data into cells, as is required in Poisson regression. Stayner et al. applied several measures of prior cadmium exposure in the Poisson regression analyses: cumulative cadmium exposure (i.e., the integral of cadmium exposure over time, expressed in units of mg-days/m(3)) and measures formed by "lagging" cumulative exposures by 5, 10, 15, or 20 years. In the lagged analyses, the exposure variable was cumulative exposure achieved up to 5, 10, 15, or 20 years prior to the observation time. This technique was used to discount the most recent exposures that may be etiologically irrelevant to cancer risk because of an apparent minimum delay between exposure and the effect of that exposure upon cancer risk for many chemical carcinogens. Stayner et al. found that lagging exposures 5 years slightly increased the magnitude of the cadmium parameter exposure estimates in the Poisson regression analyses, which is to be expected since the lagged exposure variables are smaller than the unlagged variables. However, lagging the exposures for 10, 15, or 20 years reduced the magnitude of the exposure parameters. This means that the association between lung cancer and cadmium exposure was reduced if exposures were lagged more than 5 years. Stayner et al. consequently concluded that 5 years was the most appropriate lag period. Only results derived from analyses involving a 5-year lag will be discussed further.

The results of the Poisson regression analyses involving cumulative cadmium exposures lagged 5 years are summarized in Table VI-8. Table VI-8 indicates that the deviance associated with the linear model was much larger than that associated with the remaining three models fit by Poisson regression. This means that the linear model fit the data much more poorly than the other models that were applied to the data using Poisson regression. (The deviance differs by a constant from the negative of twice the log-likelihood, and thus, like the log-likelihood, can be used to assess the significance of parameters and assess improvements in fit (BEIR IV, 1988). The power model produced the smallest deviance; however, it predicted unrealistically low background lung cancer rates (e.g., about a factor of 100 lower than that observed for U.S. white males aged 70 during 1970-1979) and for this reason was considered to be inappropriate for assessing risk from cadmium exposure(1). Of the remaining two models, the additive relative rate (Additive RR) model yielded a better fit (i.e., lower deviance) than the exponential model.


__________
  Footnote(1) If the power model had been used for risk assessment it
would have predicted  a much higher risk than the other models. Since this
model predicts a very low background, it predicts a very large
differential in lung cancer mortality between unexposed and exposed.
However, when estimating lifetime risk the background lung cancer is
estimated from general mortality rates and is hence the same no matter
which model is applied. Consequently, the very large differential
predicted by the power model would result in a very large risk if used to
estimate lifetime risk.


Table VI-8. - RESULTS FROM POISSON REGRESSION MODELS AND COX REGRESSION
  MODEL FITED TO 5-YEAR LAGGED DATA BY STAYNER ET AL. (EX. L-140-20)
_____________________________________________________________________________
                              |            |          |  Exposure parameter
        Model                 | Degrees of | Deviance |  estimate (beta)
                              |  freedom   |          |([ug/years/m(3)]-(1))
______________________________|____________|__________|______________________
                              |            |          |
Poisson Regression:(a)        |            |          |
   Linear.....................|      213   |   97.71  |    0.00000008
   Exponential................|      213   |   82.29  | (c)0.00012
   Power function.............|      213   |   79.28  | (d)0.58
   Additive RR................|      213   |   80.70  | (c)0.00061
Cox Regression:(b)            |            |          |
   Additive RR................|............|..........| (d)0.00026
______________________________|____________|__________|______________________
  Footnote(a) All models include categorical variables to control for
calendar year and Hispanic ethnicity, and a continuous variable to control
for age.
  Footnote(b) Model included categorical variables to control for calendar
year and Hispanic ethnicity.  Age was controlled by matching on survival to
the same age.
  Footnote(c) p< 0.05
  Footnote(d) p< 0.01

The same qualitative results were obtained by Stayner et al. using Cox regression. The additive relative rate model fit the data better than the exponential model, but not as well as the power model; however, the power model predicted unrealistically low background lung cancer rates.

Based on these results, Stayner et al. selected the additive relative rate model as the best model to represent the dose response for cadmium and lung cancer. Table VI-8 also contains the results of applying this model using Cox regression.

As indicated in Table VI-8, except for the linear model (which fit the data very poorly), all of the models fit using Poisson regression indicated a statistically significant effect of cadmium exposure upon lung cancer (i.e., the estimate of beta, the cadmium exposure parameter, was statistically significantly greater than zero in each of the three analyses.) Similarly, the additive relative rate model fit using Cox regression also indicated a statistically significant effect of cadmium exposure upon lung cancer.

Stayner et al. used the results from the relative rate model to estimate the excess lifetime risk of lung cancer based on the method described by Gail (Ex. 8-561), which was also used by OSHA in the proposed rule (Ex. 18). The formula for the excess lung cancer risk is:


  The sum of from i=20 to 100 of (RR(i) - 1) q(1) (i) exp

  [1 the sum of j=20 to 1 of {(RR(j) - 1) q(1)(j) + q(a)(j)}]

Where:

RR(i)   is the risk ratio for lung cancer predicted by the model based on
        the exposure scenario assumed,
q(1)(i) is the background age-specific lung cancer mortality rate at
        age i,
q(a)(i) is the background age-specific mortality rate for all causes of
        death at age i,
and I   is the oldest age through which excess risk is accumulated.

For this calculation it was assumed that persons were exposed occupationally to a constant cadmium concentration for 45 years beginning at age 20. The q(l)(i) and q(a)(i) were based on mortality rates for lung cancer and all causes, respectively, from 1984 for U.S. males (all races). The risk ratio used in the calculation is RR(i) = 1 + BetaX(i), where beta is the lung cancer potency derived from the additive relative rate model obtained either by Poisson regression or Cox regression, and X(i) is the measure of exposure associated with age i. For each year, the cadmium exposure is calculated for the midpoint of the year. For example, if cadmium exposure is lagged 5 years and risk is estimated from exposure to an 8-hour time weighted exposure level, Z, beginning at age 20, then X(i) = 0 for i< 25, X(25) = 0.5*Z, X(29) = 4.5*Z, and for i>70 (i.e., after age 70) X(i) = 45*X.

In their original work, Stayner et al. (Ex. L140-20) accumulated risk through age 74 (i.e., set I=74), which is the same as the approach used by OSHA in the proposed rule. However, in an addendum (Ex. L-163), Dr. Stayner noted that accumulating excess risk only through age 74 will result in an underestimate of the cadmium-induced risk of lung cancer because any increased risk that occurs after 75 years of age is not included. Dr. Stayner corrected this oversight in the addendum by accumulating the excess risk up through age 100 (i.e., set I = 100), an age that exceeds the vast majority of human lifespans. OSHA believes that this approach provides a better estimate of the total risk from cadmium than truncating the risk calculation at age 74. This approach has been followed in all calculations of risk derived from human data presented herein(2).


___________
  Footnote(2)  Each term in the sum used to calculate the excess risk is
the product of the probability that a person lives to the given age, times
the excess risk of dying of cadmium-inducaed cancer given that he lived to
that age.  Thus, the method adjusts for risk of death from non-cadmium
causes and does not assume that a person will live to be 100 years old.
However, truncating the sum at 100 is equivalent to assuming that a person
will not live past 100 years of age.  Although there is a small
probability that a person will live past 100 years of age, there is a much
larger probability (on the order of 0.5) that a person will live past age
74.  Consequently, accumulating excess risk through 100 years of age
provides a much better approximation to the total excess risk than
accumulating risk only through age 74, which is equivalent to assuming
that no one lives past age 75.  A similar accounting for total excess risk
was made in the risk assessments used by OSHA to support its standards for
arsenic and for benzene.

Dr. Stayner also noted that their original estimates (Ex. L-140-20) had been based on respiratory cancer rates rather than rates for lung cancer, and rates for lung cancer only were used in calculations reported in the addendum.

Dr. Stayner also reported in the addendum on the results of extending their analyses to include applying the multistage model of cancer to the data on the Thun et al. cohort with follow up through 1984. The multistage model was proposed for risk assessment by Crump and Howe (1984) and is based on a multistage theory of carcinogenesis that assumes that a single cell line has to go through a number of discrete stages in order to produce a tumor. Dr. Stayner assumed a five-stage model for lung cancer, explaining that "this is the number of stages that has generally been observed in other multistage analyses of lung cancer." He also assumed that cadmium affected one of the five stages. He fit the multistage model to the Thun et al. data varying the stage affected by cadmium and found that the model fit best when it was assumed that cadmium affected the third stage. Dr. Stayner fit the model to the data using a Cox regression approach. He then calculated estimates of excess using the same approach based on Gail's formula that he applied to the results from the additive relative rate model. These estimates were based on a multistage model with five stages with cadmium exposure affecting the third stage.

Table VI-9 presents estimates of the excess risk of lung cancer from 45 years of occupational exposure at various TWA exposures that were reported by Stayner in his addendum. The relative rate model obtained from Cox regression provides estimates of risk that are about twice as large as those obtained from the same model using Poisson regression. The estimates obtained from the multistage model fall in between the estimates derived from the other two approaches. The expected number of excess lung cancer deaths from 45 years of occupational exposure at a TWA of 5 ug/m(3) cadmium is estimated as being between 3.9 and 9.0 per 1000 workers. If the TWA is 100 ug/m(3) cadmium, the corresponding number of excess lung cancer deaths is estimated as being between 73 and 157 per 1000 workers.


TABLE VI-9. - ESTIMATES OF EXCESS LUNG CANCER DEATHS PER 1000 WORKERS FROM
  45 YEARS OF OCCUPATIONAL EXPOSURE TO CADMIUM OBTAINED BY STAYNER ET AL.
  BASED ON THE ADDITIVE RELATIVE RATE MODEL APPLIED USING BOTH POISSON
  REGRESSIONA AND COX REGRESSION WITH A 5-YEAR LAG FOR CADMIUM EXPOSURE,
  AND ON THE MULTISTAGE MODEL
___________________________________________________________________________
          Exposure           |   Excess deaths (per 1000 workers)
          (ugm(3))           |_____________________________________________
                             |       Relative Rate Model     |
_____________________________|_______________________________| Multistage
           TWA               |   Poisson     |     Cox       |   model(c)
                             | regression(a) | regression(b) |
_____________________________|_______________|_______________|_____________
                             |               |               |
  l..........................|          1.8  |          0.8  |       1.1
  3..........................|          5.4  |          2.3  |       3.3
  5..........................|          9.0  |          3.9  |       5.5
  7..........................|         12.5  |          5.4  |       7.7
 10..........................|         17.8  |          7.7  |      11.0
 20..........................|         35.1  |         15.3  |      21.8
 40..........................|         68.2  |         30.3  |      42.9
 50..........................|         84.1  |         37.7  |      53.2
100..........................|        157.0  |         73.0  |     102.2
200..........................|        275.7  |        137.6  |     189.1
_____________________________|_______________|_______________|____________
  Footnote(a) Beta = 0.00061 (ug-years/m(3))(-1).
  Footnote(b) Beta = 0.00026 (ug-years/m(3))(-1).
  Footnote(c)  The multistage model was fitted assuming five stages.
Results presented assume the third stage affected by cadmium exposure,
which maximized the likelihood.

Update of OSHA's Risk Assessment Based on the Thun Cohort

OSHA has also updated its risk estimates presented in the Proposed Rule (Ex. 18), based on the additional followup presented in the Stayner et al. risk assessment. OSHA's earlier risk assessment was based on the result of the life-table analysis conducted by Thun et al. (Ex. 4-68) on the followup of the cohort through 1978. The data set studied by Stayner et al. contained the results of additional followup through 1984.

The results of the life-table analysis conducted by Stayner et al. are contained in Table VI-7. The information from this table that will be used to update the risk assessment is contained in the section of the table labeled "EXPOSURE." This information consists of observed and expected numbers of lung cancer deaths in the cohort categorized by cumulative cadmium exposure and by Hispanic versus non-Hispanic. Expected numbers of deaths were calculated based on the mortality experience of U.S. white males.

In its earlier risk assessment OSHA applied both an absolute risk model and a relative risk model to the data from the Thun et al. study. However, the number of person-years in each category in Table VI-7, which is needed for the application of the absolute risk model, was not reported by Stayner et al. Consequently, only the relative risk model will be applied to the update. (It should also be noted that the linear model used by Stayner et al. (Ex. L-140-20), which is similar in functional form to the absolute risk model, provided a poor fit to the data.) The relative risk model assumes that the lung cancer mortality (hazard) rate at age t is given by:


h(t) = h(0)(t) * (1 + beta * X) = h(0)(t) + [h(0)(t) * beta * X],

where h(0)(t) represents the hazard rate in the absence of cadmium exposure, X is a measure of cadmium exposure, and beta is the slope of the dose response (i.e., beta is the carcinogenic potency of cadmium). The measure of cadmium exposure available from the Stayner et al. paper was cumulative exposure, which was the same measure used by OSHA earlier. The specific measure used in the fitting was the median exposures in units of mg/m(3)-days derived from the life-table analyses and converted into units of ug/m(3)-years by Stayner et al. by multiplying by 1000 and dividing by 365; the resulting exposures as reported by Stayner et al. are 795, 2466, 5699 and 10,836 ug/m(3)-days corresponding to the four cumulative exposure groups displayed in Table VI-7. (It was necessary for Stayner et al. to divide by 365 instead of 240, the number of work days in a year, because in calculating the exposures used to obtain the categorization shown in Table VI-7 it was assumed that one-month exposure in a work category meant 30 days of exposure.) The background rates used in the life-table analysis in Table VI-7 were for U.S. white males (which includes Hispanics). Since the cohort involves a sizable fraction of Hispanics (a larger fraction than in the U.S. in general), additional parameters were included in the model to control for the possibility that the background rates in the population differed between Hispanics and non-Hispanics and between either of these groups and U.S. white males in general. Since Thun et al. did not report their data separately for Hispanic and non-Hispanic, it was not possible for OSHA to conduct analyses similar to those in the risk assessment reported in the Proposed Rule.

The resulting model for the expected number of lung cancer deaths in a group with a cumulative cadmium dose of X is


E(O) * exp(a(i)) * (1 + beta * X),

where

E(o) is the expected number of cancers (obtained from Table VI-7) in the
group based on rates for U.S. white males,

a(i) is a category variable with i = H applying to Hispanics and
i = NH applying to non-Hispanics.

The model was fit both with the restriction a(NH) = 0 (corresponding to applying the U.S. white male rates to the non-Hispanic cohort) and without this restriction. The observed deaths in a group were assumed to have a Poisson distribution whose expectation is given by the above expression. Thus, this analysis is a form of Poisson regression, with the principal difference between this analysis and the Poisson regression conducted by Stayner being that this analysis utilizes mortality rates from an external population to define background mortality rates, whereas Stayner et al. did not utilize an external population in their Poisson regression model. The parameters of the model were estimated by maximizing the likelihood (e.g., by Poisson regression), utilizing the computer program AMFIT (BEIR V, 1990).

A summary of the results of fitting this model to two different cases is presented in Table VI-10. In Case I, lung cancer mortality rates for U.S. white males are used as background rates for non-Hispanic white males in the cohort. In Case II, background rates for non-Hispanic white males in the cohort are assumed to differ from rates for U.S. white males by the multiplicative constant exp(a(NH)), which is estimated from the data. In both analyses, background rates for non-Hispanic white males in the cohort are assumed to differ from rates for U.S. white males by the multiplicative constant exp(a(H)).


TABLE VI-10. - RESULTS OF APPLYING OSHA'S MODIFIED RELATIVE RISK MODEL
  TO THE 1984 FOLLOWUP OF THE THUN COHORT
__________________________________________________________________________
                       | Case I(a) (a(NH)=0) | Case II(a)(a(NH) estimated)
_______________________|_____________________|____________________________
                       |                     |
a(H) (s.e.)............|    -1.4 (0.60)      |         -1.8 (0.91)
a(NH)..................|  0                  |         -0.48 (0.77)
Beta(b)................|  0.00027 (0.000098) |       0.00054 (0.00057)
Deviance...............| 10.29               |       9.88
_______________________|_____________________|_____________________________
  Footnote(a) Case I assumes lung cancer mortality rates for U.S. white
males are appropriate background rates for non-Hispanic white males in
this cohort.  Case II permits background rates for non-Hispanic white
males to differ from rates for U.S. white males by the multiplicative
constant, exp (a(NH)).
  Footnote(b) Units are (ug-years/m(3))(-1).

This analysis does not indicate that mortality rates for U.S. white males are inappropriate for the non-Hispanics in this cohort. The reduction in the deviance when a(NH) is estimated is only 10.29-9.88 = 0.41, which is not significant (p = 0.52) based on the chi-square distribution with one degree of freedom. On the other hand, a(H) is significantly less than zero in both cases, implying that the Hispanics in this cohort had a lower background mortality rate from lung cancer than U.S. white males.


TABLE VI-11. - OBSERVED AND PREDICTED LUNG CANCER DEATHS FROM THE RELATIVE
                RISK MODEL APPLIED TO THE 1984 UPDATE TO THE THUN COHORT
___________________________________________________________________________
                         |              |      Number of lung cancers
                         |  Number of   |            predicted
Exposure (ug-years/m(3)  | lung cancers |__________________________________
                         |   observed   |   Case I(a)   |    Case II(a)
                         |              |   (a(NH)=O)   | (a(NH) estimated)
_________________________|______________|_______________|__________________
                              Non-Hispanics
___________________________________________________________________________
795......................|         1    |         4.1   |      3.0
2466.....................|         7    |         4.4   |      3.8
5699.....................|         6    |         4.0   |      3.9
10836....................|         7    |         9.5   |     10.3
_________________________|______________|_______________|__________________
                               Hispanics
___________________________________________________________________________
795......................|         1    |         0.71  |      0.50
2466.....................|         0    |         0.67  |      0.63
5699.....................|         0    |         0.75  |      0.80
10836....................|         2    |         0.87  |      1.0
                         |              |_______________|_________________
                         |              | X(2)=8.5 (NS) | X(2)=8.8 (NS)
                         |              |          6 df |          5 df
_________________________|______________|_______________|________________
  NS = nonsignificant lack of fit
  df = degrees of freedom
  Footnote(a) Case I assumes lung cancer mortality rates for U.S. white
males are appropriate background rates for non-Hispanic white males in
this cohort.  Case II permits background rates for non-Hispanic white
males to differ from rates for U.S. white males by the multiplicative
constant, exp (a(NH)).

Table VI-11 indicates that both of these cases provide an adequate fit to the data from the 1984 update of the Thun cohort. For Case I (a(NH) "identical with" 0), the chi-square is 8.56 [6 degrees of freedom (d.f.), p = 0.20], and for Case II (a(NH) estimated) the chi-square is 8.82 (5 d.f., p = 0.12). The lung cancer potency estimates obtained from these analyses were beta = 0.00027 [ug-years/m(3)](-1) (Case I) and beta = 0.00054 [ug-years/m(3)](-1) (Case II). Both of these estimates are within the range of the corresponding estimates obtained by Stayner et al. when fitting the additive relative rate model, which was their preferred model, by Cox regression (beta = 0.00026 [ug-years/m(3)](-1)) or Poisson regression (beta = 0.00061 [ug-years/m(3)](-1)). (See Table VI-8.)

Thus, both Case I and Case II provide adequate descriptions of the data and both provide estimates of the carcinogenic potency of cadmium that are similar to estimates obtained by Stayner et al. using different modelling approaches. Case I is somewhat simpler than Case II since it requires one fewer parameters to be estimated.

Table VI-12 contains estimates, based upon Case I, of the number of excess lung deaths per 1000 workers exposed for 45 years to different TWA concentrations of cadmium. This table contains both MLEs and upper and lower statistical confidence limits. The MLEs were calculated in exactly the same way as the corresponding estimates in Table VI-9, which were made by Dr. Stayner, except that the potency estimate, beta = 0.00027 [ug-years/m(3)](-1), corresponding to Case I was used in the calculation. The 95% lower and upper confidence limits in Table VI-12 were calculated in this way also, except that the lower and upper confidence limits for beta were used instead of the MLE. The 95% lower limit was beta = 0.00013 [ug-years/m(3)](-1) and the 95% upper limit was beta = 0.00046 [ug-years/m(3)](-1), both of which correspond to Case I. These confidence limits were calculated using the likelihood ratio approach. MLEs based on Case II (with aNH estimated) are about twice as large as those in Table VI-12 and the corresponding lower limits on risk calculated using the likelihood approach are zero(3).


__________
  Footnote(3) With this approach, a 95% lower (upper)limit is calculated
as the value of beta smaller (larger) than the maximum likelihood estimate
that satisfies the equation [2(*)(L(MAX)--L(beta))](alpha)(delta)=1.645
where L(MAX) is the maximum value of the log-likelihood and L(beta) is the
log-likelihood expressed as a function of beta (Cox and Hinkley, 1974).


TABLE VI-12. - OSHA FINAL ESTIMATES OF EXCESS
  LUNG CANCER DEATHS PER 1000 WORKERS WITH 45
  YEARS OF OCCUPATIONAL EXPOSURE TO CADMIUM
  BASED ON THE RELATIVE RISK MODEL ASSUMING
  U.S. LUNG CANCER RATES ARE APPROPRIATE FOR
  NON-HISPANIC WHITE MALES IN COHORT
  [a(NH)=0](a,b)
________________________________________________
  Exposure (ug/m(3)) TWA |    Risk per 1000
_________________________|______________________
                         |
1........................|   0.6 (0.3, 10)
3........................|   1.8 (0.9, 3.0)
5........................|   3.0 (1.5, 5.1)
7........................|   4.2 (2.1, 7.1)
10.......................|   6.1 (2.9, 10.1)
20.......................|  12.1 (5.9, 20.1)
40.......................|  23.9 (11.7, 39.6)
50.......................|  29.8 (14.5, 49.1)
100......................|  58.3 (28.8, 95.0)
200......................| 112.1 (56.5, 177.9)
_________________________|______________________
  Footnote(a) Estimates derived using data from
the 1984 update by Stayner et al. (Ex. L-140-20)
of cadmium smelter workers.
  Footnote(b) Numbers in parentheses are 95%
upper and lower confidence limits.

As indicated by Tables VI-9 and VI-12, all of the estimates of excess risk of lung cancer obtained from the Thun cohort are similar despite the fact that the underlying analytic methods differed in several respects. Estimates of excess risk from 45 years of occupational exposure computed using the various approaches range from three excess deaths per 1000 workers to nine excess deaths per 1000 workers exposed to a TWA 5 ug/m(3) and from 58 to 157 excess deaths per 1000 workers exposed to a TWA of 100 ug/m(3). As Table VI-3 indicates, these estimates are somewhat higher than those obtained by OSHA in the proposed rule. OSHA believes that the differences between the new estimates and those in the proposed rule are most likely attributable to three factors: (1) The new estimates are based on additional followup of the Thun cohort; (2) the estimates appearing in the proposed rule were not adjusted for ethnicity (Hispanic versus non-Hispanic); (3) the estimates were adjusted with the assumption that a person could live beyond age 74 years. Because the new estimates are based upon more complete data and more reliable quantitative methods, OSHA believes that the new estimates are more reliable than those that appeared in the proposed rule.(4)


__________
  Footnote(4) The data necessary for such an adjustment were not published
in Thun et al. (Ex. 4-68) and consequently were not available to OSHA when
preparing the proposed rule.

Discussion of Issues Related to Risk Assessment for Lung Cancer Based Upon Human Data

Potential for Confounding by Arsenic Exposure in Thun Cohort

Several commenters questioned whether there was an excess of lung cancers among workers hired in or after 1940, and whether an excess of lung cancer in persons hired prior to 1940 could have been due to arsenic rather than cadmium (Exs. 144-8B, 38). Dr. Schulte (Ex. 144-8c) addressed the lung cancer response among workers hired in 1940 or later as follows:

We have never felt that an analysis of the subcohort hired between 1940 and 1969 was scientifically justified. The analysis was only done because ASARCO desired it on the basis of thier arsenic assumptions. We do not feel this analysis is justified for several reasons: (1) We do not see an important change in the arsenic content of the feedstock in 1940, (2) the subgroup hired after 1940 is too small to provide sufficient statistical power for a scientifically definitive analysis and (3) the latency period for workers hired after 1940 was not long enough to scientifically justify analyzing this subcohort that said, we can interpret the post-1940 data with the aforementioned caveats. Clearly, there was an excess of lung cancer in the medium and high exposure groups combined (OBS=13, EXP=5.71, SMR=228, 95%CI=121-389) and no excess in the low group; these data are supportive of a dose-response pattern. The fact that the risk estimate for the high exposure group was lower than that for the medium group can be explained by the low number of workers in this group, and thus, the low statistical power to describe the effect...."

Thus, an analysis by NIOSH of workers first employed after 1940 did demonstrate an excess of cancer in the medium and high exposure groups and was supportive of a dose-response pattern.

Stayner et al. also addressed the issue of whether there was a lung cancer effect in workers first hired in 1940 or later by adding a category variable to their analysis representing whether or not year of first hire was in 1940 or later (Ex. L-140-20). When this variable was added, it was not significant, which indicates that there was no statistically significant difference between the lung cancer mortality rates among persons first hired before 1940 and those hired later that was not already being explained by the model. Moreover, if arsenic exposure was higher prior to 1940 and if this exposure was wholly or partially responsible for the excess in lung cancers observed in this cohort, then inclusion of the category variable for year of hire should reduce the magnitude of the coefficient for cadmium exposure. However, when the category variable representing year of hire was added, the magnitude of the cadmium coefficient increased, which is not consistent with the hypothesis that arsenic exposure prior to 1940 was largely responsible for the increased lung mortality seen in this cohort.

This analysis by Stayner et al. was commented on by Mr. Leonard Ulicny of SCM Chemicals as follows:

"This analysis is significantly flawed, as the year of hire variable and the cumulative exposure variable are not truly independent. ... For nearly any composite of work departments, the estimates of cadmium inhalation in workers hired before 1940 will be higher than estimates for those hired after 1940. Since the 'independent' variables are in fact related, regression analysis cannot distinguish between them" (Ex. 144-17).

OSHA agrees that cadmium exposure is very likely to be correlated with the year of hire variable used by Stayner et al. However, OSHA also believes that there were significant cadmium exposures after 1940 and there were considerable differences in cadmium exposures, both among men first hired prior to 1940 and among those first hired subsequently. Consequently, OSHA believes that the correlation between the two variables is far from perfect. Indeed, if it were perfect or nearly so, inclusion of the year of hire variable would have resulted in a decrease in the magnitude of the cadmium exposure variable. However, when the year of hire variable was included, the cadmium exposure variable actually increased in magnitude.

If there was a perfect correlation between cadmium exposure and arsenic exposure (or a surrogate for arsenic exposure such as year of hire), which OSHA does not believe is the case, then an internal analysis of the Thun cohort would be incapable of separating the effects of arsenic and cadmium. However, it would still be possible to address the potential magnitude of an effect of arsenic using the carcinogenic potency of arsenic estimated from other studies. Such an analysis has been conducted by Thun et al. (Ex. 4-68) and will be reviewed below. Thun et al. (Ex. 4-68) addressed the question of whether arsenic exposure may have been largely responsible for the excess lung cancer incidence observed in this cohort. This was done by estimating the potential number of arsenic-related cancers in their cohort using the carcinogenic potency of arsenic estimated by OSHA in its final rule on its most recent arsenic standard (48 FR 1864; Jan. 14, 1983). Thun et al. estimated that the average exposure to arsenic was 500 ug/m(3) in the areas of highest arsenic exposure (near the roasting and calcine furnaces), based on exposure measurements taken in 1950. After reducing this exposure by 75% to account for respirator use, exposure in the high arsenic areas was estimated as 125 ug/m(3). They further took into account that an estimated 20% of person-years were spent in high arsenic work areas, resulting in an average arsenic exposure of 25 ug/m(3) during work hours for the 576 workers hired in or after 1926. Thun et al. estimated that these workers worked an average of 3 years and consequently accumulated 1728 person-years of arsenic exposure. Thun et al. estimated, based on the OSHA risk assessment model for arsenic (48 FR 1864; Jan. 14, 1983), these arsenic exposures would account for only 0.77 lung cancers and consequently would not account for the excess lung cancer mortality in this cohort.

Thun et al. apparently based this estimate on OSHA's preferred estimate of lifetime risk of 40/1000 from a 45-year working lifetime of exposure to 50 ug/m(3) arsenic (48 FR 1890; Jan. 14, 1983), as follows:


[(1728 PY)*(500 ug/m(3))*(0.25)*(0.2)]/[(45 PY)*(50 ug/m(3))]*[40/1000]
     = 0.77 lung cancers.

OSHA has updated this calculation to reflect the additional followup of the cohort through 1984. OSHA has also adjusted the estimate of the average person-years work, and has made some additional analyses to account for the fact that the Thun et al. estimate of the number of lung cancers attributable to arsenic exposure represents the ultimate number after complete followup rather than after the partial followup currently available, and to incorporate additional information on arsenic exposures in the cohort.

Addressing first the estimate of the average person-years of work, Lamm (Ex. 144-8) calculated an average of 6.94 years of work at the plant per cohort member which is more than double the value of 3 years estimated by Thun et al. To resolve this discrepancy, OSHA asked Dr. Stayner to calculate the average duration of work for the cohort of men first exposed in or after 1926 based on followup through 1984. Dr. Stayner obtained an arithmetic mean of 7.15 years and a geometric mean of 3.04 years (Stayner, personal communication). Since the former figure is very close to the figure reported by Lamm and the latter figure is very close to the value reported by Thun et al., OSHA concludes that Thun et al. must have used the geometric average, whereas Lamm reported the arithmetic average. Since the average is used to calculate the total person-years of exposure reported, OSHA concludes that the arithmetic average is appropriate for this application and the geometric average is inappropriate. (The total person-years is equal to the product of the number of persons in the cohort times the arithmetic average of person-years of exposure, not the geometric average of person-years of exposure.) The arithmetic average of 7.15 years of exposure corresponds to a total person-years of exposure of 7.15*606 = 4333 person-years in the cohort reported on by Stayner et al. (Ex. L-140-20), where 606 is the number of persons in Stayner et al.'s cohort.

We now consider the adjustments to Thun et al.'s methodology needed to account for the fact that followup of the cohort currently is only through 1984 and consequently is incomplete. Since Thun et al. used OSHA's lifetime risk estimate for arsenic in their calculation, the 0.77 cancers represent the ultimate number of cancers in the cohort after complete followup, rather than the number of cancers through 1984. Stayner et al. (Ex. L-140-20) reported that only 27% of the cohort of workers hired in or after 1926 had died after followup through 1984 (162 out of 606). Consequently, the estimate of the ultimate number of lung cancers attributable to arsenic exposure should be multiplied by 0.27 to estimate the number of lung cancers attributable to arsenic exposure after followup through 1984.

Considering next the estimates of arsenic exposures, the average arsenic exposure of 500 ug/m(3) in high arsenic exposure areas that was used by Thun et al. in their calculation was based on arsenic measurements made in 1950. Two additional sources of data on arsenic exposures are also available: a 1945 report by the University of Colorado, Division of Industrial Hygiene on plant dust and fumes, including arsenic measurements (cited in Ex. L-140(23) and Ex. 144-8c), and urinary arsenic level measured in the high arsenic areas from 1960 to 1980 (Ex. 4-68). OSHA has reviewed the data from each of these sources and used these data to quantify arsenic exposures in the cohort.

In 1991 Dr. Schulte of NIOSH (Ex. 144-8c) used the 1945 exposure data to estimate average arsenic exposures in high arsenic exposure areas. Data tables contained in the 1945 report provide estimated 8-hour arsenic exposure values for 18 sample groups that were collected in the high arsenic exposure operations of sampling, mixing, roasting and calcine. Estimates of work shift exposure were calculated by combining the arsenic concentrations obtained from short-term samples that were collected while various tasks were being performed, or in the general areas where the worker was during the work shift. Based on these data, Dr. Schulte presented several summary measures of arsenic exposure, including both arithmetic and geometric averages. Both types of averages were calculated both unweighted and weighted by the number of workers in each operation. OSHA has substantially verified the NIOSH results for the weighted averages, which were 936 ug/m(3) for the weighted arithmetic mean and 157 for the weighted geometric mean. OSHA also concludes that the weighted means are more appropriate than the unweighted means since estimates of overall exposure should take into account the number of workers exposed to each level (e.g., if 100 workers are exposed to 100 ug/m(3) and one worker to 1000 ug/m(3), it is not reasonable to assume an average exposure of [100+1000]/2 = 550 ug/m(3) for the entire cohort.) OSHA also concludes that the arithmetic weighted average is a more appropriate summary measure for assessing risk than the geometric average because the arithmetic average is more likely to be closely correlated with lung cancer risk than the geometric average. (To cite an extreme example of the unreasonableness of the geometric average in this context, if a single sample value is zero, then regardless of the values of the remaining samples, the overall geometric average is zero.) In summary, based on Schulte's data, OSHA concludes that the weighted average of 936 ug/m(3) is the most reasonable summary measure of arsenic exposure in the high arsenic exposure areas from the 1945 data for purposes of risk assessment. Using the respirator adjustment factor of 0.25 introduced by Thun et al. results in a average exposure of 0.25 * 936 ug/m(3) = 234 ug/m(3).

Turning now to the urinary arsenic data, Thun et al. (Ex. 4-68) noted that the average urinary arsenic level measured in the high arsenic areas from 1960-1980 were consistent with an average inhaled arsenic concentration of 14 ug/m(3). If one assumes that the reduction in airborne arsenic levels paralleled the reduction in airborne cadmium levels over time in these areas (This does not require assuming that individual exposures to cadmium and arsenic were proportional since workers generally moved among several areas.), then the urinary arsenic data can be used in conjunction with the cadmium exposure estimates (Ex. 4-68, Table 1) to estimate the arsenic exposures prior to 1960. Since the (respirator-adjusted) cadmium exposures at the calcine furnace are estimated to be 0.15 ug/m(3) after 1965, 0.4 ug/m(3) between 1960 and 1964, and 1.5 ug/m(3) in earlier years, the estimate of arsenic exposure previous to 1960 at the calcine furnace is estimated to be between 14 * 1.5/0.4 = 52.5 ug/m(3) and 14 * 1.5/0.15 = 140 ug/m(3). Similarly, since the estimated cadmium exposures in the roaster area were 0.6 ug/m(3) after 1950 and 1.0 ug/m(3) before 1950, arsenic exposures in the roaster area are estimated as 14 ug/m(3) between 1950 and 1960, and 14 * 1.0/0.6 = 23.3 ug/m(3) prior to 1950. Since these estimates are based on urinary levels, they reflect what was actually inhaled and require no additional adjustment for respirator use.

In summary, OSHA has developed three sets of estimates of arsenic exposure in high arsenic exposure areas for the period prior to 1960. After adjusting for respirator use, these estimates are: 125 ug/m(3) based on 1950 monitoring data, which was used by Thun et al. to quantify arsenic exposure; 234 ug/m(3) based on 1945 monitoring data; and between 14 ug/m(3) and 140 ug/m(3) based on changes over time in cadmium exposures and urinary arsenic levels measured between 1960 and 1980.

We note that the original respirator-adjusted estimate of arsenic exposure of 125 ug/m(3) made by Thun et al. is within the range of the other estimates summarized in the previous paragraph. If this estimate of arsenic exposure is retained and the person-years of arsenic exposure are adjusted as explained earlier and the factor of 0.27 is added to account for the fact that most of the cohort members were still alive at the end of the most recent followup, the resulting estimate of the number of cancer deaths in the cohort by 1984 attributable to arsenic exposure becomes:


[(4333 PY)*(125 ug/m(3))*0.2]/[(45 PY)*(50 ug/m(3))]*[40/1000]*0.27
     = 0.52 lung cancer deaths.

Even if the highest OSHA estimate of average arsenic exposure (234 ug/m(3)) is used, instead of 125 ug/m(3), the estimated number of lung cancer deaths attributable to arsenic exposure is 0.97. On the other hand, Stayner et al. observed 24 lung cancer deaths and an overall excess of eight (including an excess of 11 among non-Hispanics only). Therefore, even though OSHA's analysis differs from that of Thun et al. in several respects, the overall conclusion is the same: it is unlikely that arsenic accounts for the excess lung cancer deaths observed in this cohort, or even for a substantial proportion of that excess.

Any error in the estimate of respirator use will result in uncertainty in the estimate of arsenic exposure for those members of the cohort who were employed in high arsenic exposure areas. If the arsenic exposure is underestimated the contribution of arsenic to the excess lung cancer risk would be underestimated. If, on the other hand, the arsenic exposure was overestimated, the contribution of arsenic to the excess lung cancer risk would be overestimated. As discussed in Section V on health effects, OSHA is of the opinion that Drs. Thun and Smith made reasonable estimates of both arsenic and cadmium exposures, but realizes that the actual certainty of the estimates will never be known.

Dr. Schulte of NIOSH (Ex. 144-8C) pointed out three factors that indicate that the average concentrations derived from the 1945 monitoring data above may be substantial overestimates of actual arsenic exposure.

"First, the 1945 report states that three of the departments with the highest arsenic concentrations (crusher, including unloading; Godfrey Roasters; and baghouse) 'in recent years have operated about one month in every two or three.' The TWA should be multiplied by 0.5 or 0.33 to estimate the actual average arsenic exposure to take this work pattern into account. Second, loading and unloading railroad cars are among the jobs with the highest exposure, but not only are these jobs done intermittently, they should not be equated with departmental averages throughout the departments where arsenic exposure potentially occurs. Third, particle size sampling was not done during the 1944 and 1945 surveys. Many of the samples taken were of dust, and thus a portion of the sample was of non-respirable size. This is especially true in the jobs involving loading and unloading, which included some of the highest arsenic values."

Dr. Thun (Ex. L-140-23) cites two additional reasons why the estimated average arsenic exposure of 500 ug/m(3) may be a substantial overestimate. One is that an estimate of the average exposure derived from the urinary arsenic data is so much lower. The other is the fact that high arsenic exposure jobs were frequently staffed with short-term employees who were not included in the study and whose arsenic exposures therefore did not affect its results.

Dr. Lamm (Ex. 144-8) disputed each of the factors used by Thun et al. to estimate arsenic exposures in this cohort. Lamm posits a mid-range value of 8,700 ug/m(3) (representing simply one-half of the highest recorded measurement), or 2650 ug/m(3) (obtained by "exclusion of the highest level and taking a mid-range of other high levels"). OSHA considers the weighted arithmetic average used in its analysis to estimate exposures from the 1945 data to be a more appropriate scientific approach than Lamm's use of a mid-range of certain selected "high levels." Since other data are available which indicate lower levels of arsenic and because of the potential for overestimation of arsenic exposure as described by Dr. Schulte and Dr. Thun, OSHA considers that, to the extent that its estimate of 936 ug/m(3) based on the 1945 data is in error, it is more likely to overestimate rather than underestimate arsenic exposures.

Dr. Lamm also questioned the estimate by Thun et al. that 20% of the person-years of exposure were spent in high arsenic exposure areas, citing that the controls in his case control study spent "more than two-thirds of their time (68%) in the high cadmium areas." However, Thun (Ex. L-140-23) explains that he and his co-investigators used several sources of information in making this estimate.

"The primary source of information was biomonitoring data which gave department numbers, and thus could be used to estimate the percent of time spent in high arsenic areas. We also used information provided in the doctoral thesis of Jeffrey Lee which included information on shift assignments. The 20% estimate was discussed with Dr. Lowell White of ASARCO and with Mr. Ernie Lovato, president of United Steelworkers Local 557, who confirmed the 20% estimate. Finally, personnel records showing two-week shift assignments were reviewed to see if they confirmed the 20% estimate."

Dr. Thun also points out that Lamm's 68% figure does not contradict this estimate because, first, Lamm's figure is for high cadmium areas, not high arsenic areas and, second, Lamm's figure is derived from a small sample of workers from his case control study (comprising only 12% of the cohort).

The Dry Color Manufacturers' Association (DCMA) commented that Thun et al. underestimated the percentage of the workforce exposed to 'high arsenic' work areas by assuming only 20% were so exposed when in reality everyone was exposed because entry-level positions were in these high-exposure areas (Ex. 120). This comment appears to be based on a misconception regarding what the 20% figure is supposed to represent. It represents the percent of person-years spent in high-arsenic exposure areas, not the percent of workers that ever worked in these areas. Even if DCMA were to be correct in its claim that 100% of workers were exposed in high arsenic areas, this would in no way contradict the 20% estimate made by Thun et al.

Dr. Lamm contends that the "arsenic content of the fines used as feedstock prior to 1940 were considerably higher than those used after 1940" (Ex 144-7B). Lamm presents a graph of mean arsenic in the feedstock from 1926 through 1958 (Ex. 144-7, Figure 5) that indicates levels were two to three times as high between 1926 and 1940 as between 1940 and 1958.

However, Thun (Ex. 33) presents a graph of total pounds of arsenic processed by year from 1924 through 1958 that indicates no particular trend after 1926. In fact, the highest amount of arsenic processed between 1927 and 1958 appears to have occurred in 1953. The lack of a difference in amount of arsenic processed during the two time periods is corroborated by Table 8 in Lamm et al. (Ex. 144-7) that indicates the average arsenic produced was 5,150 kg/yr between 1928 and 1940 and 4,356 kg/yr between 1941 and 1958. OSHA believes that it is questionable as to whether the amount of arsenic processed or the arsenic concentration of the ore is a better indicator of worker exposures. It seems reasonable that the amount of fugitive dust present would be likely to be related to the total amount of ore processed and not to the arsenic content. If so, then the total amount of arsenic processed is likely to be a better predictor of the total amount of arsenic released into the air. At the outside, these data do not suggest a large difference between the arsenic air concentrations between 1926-1940 and post-1940. Thus, OSHA finds Dr. Lamm's argument to be less than compelling and concurs with Dr. Thun's conclusion that "[t]hese data [on amount of arsenic processed by year] do not support the hypothesis that arsenic contamination was markedly higher from 1926-1940 than thereafter" (Ex. 33).

Dr. Lamm indicates that he calculated, based on computerized work histories supplied by NIOSH, an average employment of members of the Thun cohort of 6.94 years as opposed to the 3 years used by Thun et al. in their arsenic calculation. As explained earlier, based on the followup through 1984, Dr. Stayner of NIOSH recently calculated an arithmetic average employment duration of 7.15 years which agrees substantially with the value calculated by Dr. Lamm. This figure of 7.15 years was incorporated by OSHA into its calculations as described earlier.

Finally, Dr. Lamm contends that Thun et al. were in error in reducing exposures by 75% to account for respirator use. Lamm argues that respirator use was not considered in OSHA's risk assessment of arsenic and therefore not using a respirator reduction factor would be equivalent to assuming that respirator protection in this cohort was comparable with that in the cohorts upon which OSHA's risk assessment for arsenic was based.

But OSHA relied mainly on data from the Anaconda, Montana and the Tacoma, Washington copper smelters in its risk assessment for arsenic. Arsenic exposures at the Tacoma smelter were estimated from urinary arsenic measurements and based on correlations of air measurements and urinary arsenic levels collected from workers who did not wear respirators during the time the data were being collected (48 FR 1878, Jan. 14, 1983). Air levels estimated from such measurements thus would reflect actual inhaled doses and further correction for respirator use would not be appropriate. As stated by OSHA in connection with the final arsenic rule, "they [estimates of exposure in the Tacoma smelter based on urinary arsenic data] take into account the protection afforded by the respirators that were sometimes worn." (48 FR 1887, Jan. 14, 1983).

Respirator use was also accounted for in some analyses of data from the Anaconda smelter: "Because respirators generally were worn in the heavy exposure areas, Lubin et al. reduced the assigned exposure level to 1.13 mg/m(3) in the heavy exposure category for some of their multivariate analyses" (48 FR 1870, Jan. 14, 1983). Moreover, some of the risk assessments of data from the Anaconda smelter upon which OSHA relied omitted data from the heavy exposure category [Ex. 206 ref. in 48 FR 1864, Jan. 14, 1983). This constitutes a de facto correction for respirator use, because the group most likely to have used respirators were not included in the analysis.

On the other hand, the exposure estimates in the epidemiological study of the Anaconda smelter by Higgins et al. did not include adjustments for respirator use because the authors did not believe such adjustments were appropriate. Thus, some of the risk assessments cited by OSHA included adjustments for respirator use and at least one did not. The estimates of lifetime risk from arsenic exposure preferred by OSHA were consistent with estimates made from studies cited above in which adjustments for respiratory use were either made or judged not to be appropriate.

It should also be noted that, like the estimates of arsenic exposure made in the Tacoma arsenic smelter based on urinary data, estimates of arsenic exposure at the Colorado cadmium smelter based on the urinary arsenic data (discussed earlier in this section) did not require nor employ any adjustment for respirator use. Nevertheless, estimates of arsenic exposure derived from these data were generally lower than estimates made from other types of data for which adjustment for respirator use was deemed necessary and therefore employed.

After review and modification of the Thun et al. analysis of the potential effect of arsenic and consideration of the above points, OSHA concludes that Dr. Lamm's contention that Thun et al. seriously underestimated the potential impact of arsenic upon lung cancer in this cohort is not convincing. OSHA's preferred estimate of the expected number of deaths from lung cancer due to arsenic in the cohort studies by Stayner et al., which were first exposed in 1926 or later, is 0.52 out of a total of 24 that have been observed. Further, based on its analysis described above, OSHA believes that it is unlikely that the number could be substantially larger than this preferred value.

Potential for Confounding by Smoking in the Thun Cohort

In addition to arsenic exposure, tobacco smoking is also a risk factor for lung cancer in the Thun cohort. As explained by Dr. Thun in his testimony on the proposed rule (Ex. 33), smoking information was available for 43% of the workers in the study. This information was collected by the company in 1982 using a questionnaire mailed to surviving workers or next of kin of deceased workers and using medical records. The questionnaire inquired about the ages of starting and stopping smoking and the usual amount smoked. Dr. Thun compared the smoking habits of the cadmium workers as of July 1, 1965 to smoking habits for U.S. white males in 1965 as determined by a Health Interview Survey (HIS) of the National Center for Health Statistics. Table VI-13 was used by Dr. Thun to summarize these data. This table compares smoking habits of Hispanic and non-Hispanic workers separately to smoking habits of U.S. white males.


TABLE VI-13. - SMOKING HABITS AMONG HISPANIC AND NON-HISPANIC WORKERS AND
                AMONG WHITE MALES IN U.S. GENERAL POPULATION
                                [In percent]
____________________________________________________________________________
                             | Non- and  |        Current smokers
            Group            |  former   |__________________________________
                             | smokers   |    < 15  |   15-24   |   >25
_____________________________|___________|_________|___________|____________
                             |           |         |           |
U.S. Males...................|     48.5  |   13.3  |    24.1   |   14.1
Cadmium Workers:             |           |         |           |
   Hispanic..................|     43.7  |   31.0  |    20.7   |    4.6
   Non-Hispanic..............|     50.9  |   13.5  |    21.5   |   14.1
_____________________________|___________|_________|___________|____________
  SOURCE:  Thun (Ex. 33).

Although Dr. Thun cautioned that smoking information was collected differently from the cadmium workers than in the 1965 HIS and therefore may not be comparable, the smoking pattern for non-Hispanics is very similar to the smoking habits for U.S. white males in general. These data do not suggest that differences in smoking habits could have accounted for the 11 excess lung cancers (SMR = 211, p < 0.01) observed by Stayner et al. (Ex. L-140-20) among non-Hispanics. If anything, they suggest that smoking rates may have been slightly lower among non-Hispanics than in the referent group of U.S. white males used by Stayner et al. which could cause the effect of cadmium to be underestimated in the life-table analysis.

The smoking data on Hispanics in Table VI-13 indicate that the Hispanic workers were more likely to be current smokers, but smoked fewer cigarettes per day. Dr. Thun pointed out that this pattern is typical of Mexican-American men. Dr. Thun also pointed out that "[s]everal studies have shown low rates of lung cancer among Hispanics in the Southwest", and "[i]n Denver between 1969-1971, Hispanic males had less than half (0.41) the incidence of lung cancer compared to other whites." Thus the lower mortality from lung cancer among Hispanics found in this study is consistent with generally lower rates among Mexican-American men. These findings support the use of U.S. white males as the referent population for non-Hispanics in this study, as was used by Stayner et al. in their life-table analysis, but suggests that this referent population has a higher background rate of lung cancer than the population of Hispanics in this study. This could have caused Stayner et al. to underestimate the effect of cadmium among Hispanics in their life-table analysis.

As noted by Stayner et al. (Ex. L-140-20), the modelling procedures they used based on Poisson regression and Cox regression greatly reduced the potential for confounding by cigarette smoking. This is because these modelling procedures rely on internal comparisons within the cohort rather than upon comparisons with an external control population. Consequently, "[i]n order for smoking to confound this analysis, one would have to propose that smoking habits varied between the exposure categories used in the analysis, which seems unlikely." Moreover, Dr. Thun showed (Ex. 33) that the percent of Hispanics, which may be acting as a partial surrogate for cigarette smoke, was evenly represented in the low, medium and high cadmium exposure groups.

In summary, the smoking data that are available indicate that smoking patterns among non-Hispanics in this cohort were comparable with those of U.S. white males in general and therefore cannot explain the significant excess of lung cancers among non-Hispanics in this cohort. The smoking pattern among Hispanics in the cohort indicated that the Hispanic smokers tended to smoke less than U.S. white males. This pattern is consistent with generally lower smoking rates among Mexican-American men and may explain at least partially the lower lung cancer rates observed among Hispanics in this study. In neither Hispanics nor non-Hispanics did the smoking data indicate a higher than normal prevalence of smoking that could account for the observed increased lung cancer mortality rate in this cohort. Moreover, even if the rate of smoking in this population had been different from that expected among any comparable mixture of Hispanics and non-Hispanics, such differences would be unlikely to affect the statistical analysis conducted by Stayner et al. that did not make use of an external control population.

Summary of OSHA's Review of Potential Confounders in the Thun Cohort

OSHA has carefully considered the evidence regarding whether the excess lung cancers in the Thun cohort can be attributed to arsenic exposure, and particularly to arsenic exposure prior to 1940. The analysis presented by Dr. Schulte of NIOSH demonstrates that the increased lung cancer mortality in this cohort is not confined to workers first employed prior to 1940. Similarly, results obtained by Stayner et al. were not consistent with the increased lung cancer in the cohort being a result of arsenic exposure that occurred prior to 1940. Finally, even though there are several reasons stated by Dr. Schulte and Dr. Thun as to why the analyses conducted by Thun et al. and modified by OSHA could overestimate the effect of arsenic exposure, these estimates indicate that arsenic exposure does not explain the increased lung cancer mortality observed in this cohort. OSHA also finds no evidence of excess smoking in the population that could explain the excess of lung cancers. Thus, OSHA concludes that the excess in lung cancer among workers in the Thun cohort who were first employed in or after 1926 is unlikely to be due to arsenic exposure or to cigarette smoking and is more likely to be attributable to cadmium exposure. This conclusion is strengthened by the knowledge that lung cancer has unequivocally been related to inhalation of cadmium in animals.

Review of Other Issues

The Low SMR in the Low Exposure Group in the Thun Cohort

Several commenters (Exs. 38, 19-30), noted the low SMR (SMR = 53, based on two lung cancer deaths) in the low exposure group (< 584 mg-days/m(3)) of the Thun et al. study (Ex. 4-68) and suggested that this was evidence of a threshold for the carcinogenic effect of cadmium. Dr. Thun addressed this issue in his testimony at the OSHA hearing (Ex. 33) by presenting results for followup through 1984 and by calculating the expected number of lung cancers using both U.S. and Colorado rates. His findings with respect to the low exposure group (< 584 mg-days/m(3)) are summarized in Table VI-14. In none of the four cases is the deficit in lung cancers statistically significant. Moreover, the life-table analysis of this cohort by Stayner et al. (Ex. L-140-20) demonstrates that the deficit is also not statistically significant when considering only non-Hispanics (1 case observed and 3.35 expected, p = 0.15, see Table VI-7). Further, the deficit in the lung cancer rate in the low exposure group is smaller when Colorado mortality rates are used (Table VI-14). Moreover, as noted by Dr. Thun (Ex. 33), several studies have shown low rates of lung cancer among Hispanics in the Southwest (possibly as a result of reduced smoking among Hispanics) and 40% of the cohort are Hispanics whereas in 1980 only 6.5% of Colorado males were Hispanic (USDOC, 1980). OSHA also suggested that this could be due to a healthy worker effect. The healthy worker effect is seen in worker cohorts because the health status of people who are accepted for employment is better than the health status of the general population used for comparison (Ex. 50). Thus, OSHA concludes that the deficit in lung cancers in the low exposure group may be attributed to: (1) Random fluctuation, (2) the healthy worker effect, (3) differences between U.S. and Colorado lung cancer rates, and (4) an excess of Hispanics in the cohort, and does not indicate a threshold for the carcinogenic effect of cadmium. Similarly, Dr. Thun (Ex. 33) expressed the opinion that the apparent deficit of lung cancers among workers exposed to less than 584 mg-days/m(3) (the low exposure group) "is an artifact of the lower smoking habits of the cadmium workers and should not be interpreted as showing a 'safe' level of cadmium."


TABLE VI-14. - EXPECTED AND OBSERVED NUMBERS OF LUNG CANCERS IN THE LOWEST
  EXPOSURE GROUP (<  or = 584 mg-days/m(3)) OF THE THUN COHORT (EX. 33)
  HIRED IN OR AFTER 1926
_____________________________________________________________________________
                     |  Based on rates of U.S.   |     Based on rates of
                     |      white males          |   Colorado white males
    Followup period  |___________________________|___________________________
                     |        O/E(a)      | SMR  |    O/E(a)          | SMR
_____________________|____________________|______|____________________|______
                     |                    |      |                    |
Through 1978.........| (b)2/3.76 (p=0.28) | 0.53 | (b)2/2.64 (p=0.51) | 0.76
Through 1984.........| (b)2/6.06 (p=0.06) | 0.33 | (b)2/4.37 (p=0.19) | 0.46
_____________________|____________________|______|____________________|______
  Footnote(a) Observed/expected.
  Footnote(b) Probability of observing two or fewer lung cancers assuming
lung cancers are Poisson distributed with mean given by the respective
expected number.

Lamm et al.'s Case Control Analysis

Lamm et al. (Ex.144-7B) conducted a case control analysis of 25 lung cancer cases in the Thun cohort. Three controls per case were selected from this cohort. Controls were matched to cases by date of hire and age at hire. Explanatory variables were cumulative cadmium exposure, cigarette smoking history, and "plant arsenic exposure status at time of hire."

Lamm et al. stratified his cadmium exposures by period of hire (pre-1926, 1926-1939, and 1940-1969). Within each period mean cumulative cadmium exposure was nearly equal among cases and controls (ratio of exposure-cases/controls - was 0.99, 1.05, and 0.91, for the three periods of hire, respectively). Among those for whom smoking histories were available (72% of the cases and 57% of the controls), the odds ratio for smoking was 8.2 (p=0.046). Lamm et al. interpreted these findings as suggesting "that the cumulative cadmium exposure .... is not a major determinant of lung cancer within the study group."

Lamm et al. matched for both age and date of first hire in their case control study. It seems likely that date of hire could be significantly correlated with cadmium exposure in this analysis because higher cadmium exposures would be expected to occur in persons with earlier periods of employment, both because estimated airborne cadmium levels were higher in earlier years and because persons employed earlier had longer to accumulate exposure. Such a correlation could mask any effect of cadmium exposure in Lamm et al.'s analysis because they only compared the exposures of cases and controls for those that had similar dates of hire. Lamm et al. raised the possibility that such "overmatching" may have been the reason there was no difference in cadmium exposures between cases and controls. They point out that a considerable range of exposures exist "in this study, ranging from about 1 to 30 mg/m(3) for those hired prior to 1940 and from about 0.3 to 17 mg/m(3) for those hired subsequently." However, since each of the ranges are determined by the exposures of only two members of the cohort (the most highly exposed and the least exposed in each group), presentation of these ranges does not adequately address the issue of whether or not the cadmium exposure and age at first hire are highly correlated.

Because of the potential flaw due to "overmatching" in the methodology used by Lamm et al., OSHA does not accept the conclusion from this study that "exposure to arsenic and cigarette particulates, rather than cadmium exposure, may have caused the lung cancer increase of these workers." NIOSH demonstrated a significant excess of lung cancer in medium and high cadmium exposure groups among workers first hired in 1940 and later, when Lamm et al. contend arsenic exposures were lower (Ex. 144-8c). Similarly, in Stayner et al.'s analysis (Ex. L-140-20) the magnitude of the cadmium variable increased rather than decreased when they controlled for first employment prior or subsequent to 1940. Finally, the Thun et al. analysis of the potential impact of arsenic, as reviewed and modified by OSHA earlier in this section, indicates that arsenic exposures would account only for a very small fraction of the total lung cancers observed in this cohort.

Disaggregation of Data and Use of Multistage Model

Dr. Thomas Starr commented that "OSHA should redo its analysis with the individual person-year information collected by Thun et al.," noting that the latter data provided just three data points, and they are only very crudely characterized by the median cumulative exposure for each category" (Ex. 38). He also recommended that "OSHA should reanalyze the Thun et al.," (Ex. 4-68) data with the multistage dose-response approach model utilizing an approach similar to that described by Crump and Howe (1984)." The risk assessment by Dr. Stayner and his associates at NIOSH (Exs. L-140-20, L-163) incorporates both of these recommendations. The Poisson regression conducted by Stayner et al. divided the person-years into about 200 cells, defined by categorizations on age, calendar year, ethnicity (Hispanic versus non-Hispanic), and cumulative cadmium dose. The Cox regression conducted by Stayner et al. did not involve any grouping of person-years. Finally, Dr. Stayner et al. did reanalyze the Thun et al. data, using the followup through 1984 rather than that through 1978 that was reported in Thun et al. (Ex. 4-68), based upon the multistage dose response approach described by Crump and Howe (1984). These methods all gave very similar estimates of excess risk (Table VI-9), and are very similar to OSHA's estimates (Table VI-12) in its final risk assessment that were derived from more aggregated forms of the data.

Summary of Cancer Risk Assessment Based on Both Animal and Human Data

In its proposed rule, OSHA assessed cancer risk from occupational exposure to cadmium by applying several different risk assessment models to the animal data of Takenaka (Ex. 4-67) (results shown in Tables VI-1 and VI-2) and by applying relative and absolute risk models to the data on followup through 1978 of the Thun cohort (results shown in Tables VI-3 and VI-4).

Since writing the proposed rule significant new data have become available that prompted OSHA to conduct additional risk calculations. Oldiges et al. (Ex. 8-694-D) showed that inhalation exposure to several types of cadmium caused lung tumors in male and female rats. Stayner et al. (Ex. L-140-20) reported on additional followup of the Thun cohort through 1984 and developed new estimates of risk from this update.

OSHA maintains that the animal carcinogenicity data on cadmium is relatively extensive and of high quality. It shows unequivocally that several different forms of cadmium induce lung tumors when inhaled by male or female rats. These data are satisfactory for establishing estimates of human risk according to toxicological methods generally employed by federal regulatory agencies, including OSHA.

OSHA has carefully evaluated the human data from the Thun cohort and the likelihood that the excess lung tumors were caused by either smoking or arsenic. It concludes that these tumors are unlikely to be explained by either of these factors, but are more likely attributable to cadmium. OSHA further concludes that the human data on cadmium from the Thun cohort are substantially strengthened and corroborated by the animal data. First, the animal data show unequivocally that inhalation of cadmium can induce lung tumors, the same type of tumors that are seen in excess in the Thun cohort are in essential agreement.

In its new risk estimates based on the recently developed animal data (Table VI-6), OSHA utilized three dose-response models, all of which are different versions of the multistage model of cancer: the multistage model, the Armitage-Doll multistage model and the multistage-Weibull model. The Armitage Doll model makes more detailed biological assumptions regarding the effect of exposure and will always predict a linear response. The multistage and multistage-Weibull models are flexible dose response models that can assume a variety of curve shapes, both linear and non-linear. However, upper confidence limits calculated from these models will be linear. Furthermore, when applied to data with only two dose groups (a treated group and a group of control animals), the multistage and multistage-Weibull models become linear by default, because there is not enough information from only two doses to estimate the shape of the dose response. The multistage model is applied to the quantal data specifying whether an animal had a lung tumor response, and the Armitage-Doll and multistage-Weibull model also utilized the time at which the response was observed.

Regarding the cancer risk assessment models applied to epidemiological data, OSHA has reviewed the new risk assessment by Stayner et al. based on the updated followup through 1984 of the Thun cohort, and has also updated its own risk assessment utilizing this new information. Stayner et al. employed three dose response models: the power model, the exponential model, and the additive relative rate model. Although all three of these models are linear at the very lowest cadmium exposures, they have different amounts of non-linearity at higher doses, with the power model being the most nonlinear and the additive relative rate model being essentially linear. Although the power model gave the best fit tot the data, this model predicted unrealistically low rates of lung cancer in unexposed subjects and consequently was discounted by Stayner et al. [This model would also predict much higher lung cancer risks than the other two models at exposures in the range OSHA is considering regulating (1ug/m(3)-100 ug/m(3)).] Between the two remaining models, the additive relative rate model fit the data slightly better than the exponential model and was selected by Stayner et al. as the best model of the three for risk assessment. OSHA believes that this is a reasonable scientific conclusion from the results of the Stayner et al. analysis.

Dr. Stayner (EX. L-163) also reported on results from applying the multistage model to the updated thun cohort. Application of this model to the Thun cohort had been recommended by Dr. Starr (Ex. 38). This model assumes that lung carcinogenesis is a five-stage process. Dr. Stayner estimated that cadmium affects the third stage in this process.

Dr. Starr also criticized OSHA's use of "purely linear models of absolute and relative risk." He stated that "Today, it is nearly universally accepted that the process of carcinogenesis, and chemical carcinogenesis in particular, is multistage in character with numerous distinct events required for the conversion of normal cells to malignant ones. Thus, the use of the one-hit model (or linear models derivable from it) for low-dose purposes are wholly inadequate."

OSHA agrees that there is good scientific evidence that carcinogenesis is generally a multistage process. However, OSHA disagrees with Dr. Starr's conclusion that a multistage process is incompatible with a linear dose response. It is plausible that a carcinogen affects a multistage process by increasing the rates at which the different stages occur by an amount that is proportional to the amount of the ultimate carcinogen present. If this is the case, then the dose response shape is determined mainly by two factors: the number of stages affected by the carcinogen (not the total number of stages as implied by Dr. Starr), and the interaction of the carcinogen with background carcinogenesis. If the carcinogen acts on a single stage, then the dose response will be linear. Even if the carcinogen affects multiple stages, if its effect is to increase the background rate at which these stages occur, then the process can still generate essentially a linear dose response (Crump, 1985).

Dr. Stayner's application of the multistage model to the data from the thun cohort provides direct evidence that the multistage model is compatible with a linear dose response. He applied a multistage model assuming five stages and one of the stages was affected by cadmium. The excess risks he estimated from this model were intermediate between those he obtained with two versions of the relative rate model, which is a linear model (Table VI-9).

In its updated risk assessment based on followup; through 1984 of the Thun cohort. OSHA applied a linear version of the relative risk model. OSHA determined using a goodness-of-fit test (Table VI-11) that this linear model was completely compatible with the dose response data from this cohort.

Three of the ten animal data sets analyzed by OSHA using the multistage model (Table VI-6(, and seven of the nine animal data sets analyzed using the multistage-Weibull model, involved three or more treatment groups and therefore provided information on the shape of the dose response curve. In none of these ten model fits did a non-linear version of the model fit significantly better than a linear model. (Most significant of the ten p-values for a test of departure from linearity was p = 0.17.) In all but two of these ten cases, the best-fitting model was linear . [Even if the true dose response is linear, the probability is only one-half that the best fitting model to a data set will be linear (Crump et al., 1977).] Moreover, the Armitage-doll model, which is a linear model, provided an adequate fit to all seven data sets in which there were three or more treatment groups, based on a standard chi-square goodness of fit test.

OSHA concludes that a linear dose response model is compatible with both the animal and human data on lung cancer in cadmium-exposed cohorts, and that it is also compatible with multistage mechanisms of carcinogenesis.

OSHA believes that the theoretical understanding of cancer is not sufficient to unequivocally establish a single dose response function for cadmium-induced lung cancer. In view of this uncertainty, the fact that linear models are compatible with both the current theoretical understanding of cancer and the animal and human data on cadmium, and the fact that inappropriate application of a non-linear model could seriously underestimate human risk, OSHA maintains that in order to be compatible with current scientific understanding of carcinogenesis and cadmium dose response data, and to satisfy its mandate to protect worker health, OSHA should give particular consideration to estimates of cancer risk obtained from linear models.

Among the various linear models applied to either the animal and human data, OSHA is unable to select a single one as being most appropriate. However, OSHA does not believe that this is a serious problem because the different linear models generally provided similar estimates of risk when applied to the same data.

All of the risk estimates derived from the Thun cohort (Tables VI-9 and VI-12) are based upon linear models. All of the risk estimates derived from animal data (Table VI-6) are based upon linear dose responses except the fits of the multistage-Weibull model to the Oldiges, et al. data on male rats exposed to CdCl(2) or to CdO fume.

Comparing the estimates of excess risk of lung cancer made from animal data (Table VI-6) with those made from human data, OSHA finds that they are in essential agreement. Estimates based upon human data of the excess risk from lung cancer from 45 years of exposure at a TWA of 5 ug/m(3) range from three excess deaths per 1000 workers to nine excess deaths per 1000 workers (Tables VI-9 and VI-12). The corresponding ranges in excess risk made from the three models applied to animal data are as follows: multistage model, 0.061-28 excess deaths; Armitage-Doll model, 3.3-38 excess deaths; multistage-Weibull model, 0.095-35 excess deaths. The corresponding ranges for numbers of excess deaths estimated per 1000 workers from 45 years of exposure to a TWA of 100 ug/m(3) are as follows: Human data, Thun cohort, 58-157 excess death; animal data, multistage model, 24-433 excess deaths, animal data, Armitage-Doll model, 6.8-726 excess deaths, animal data, multistage-Weibull model, 37-512 excess deaths. Although estimates made from animal data tend to be somewhat higher than those made from human data, the ranges of risk obtained from each of the three models applied to animal data overlap the range obtained from the human data. The excess risks from exposure to CdO-fume estimated using animal data are somewhat smaller than the risks estimated from the other forms of cadmium; Dr. Oberdorster hypothesized that this could be attributed to a lower uptake by the lung of the inhaled fume (Ex. 141). The ranges of risks from the animal data are much narrower if the CdO data are omitted. However, the resulting ranges still overlap the estimates obtained from the human data.

It may be that the estimated obtained from human data are somewhat more reliable than those obtained from the animal data because the former do not involve the uncertainty of cross-species extrapolation. However, the similarity of results from animal and human studies serves to strengthen OSHA confidence in estimates obtained from both types of data.

Based on human data, OSHA's preferred estimates of the excess lung cancer risk from 45 years of occupational exposure to cadmium are 58 to 157 excess deaths per 1000 workers from exposure to a TWA of 100 ug/m(3) and three to nine excess deaths per 1000 workers from exposure to a TWA of 5 ug/m(3). OSHA notes that estimates of excess deaths exceed one death per thousand workers even at a TWA exposure of 5 ug/m(3).

Risk Assessment for Kidney Dysfunction

In its proposed rule, OSHA quantified the risk of kidney dysfunction due to cadmium exposure using the study by Falck et al. (Ex. 4-28) of workers at a refrigeration compressor production plant and the study by Ellis et al. (Ex. 4-27) of workers at the cadmium smelter in Colorado. OSHA has since identified four additional studies that contain useful quantitative information on kidney effects from cadmium exposure. These are: (1) the study by Elinder et al. (Ex. L-140-45) of a cohort of 60 workers who had previously been exposed to cadmium through welder fume and dust associated with the use of cadmium soldiers; (2) the study by Jarup et al. (Ex. 8-661) of 440 workers exposed to cadmium at a Swedish Battery plant; (3) the study by Mason et al. (Ex. 8-669) of 75 workers exposed to copper-cadmium alloy in a factory in the United Kingdom; (4) the study by Thun et al. (Ex. 19-43B), which is based on the same population of smelter workers studied earlier by Ellis et al. However, the data reported by Thun et al. (Ex. 19-43) are not in a form that is suitable for quantitative modelling and consequently data from this study were not modelled by OSHA.

Table VI-15 summarizes some essential features of these studies. Many differences are noted in these studies, including size of study; type of cadmium; type, extent and quality of exposure data; type of urine sample; control of samples for pH; and definition of kidney dysfunction. Each of these differences could result in differences in quantitative estimates obtained from these studies.


    Features of Cadmium Studies of Kidney Effects

(For Table VI-15, Click Here)

In each of these studies, data were obtained on the cumulative cadmium exposure of each subject and on the concentration of a small molecular weight protein in urine [B2-microglobulin or retinol binding protein (RBP)], an excess of which is considered evidence of kidney dysfunction. Data on the protein levels in urine were available in two distinct forms. First, in each of the published reports of these studies, a cutoff level was specified that defined the boundary between normal individuals and those with proteinuria. The information available in these published reports was in terms of the "zero-one" variable that specified whether or not measured protein levels in a subject exceeded (coded as "one") or did not exceed (coded as "zero") the cutoff level. Such data are called quantal data. Second, the unpublished concentrations of marker proteins in urine from the Mason et al. study were made available to OSHA by Mr. Mason (Ex. 12-45). These type of data are called continuous data. Different types of statistical models are required for these two types of data.

The Office of Budget and Management (OMB)(Ex. 17-D) questioned OSHA's use of quantal data from the Falck et al. and Ellis et al. studies for risk modelling in the Proposed Rule. In both of these studies, data were only available in quantal form as presented in the published papers. Thus, OSHA had no choice over either whether to use quantal data in its modelling or the cutoff used to define kidney dysfunction in these studies. In the risk modelling to be presented below, OSHA reanalyzes the quantal data from the Falck et al. and Ellis et al. studies by Elinder et al., Jarup et al. and Mason et al. In all of these studies except Mason et al., data were available only in a quantal form derived from a cutoff selected by the original authors for defining kidney dysfunction, However, since OSHA had access to the unpublished continuous data from Mason et al. study, OSHA was able to explore the effect of different cutoffs for kidney dysfunction upon the analysis. (Such an analysis was suggested by OMB, Ex. 17-D). OSHA also applied models appropriate for continuous data to the continuous data of Mason et al.

Risk Assessment Based Upon Quantal Data

In the proposed rule, OSHA presented quantitative estimates based on the logistic model, which can be expressed as


Ln{P(X)/[1-P(X)]} = alpha + t (*) Ln(X)

or, equivalently,

P(X) = e(alpha) X(t)/(1 + e(alpha) (*) X(t))

where

Ln           indicates the natural logarithm,
X            is cumulative occupational exposure in ug-years/m(3)
P(X)         is the probability of kidney dysfunction in a person with a
             cumulative cadmium exposure of X, and
alpha and t  are regression parameters estimated from the data(5).


__________
  Footnote(5) In the proposed rule, B was used in place of t in the
logistic regression model.  However, both the proposed rule and the
present document also used B as a potency parameter in the cancer models
(i.e., B multiplies cumulative cadmium dose, X, in the cancer models)
whereas the parameter designated as B in the logistic regression model in
the proposed rule is a shape parameter (i.e. cumulative dose, X, is raised
to the power B).  To avoid confusion, the symbol for this shape parameter
has been changed to t.

This logistic regression model is widely used to model quantal responses (in this case 0 = normal kidney function and 1 = abnormal kidney function) as a function of one or more explanatory variables (in this case cumulative exposure to cadmium). When used to define a dose response as in the present situation, it provides a flexible model that is able to accommodate a wide range of dose response shapes ranging from threshold- like (when the parameter t is large) to a shape that is linear (t = 1) or even supralinear (t < 1) at low doses.

The model as formulated above cannot accommodate any level of kidney dysfunction in persons without cadmium exposure [since the model definition implies the restriction P(0) = 0]. However, kidney dysfunction is not restricted to persons with prior cadmium exposure. (In several studies the cutoff used to define proteinuria was either an upper quantile, or a statistical upper confidence limit, on the level of the maker protein in unexposed subjects, which implies that a quantifiable number of unexposed subjects would have urinary levels above the cutoff; e.g., Mason et al. defined their cutoff as the 95% upper quantile on the level of retinol binding protein in unexposed subjects; consequently, by definition, a background approaching 5% would be expected.) The need for incorporating background response in the model was noted by several commenters. Dr. Starr pointed out this problem in his oral testimony, noting that "some adjustment must be made in the analysis procedure to account for that background response rate, which is not attributable to cadmium exposure"(Tr. 6/8/90). ENVIRON Corp. (Ex. 19-43G) indicated that, because of this (unexposed workers identified as having kidney disease), OSHA should use an independent background parameter that properly accounts for this false positive response rate. Mr. Edwin Seeger, representing the Cadmium Council, pointed out that (Ex. 19-43) the Council believes that OSHA's renal dysfunction risk estimate is biased high at low dose levels because it does not take account of the "normal incidence of elevated B(2)-microglobulin levels in the worker population even in the absence of cadmium exposure."

OSHA has therefore modified the logistic model in this revised quantitative risk assessment to include a background term. The modified model is


P(X) = (delta + e(alpha*) X(t))/(1 + (alpha*) X(t)),

where delta is the probability of kidney dysfunction among persons not exposed to cadmium (delta = P(0)). Thus, this revised model is able to accommodate kidney dysfunction in persons unexposed to cadmium by estimating a positive value for delta.

This model has been applied to quantal data from five of the six of the studies described above. The Thun et al. study was excluded because the data were not reported in a form that is amenable to this type of analysis. The model was fit to the data from Elinder et al. (Elinder et al. Table II) and Jarup et al. (Jarup et al. Table 1), in which the subjects in the study were grouped according to their cumulative cadmium exposure. These data sets are listed in Table VI-16. Data for the Ellis et al. study used in the analysis were read from Figure 3 of the Ellis et al. study (Ex. 4-27) by OSHA and also an OSHA contractor (Ex. 16-B). Both data sets were analyzed and similar results were obtained. Results obtained from the OSHA reading are reported on herein (Ex. 4-27-A).


TABLE VI-16. - KIDNEY DYSFUNCITON VERSUS CUMULATIVE CADMIUM EXPOSURE FROM
                JARUP ET AL. (1988) AND ELINDER ET AL (1985)
________________________________________________________________________
  Cumulative Cadmium Exposure (ug-years/m(3))    |  Number with kidney
_________________________________________________|  dysfunction/number
        Range          |       Midvalue          |      examined
_______________________|_________________________|______________________

                         Jarup et al. (1988)(a)
________________________________________________________________________
                       |                         |
<359...................|                131      |          3/264
359-< 1710..............|                691      |           7/76
1710-< 4578.............|               3460      |          10/43
4578-< 9458.............|               6581      |          10/31
9458-< 15,000...........|             12,156      |           5/16
15,000+................|             21,431      |           5/10
_______________________|_________________________|______________________

                         Elinder et al. (1985)(b)
________________________________________________________________________
                       |                         |
<1000..................|                500      |           3/16
1000-< 2000.............|               1500      |           7/22
2000-< 3000.............|               2500      |            4/9
3000-< 5000.............|               4000      |            5/8
5000+..................|               7500      |            5/5
_______________________|_________________________|______________________
  Footnote(a) Definition of kidney dysfunction: B(2)-microglobulin
> 310 ug/g creatinine.
  Footnote(b) Definition of kidney dysfunction: B(2)-microglobulin
> 300 ug/g creatinine.

Data from the Falck et al. (Ex. 4-28) study came from Table III of that paper. Falck et al. omitted three subjects, all of whom had elevated B(2)-microglobulin in their spot kidney sample, from their statistical analysis because one (#14) was a controlled diabetic, one (#30) had a history of kidney infection, and one (#32) was hypertensive. OSHA, however, considers that it may not be appropriate to eliminate these subjects when estimating the effect of cadmium exposure in the general population because of the following statistical/study design concerns and biological reasons: (1) it is not clear from Falck et al. (Ex. 4-28) that the same exclusion rules were applied to subjects exposed to cadmium who had negative spot samples; (2) it is not clear from the report that the exclusion rules were applied to the controls; (3) these conditions also may occur among persons with occupational exposure to cadmium and these persons need to be protected from the adverse kidney effects of cadmium as well. OSHA therefore applied the modified logistic model to the complete data set of Falck et al. (N=33) as well as to the reduced data set obtained by omitting subjects #14, #30, and #32. OSHA also made a third analysis of the complete data set in which subject #4 (who had an elevated spot urine sample that was not confirmed by the 24-hour sample) was considered to be affected. OSHA notes that there are large differences between the B(2)-microglobulin levels measured in spot samples and in 24-hour samples; B(2)-microglobulin levels in the 24-hour samples are up to 200 times smaller than the levels in the spot samples. These differences are not explained by Falck et al. and no mention is made of controlling or testing for pH in the 24-hour samples. Thus, OSHA is of the opinion that the data from this study are less reliable for quantitative analysis as compared to the data from the remaining studies.

Mason et al. used elevated levels of RBP to define cadmium workers who were considered to have cadmium proteinuria. RBP was used instead of B(2)-microglobulin because RBP appears to be less sensitive to urine pH than B(2)-microglobulin, and because pH apparently was not adjusted or controlled in the Mason et al. study. However, since OSHA intends to base its regulations on urinary B(2)-microglobulin levels, it was necessary for OSHA to relate the RBP levels in the Mason et al. study to corresponding levels of B(2)-microglobulin, after adjusting for pH.

A regression of the logarithm of RBP level on the logarithm of urinary B(2)-microglobulin concentration, restricted to the 114 samples for which the urine pH was 5.5 or greater, revealed a highly significant relationship (p < 0.0001) that explained a large proportion of the variation in the data (R(2) = 0.84). When the same analysis was performed on the 60 samples for which the pH was less than 5.5, although a highly significant relationship was also obtained (p < 0.0001), the R(2) was considerably smaller (R(2) = 0.27). OSHA concluded from this analysis that these data indicate that there is a relationship between RBP levels and B(2)-microglobulin levels in urine samples in which the pH is 5.5 or greater that can be used to relate the level of 300 ug/g creatinine, or, equivalently, 33.8 ug/mmole creatinine, of B(2)-microglobulin (used by OSHA to define proteinuria) to a corresponding level of RBP. The regression equation obtained, based on samples with pH of 5.5 or higher, was


Ln(R) = 0.032 + 0.88 * Ln(B2),

where R and B2 represent levels of RBP and B(2)-microglobulin, respectively, in urine measured in units of ug/g creatinine. This equation indicates that an RBP level of 156 ug/g creatinine corresponds to a B(2)-microglobulin level of 300 ug/g creatinine. Based on this cutoff, 15% of Mason et al.'s matched referents (11/72) would be defined as having proteinuria. On the other hand, Mason et al. indicate that they used the upper 95th percentile for urinary RBP calculated from their matched referent population to define cadmium workers who were considered to have proteinuria. Although Mason et al. do not specify this 95th percentile, based on the raw data from this study, OSHA has calculated this percentile as 338 ug of RBP per gram creatinine. OSHA has conducted analyses of the Mason et al. data using both of these cutoff levels, that is, considering subjects with urinary levels of RBP in excess of either 156 ug/g creatinine (Mason 1) or 338 ug/g creatinine (Mason 2) as having proteinuria.

Since there are questions regarding whether supralinear dose responses are reasonable for biological responses (see, e.g., Crump, 1985), whenever a t < 1 was estimated, the model was refit with t fixed and equal to one. This occurred only with the data from the Jarup et al. study.

OSHA made a total of nine fits of the logistic model, modified to incorporate background response, to quantal data from five studies: the Elinder data; the Ellis data; three versions of the Falck data that involved minor changes in cohort definition and designation of subjects with proteinuria; two fits to the Jarup data, one with t < 1 (Jarup 1), and one with t = 1 (Jarup 2); and two versions of the Mason data using different cutoffs for defining subjects with proteinuria. The model was fit to data from the Elinder et al. and Jarup et al. studies that had been grouped by the authors into five and six exposure groups, respectively (Table VI-16). In the remaining studies the model was fit to the ungrouped data.


TABLE VI-17. - SUMMARY OF FIT OF MODIFIED LOGISTIC MODEL TO DATA
                ON KIDNEY DYSFUNCTION
_________________________________________________________________
                          |        Parameter estimate
                          |______________________________________
      Data set            |  Delta     |    Alpha     |  Beta
__________________________|____________|______________|__________
                          |            |              |
Elinder...................|   0.215    |    -23.74    |   2.92
Ellis.....................|   0.040    |    -11.1     |   1.60
Falck:                    |            |              |
  1(a)....................|   0.0663   |    -37.6     |   5.19
  2(b)....................|   0.139    |    -55.8     |   7.64
  3(c)....................|   0.0748   |    -56.4     |   7.93
Jarup:                    |            |              |
  1(d)....................|   0.0      |     -8.0     |   0.81
  2(e)....................|   0.0054   |     -9.6     |   1.0
Mason  1(f)...............|   0.17     |    -26.1     |   3.7
Mason  2(g)...............|   0.062    |    -18.6     |   2.51
__________________________|____________|______________|___________
  Footnote(a) Includes all 33 subjects.
  Footnote(b) Considers subject number 4 to have kidney dysfunction.
  Footnote(c) Omits subjects 14, 30, 32.
  Footnote(d) t unrestricted.
  Footnote(e) Restriction imposed of t = 1.
  Footnote(f) Kidney dysfunction defined as RBP > or = 18 umoles/mg
creatinine.
  Footnote(g) Kidney disfunction defined as RBP > or = 38 umoles/mg
creatinine.

Table VI-17 records the estimates of the parameters delta, alpha, and t of the logistic model for each of the nine fits to the five data sets, and Table VI-18 records the results of chi-square goodness-of-fit tests applied to each of the fits except those based on the Falck et al. data. (There were not enough affected subjects in this study to provide a valid goodness-of-fit test.) Although the logistic model was fit to the ungrouped data from Mason et al. and Ellis et al., the data were subsequently grouped into exposure groups as shown in Table VI-18 to obtain large enough expected numbers of affected subjects in each group to yield a valid goodness-of-fit test. As indicated by Table VI-18, the modified logistic model provided a good fit to each of the data sets. This indicates that there is no statistical evidence in these studies that the modified logistic regression model is not an appropriate model for modelling kidney dysfunction as a result of cadmium exposure.


TABLE VI-18. - FIT OF LOGISTIC MODEL TO DATA ON KIDNEY PROTEINUREA(a)
___________________________________________________________________________
                       |           | Number of Cases   |          |Lack of
Cadmium Exposure Range | Number of |___________________|Chi-square| fit
(ug-year/m(3))         | subjects  |Observed |Expected |  (df)(*) |p-value
_______________________|___________|_________|_________|__________|________
                       |           |         |         |          |
Elinder (Ex. L-140(45):|           |         |         |          |
    0-1000.............|       16  |      3  |     3.5 |          |
    1000-2000..........|       22  |      7  |     6.2 |          |
    2000-3000..........|        9  |      4  |     4.0 |          |
    3000-5000..........|        8  |      5  |     5.6 |          |
    5000+..............|        5  |      5  |     4.7 |          |
                       |           |         |         | 0.83 (2) |0.66(NS)
Ellis (Ex. 4-27):      |           |         |         |          |
    5-800..............|       39  |      8  |     6.7 |          |
    820-1567...........|       11  |      4  |     6.3 |          |
    1600-2500..........|       10  |      8  |     7.5 |          |
    3000-4900..........|        8  |      7  |     7.2 |          |
    5000-6100..........|        7  |      7  |     6.6 |          |
    6500-20,830........|        7  |      7  |     6.8 |          |
                       |           |         |         | 1.2 (3)  |0.76(NS)
Jarup 1(b) (Ex. 8-661):|           |         |         |          |
    0-359..............|      264  |      3  |     4.7 |          |
    359-1710...........|       76  |      7  |     5.0 |          |
    1710-4578..........|       43  |     10  |     8.9 |          |
    4578-9458..........|       31  |     10  |     9.4 |          |
    9458-15,000........|       16  |      5  |     6.7 |          |
    >15,000............|       10  |      5  |     5.3 |          |
                       |           |         |         | 2.5 (3)  |0.48(NS)
Jarup 2(c) (Ex. 8-661):|           |         |         |          |
    0-359..............|      264  |      3  |     3.9 |          |
    359-1710...........|       76  |      7  |     4.0 |          |
    1710-4578..........|       43  |     10  |     8.7 |          |
    4578-9458..........|       31  |     10  |    10.0 |          |
    9458-15,000........|       16  |      5  |     7.5 |          |
    >15,000............|       10  |      5  |     6.1 |          |
                       |           |         |         | 4.9 (4)  |0.30(NS)
Mason 1(d)(Ex. 8-669A):|           |         |         |          |
    0..................|       96  |     15  |    16.8 |          |
    30-752.............|       37  |      9  |     7.1 |          |
    810-1424...........|       15  |      7  |     7.7 |          |
    1501-3219..........|        8  |      8  |     7.0 |          |
    3752-5263..........|        8  |      8  |     7.9 |          |
    6849-13,277........|        7  |      7  |     7.0 |          |
                       |           |         |         | 0.86 (2) |0.65(NS)
Mason 2(e)(Ex. 8-669A):|           |         |         |          |
    0-192..............|      109  |      7  |     6.8 |          |
    209-1414...........|       38  |      7  |     7.0 |          |
    1424-3752..........|       10  |      6  |     6.8 |          |
    3793-5263..........|        7  |      7  |     6.5 |          |
    6849-13,277........|        7  |      7  |     6.9 |          |
                       |           |         |         | 0.15 (2) |0.93(NS)
_______________________|___________|_________|_________|__________|________
  Footnote(*) df = degrees of freedom.
  Footnote(a) Fitting was based on ungrouped data for Mason 1, Mason 2,
and Ellis.
  Footnote(b) B unrestricted.
  Footnote(c) Restriction imposed of B = 1.
  Footnote(d) Kidney dysfunction defined as RBP > 18 umoles/mg creatinine.
  Footnote(e) Kidney dysfunction defined as RBP > 38 umoles/mg creatinine.

Table VI-19 provides estimates of the numbers of cases of proteinuria per 1000 workers exposed for a 45-year working lifetime at various 8-hour TWA exposures. Estimates in this table are based on extra risk {P(X)-P(0)/[1-P(0)]}, and confidence intervals were computed by the likelihood method. The results in Table VI-19 from the Ellis data and the Falck data differ from those contained in the proposed rule. The principle reason for this is that, unlike the model used to obtain the results in Table VI-18, the version of the logistic model used in the proposed rule did not allow for background response. OSHA considers the modified logistic model that incorporates background response to be more appropriate.


TABLE VI-19. - ESTIMATE OF KIDNEY PROTEINUREA PER 1000 WORKERS WITH 45
               YEARS OF OCCUPATIONAL EXPOSURE TO CADMIUM DERIVED FROM
               MODIFIED LOGISTIC MODEL

(For Table VI-19, Click Here)

Some of the confidence intervals in Table VI-19 are fairly wide, which reflects both the small size of some of the studies and the uncertainty in extrapolating results to low TWA exposures. Considering this and the differences among the underlying studies (see Table VI-15), these analyses yield reasonably consistent results. All of the analyses predict a high incidence (24%-99.8%) of proteinuria at exposures of 100 ug/m(3). Six of the seven analyses (all except the Mason 1 analysis), provide consistent estimates of the extra risk of proteinuria at TWA cadmium exposures of 1 ug/m(3) and 5 ug/m(3), in the sense that the 90% confidence intervals from all of the analyses contain a common range. That is, with the exception of the Mason 1 analysis, all of the 90% confidence intervals for the extra risk of proteinuria at a TWA exposure of 5 ug/m(3) contain the range between 14 cases per 1000 workers and 23 cases per 1000 workers(6). Thus, within the limits of statistical variability, these six analyses are all consistent with an extra risk of proteinuria in this range, but are not all consistent with a risk of proteinuria outside this range. (The upper 95% confidence limit from the Mason 1 analysis is 12 cases per 1000 workers, which is just barely outside the range defined by the remaining six analyses.)


__________
  Footnote(6) The 95 percent upper and lower conficence bounds presented
in Table 19, when considered in combination define 90 percent confidence
intervals.

Some of the differences in the risk estimates from different studies is due to different definitions of kidney dysfunction used in the various studies. For example, considering the risk of kidney dysfunction from a TWA exposure of 5 ug/m(3) cadmium, the lowest risk was obtained from the Falck et al. study, which employed the most stringent definition of kidney dysfunction (B2-microglobulin > 629 ug/g creatinine [Table VI-15]), whereas the highest risk was obtained from the Ellis et al. study, which employed a much more liberal definition of kidney dysfunction (B2-microglobulin > 200 ug/g creatinine [Table VI-15]). Other reasons for the differences in the risk estimates include different methods for controlling pH and differences in the quality and quantity of extent of exposure data and the methods used for estimating exposures from these data (Table VI-15). Considering these differences, and also the statistical uncertainty in the five studies due to small sample sizes, OSHA believes that the analyses summarized in Table VI-19 present a consistent picture of risks of kidney dysfunction from cadmium exposure. The best estimates from these analyses of the risk of kidney dysfunction from 45 years of occupational exposure to a TWA concentration of 5 ug/m(3) cadmium are in the range of 14 to 23 cases per 1000 workers.

Risk Assessment Based Upon Continuous Data from the Mason et al. Study

Since it had access to the raw data from the Mason et al. study (Ex. 8-669A), in addition to modelling, as discussed earlier, OSHA was able to model the continuous data from the Mason study on the actual level of RBP in the urine as a function of cumulative cadmium exposure. This approach does not require collapsing the urinary data into a yes-no response, and consequently may make more efficient use of the data.

OSHA applied two types of models to the continuous data from the Mason study: One that made use of the matching by Mason et al. of referents to exposed subjects, and one that did not make use of this matching. The matched analysis was applied to 72 matched pairs of exposed subjects and referents. (Although Mason et al. [Ex. 8-669A] reported 75 pairs in their analysis, in the data provided to OSHA, RBP values were missing for three of the matched referents.) The unmatched analysis was applied to all of the exposed subjects and referents recorded in the data furnished to OSHA by Mr. Mason for which there was both a cadmium exposure and a urinary value of RBP reported (the cadmium exposure was assumed to be zero for referents); this analysis included 75 exposed subjects and 96 referents.

In the model fitting that made use of the matching, the quantity Log(R(e)/R(r)), where R(e) is the urinary RBP of an exposed subject and R(r) is the urinary RBP of the matched referent, was assumed to have a normal distribution with mean alpha(*) (X - X(0))(t) and standard deviation, sigma (independent of X), where


X    is the cumulative cadmium exposure of the exposed subject in
     ug-yr/m(3),
X(0) is a posited potential threshold exposure to cadmium below which
     cadmium cannot adversely affect the kidney,
and alpha and t are parameters estimated from the data.  [Throughout this
     discussion, the expression (X - X(0))(t) is taken to be zero if
     X > = X(0).]

In the model fitting that did not consider the matching, Ln(R(e)) is considered to have a normal distribution with mean delta + alpha(*) (X - X(0))(t) and standard deviation sigma, where delta is a parameter representing the mean amount of urinary RBP in persons not exposed to cadmium. The remaining parameters have the same meaning as the model applied to the matched data.

Table VI-20 shows the results of fitting these models and various simpler submodels to the continuous data from the Mason et al. study. Considering the analyses applied to the unmatched data, in Case I the mean of the logarithm of urinary RBP {E[Ln(R(e))]} is assumed to vary linearly with cumulative cadmium exposure; in Case II this mean is allowed to vary in a non-linear fashion with cumulative cadmium exposure (sub-linear if t > 1, and supralinear if t < 1). In Case III a threshold of X(0) ug-years/m(3) is assumed for the effect of cadmium; urinary RBP is assumed to be unaffected by cadmium exposure as long as the cumulative cadmium exposure is below X(0) ug-years/m(3) (the value of the threshold X(0) is estimated from the data). The mean of the logarithm of urinary RBP is assumed to increase linearly with increasing cumulative cadmium exposure for exposures higher than the threshold cadmium exposure of X(0). Case IV is a modification of Case III in which the mean of the logarithm of urinary RBP is allowed to increase non-linearly for exposures above the threshold cadmium concentration. Cases I-IV have similar meanings in the analyses applied to the matched data.


TABLE VI-20. - RESULTS OF FITTING MODELS TO CONTINUOUS DATA ON RETINAL
                BINDING PROTEIN (RBP)
_________________________________________________________________________
                                                       |  Log-likelihood
_______________________________________________________|_________________
                     Analyses Based on Unmatched Data
_________________________________________________________________________
I.    E[Ln(R(e))] = delta + alpha X                    |
      delta = 2.14; alpha = 0.000687; sigma(2) = 1.33..|     -24.18
II.   E[Ln(R(e))] = delta + alpha X(t)                 |
      delta = 2.04; alpha = 0.00406; t = 0.803;        |
        sigma(2) = 1.29................................|     -21.56
III.  E[Ln(R(e))] = delta + alpha (X-X(0))             |
      delta = 2.14; alpha = 0.000687; X(0) = 0,        |
        sigma(2) = 1.33................................|     -24.18
IV.   E[Ln(R(e))] = delta + aplha [(X-X(0))](t)        |
      delta = 2.10; alpha = 0.0149; t = 0.663;         |
        X(0) = 449; sigma(2) = 1.24....................|     -18.58
_______________________________________________________|_________________

                     Analyses Based on Matched Data
_________________________________________________________________________
                                                       |
I.    E[Ln(R(e)/R(r))] = alpha X                       |
      alpha = 0.000690; sigma(2) = 3.22................|     -42.10
II.   E[Ln(R(e)/R(r))] = alphaX(t)                     |
      alpha = 0.00515; t = 0.774; sigma(2) = 3.05......|     -40.22
III.  E[Ln(R(e)/R(r))] = alpha (X-X(0))                |
      alpha = 0.000690; X(0) = 0.0;  sigma(2) = 3.22...|     -42.10
IV.   E[Ln(R(e)/R(r)]) = alpha [(X-X(0))](t)           |
      alpha = 0.147; t = 0.408; X(0) = 1040;           |
        sigma(2) = 2.75................................|     -36.37
_______________________________________________________|_________________
  R(e) - Urinary RBP in subject exposed to X ug-year/m(3) cadmium.
  R(r) - Urinary RBP in matched.
  NOTE:  In these expression, X-X(0) is taken to be zero if X <  X(0).

With no threshold in the model (Cases I and II), a non-linear model (Case II) provides a significantly better fit than a linear model (Case I) [based upon a likelihood ratio test (Cox and Hinkley, 1974)]. The better-fitting non-linear model is supralinear (t = 0.803 < 1 for the unmatched analysis, and t = 0.774 < 1 for the matched analysis). Based on linear models only (Cases I and III), the data do not provide any evidence of a threshold, because even if a threshold is permitted in the model (Case III), it is estimated as zero (i.e., no threshold). However, a non-linear model with a threshold (Case IV) provides a significantly better fit than either a non-linear model with no threshold (Case II) or a linear model with a threshold (Case III). Again, the better-fitting model is supralinear for doses above the threshold (t = 0.663 < 1 for the unmatched analysis and t = 0.408 < 1 for the matched analysis). These conclusions hold both for analyses based on matched data and those based on unmatched data.

The extra risk of proteinuria in a person exposed to X ug-yr/m(3) is defined as [P(X)-P(0)]/[1-P(0)], where P(X) is the probability of proteinuria in a person with a cumulative cadmium exposure of X ug-years/m(3). Based on the model assumptions given above, it can be shown that for the unmatched analyses, P(X) is given by


P(X) = 1 - N{[Ln(338) - delta - alpha (*) (X - X(0))(t)]/sigma}

where N is the standard normal distribution function, and 338 ug/g creatinine is the cutoff value for RBP in urine used to define proteinuria in the Mason et al. study.

The model used for the matched analysis data provides an estimate of the ratio of the RBP level in an exposed individual relative to what his baseline RBP level would be if he had not been exposed to cadmium. (This baseline value was estimated by the RBP level in a matched control in the matched analysis.) In the calculations presented, OSHA used 111 ug/g creatinine for this baseline, which was the average value from the 96 referents in the Mason et al. study. Consequently, for the matched analyses,


P(X) = 1 - N{[Ln(3.045) - alpha (*) (X - X(0))(t)]/sigma},

where 3.045 = 338/111. Thus, the matched analysis is estimating the probability that the RBP level is about tripled (3.045 = 3).


TABLE VI-21. - ESTIMATE OF KIDNEY PROTEINUREA PER 1000 WORKERS 45 YEARS
  OF OCCUPATIONAL EXPOSURE TO CADMIUM DERIVED FROM CONTINUOUS MASON DATA
_________________________________________________________________________
  8-Hour TWA         |   Models I, III  |   Model II  |    Model IV
exposure (ug/m(3))   |                  |             |
_____________________|__________________|_____________|__________________
                          Based on Unmatched Analyses
_________________________________________________________________________
                     |                  |             |
1....................|           5.2    |         13  |          0
5....................|            28    |         54  |          0
10...................|            60    |        107  |        2.3
20...................|           139    |        224  |        204
50...................|           465    |        606  |        681
100..................|           908    |        947  |        970
200..................|          1000    |       1000  |       1000
_____________________|__________________|_____________|__________________
                         Based on Matched Analyses
_________________________________________________________________________
                     |                  |             |
1....................|           7.8    |         25  |           0
5....................|            40    |         91  |           0
10...................|            82    |        161  |           0
20...................|           170    |        287  |           0
50...................|           450    |        593  |         469
100..................|           818    |        879  |         889
200..................|           997    |        996  |         992
_____________________|__________________|_____________|__________________
  See Table 20 for definition of models.

Table VI-21 provides estimates of the numbers of cases of proteinuria per 1000 workers exposed for a 45-year working lifetime at various 8-hour TWA exposures derived from analyses of continuous data from the Mason et al. study. Models I, II, and III predict between 28 and 91 cases of proteinuria per 1000 workers exposed to a TWA of 5 ug/m(3). These estimates are considerably higher than those in Table VI-19 derived from the Mason data, and are also generally higher than those in Table VI-19 derived from other data sets. Turning to the estimates in Table VI-21 derived from Model IV (non-linear model incorporating a threshold), based on matched analyses the model predicts no risk for TWA exposures of 20 ug/m(3) or below. (This model predicted a threshold of 1040 ug-years/m(3) [Table VI-20], which is equivalent to a TWA exposure for 45 years of 1040/45 = 23 ug/m(3).) However, the model predicts a very high risk of proteinuria for exposures slightly higher than the predicted threshold (e.g., it predicts that 469 out of 1000 workers exposed for 45 years to a TWA cadmium exposure of 50 ug/m(3) will develop proteinuria.) Based on unmatched analyses, Model IV predicts no risk for TWA exposures of 5 ug/m(3) or below. (This model predicted a threshold of 449 ug-years/m(3) [Table VI-20], which is equivalent to a TWA exposure for 45 years of 449/45 = 9.98 ug/m(3).) However, this model likewise predicts a sharply increasing risk of proteinuria for exposures higher than the predicted threshold (e.g., it predicts that 204 out of 1000 workers exposed for 45 years to a TWA cadmium exposure of 20 ug/m(3) will develop proteinuria.) To further explore the relationship between cadmium exposure and urinary RBP in Mason et al. study, OSHA developed three plots showing the Mason et al. data and the fit of the various models to these data. Figure VI-1 is a log-log plot of the ratios of RBP level in cadmium-exposed subjects to the RBP level in matched referents versus cumulative exposure to cadmium. (For Figures VI-1 - VI-4, see the end of this section VI-Quantitative Risk Assessment.) This figure shows there is a considerable amount of variability in the data points about the mean curve, no matter which model is used to describe the mean. Figure VI-2 is a similar plot for the unmatched analyses, and this plot likewise shows a considerable amount of variability in the data points about the mean curves. For visual reference, the referent RBP values are plotted along the left vertical axis in Figure VI-2 (although since a log scale is being used for dose, theoretically they should be plotted infinitely far to the left at an x-axis value of minus infinity). Visually, there appears to be little difference between the fits of the different models in Figure VI-2. Figure VI-3 is exactly the same plot as Figure VI-2, except than in Figure VI-3 a linear scale has been used for cumulative cadmium exposure (the x-axis) rather than a log scale. Use of a linear scale also allows both referents and cadmium-exposed subjects to be legitimately included in the same graph. The dose response suggested by Figure VI-3 appears less "threshold-like" than that suggested by Figure VI-2 and suggests that any indication of a threshold by Figure VI-2 may be an artifact stemming from the use of a log-scale for cadmium exposure.

The models applied to the continuous data only estimated a threshold in conjunction with a supralinear dose response (t < 1). If the models were not allowed to be supralinear (i.e., constrained to be linear or sublinear, t + or - 1), the model predicted no threshold. This superlinearity causes the models to predict a very rapid rise in risk at cadmium exposures slightly above the estimated threshold (Figures VI-1 - VI-3). Consequently, if this supralinearity is in fact real, if the threshold is slightly overestimated, the risk at the estimated threshold could be significant. Moreover, there are questions regarding whether supralinear dose responses are biologically plausible (Crump, 1985).

In addition, there are plausible biological arguments that indicate a threshold may not occur for kidney effects of cadmium exposure. As pointed out by epidemiologist Richard Peto (1978), whenever a chemical exposure augments a health effect that occurs to some degree even among non-exposed persons by the same general mechanisms through which the background health effects occur, there is unlikely to be a threshold for that effect.(7) OSHA notes that proteinuria occurs among persons not exposed to cadmium and, further, there are sources of cadmium exposure other than the workplace. These facts, when applied to Peto's reasoning, argue that there may not be a threshold for kidney effects from cadmium exposure.


__________
  Footnote(7) Although Peto's argument was made specifically for cancer,
it applied more generally to any health effect that is augmented by
exposure to a chemical through the same general mechanism.

OSHA concludes that in order to insure adequate protection for exposed workers, it should give greater weight to the non-threshold models than to the models that predict a threshold. The reasons for this are threefold. First, there are plausible biological arguments that a threshold may not exist for cadmium effects upon the kidney. Second, a threshold is only estimated in conjunction with a supralinear dose response which is of questionable biological plausibility. Third, even if the estimated threshold is real, slight errors in estimation of the threshold could result in significant risk at the estimated threshold.

Thun et al. Study of Kidney Effects in Cadmium Workers

In their study of workers exposed to cadmium at the same cadmium recovery plant that was studied by Ellis et al. (Ex. 4-27), Thun et al. (Ex. 19-43B) found that increasing cadmium dose was associated with reduced absorption of B2-microglobulin and retinol binding protein. Cadmium dose remained the most important predictor of serum creatinine levels after controlling for age, blood pressure, body size, and other factors. Although Thun et al. did not report the data in their paper in a form that could used for quantitative dose response modelling, Thun et al. present a graph that represents the result of their own modelling (Ex. 19-43B, Figure VI-3). Based on values read from this graph, it appears that the probability of renal abnormality in unexposed subjects is about 0.12 and the corresponding probability among subjects with a cumulative cadmium exposure of 500 mg-days/m(3) (equivalent to a 45-year TWA occupational exposure to 30 ug/m(3)) is about 0.2, which means that the extra risk of renal abnormality from cadmium exposure is about (0.2 - 0.12)/(1 - 0.12) = 0.091, or 91 per 1000 workers. Similarly, the probability of renal abnormality among subjects with a cumulative cadmium exposure of 1000 mg-days/m(3) (equivalent to a 45-year TWA occupational exposure to 60 ug/m(3)) is about 0.34, which translates to an extra risk of about 0.25, or 250 per 1000 workers. These values appear to be within the range of the quantitative estimates obtained by OSHA from other studies, and which appear in Tables VI-19 and VI-21. Thus, although OSHA was not able to conduct an independent analysis of the data from the Thun et al. study of kidney effects in cadmium-exposed smelter workers, it appears that the quantitative predictions from this study are consistent with the those developed by OSHA from other studies.

Thun et al. Review of Studies of Kidney Dysfunction in Cadmium Workers

Thun et al. (Ex. L-140-50) reviewed seven occupational studies that examined the relation of kidney dysfunction to cumulative exposure to airborne cadmium. The seven studies included the five used by OSHA for quantitative modelling [Falck et al. (Ex. 4-28), Ellis et al. (Ex. 4-27), Elinder et al. (Ex. L-140-45), Jarup et al. (Ex. 8-661), and Mason et al. (Ex. 8-669)], the study of Thun et al. (Ex. 19-43B), which was discussed above, and a study by Kjellstrom et al. (Ex. 8-233) of the same cohort of Swedish battery workers that was studied by Jarup et al. They considered both the dose-response relationship and the potential lowest cadmium exposure at which kidney effects are detectable. Figure VI-4 reproduces Figure 1 of Thun et al. (Ex. L-140-50), which presents the dose-response data for kidney dysfunction from these studies along with a risk assessment model developed by OSHA in the proposed rule and the prevalence estimated from a metabolic model by Kjellstrom (1986d, ref. in Ex. L-140-50). Thun et al. noted that, "the OSHA model generally follows the upper range of the empirical data and agrees well with the Kjellstrom and Nordberg metabolic model."

For comparison purposes, OSHA has superimposed on Thun et al.'s figure, a representation of a dose-response model estimated by fitting the modified logistic model to the data of Mason et al. (the Mason II analysis reported in Tables VI-16, VI-17, and VI-19). This particular dose-response model appeared to give results that were within the range of the results obtained by OSHA from the remaining models (see Table VI-19 and Figure VI-4). Table VI-19 indicates that this model appears to give results that are generally intermediate between those obtained using the OSHA model from the proposed rule and the metabolic model, and that, overall, the three models are in good agreement.

After reviewing the available data on the dose response of renal dysfunction, Thun et al. (Ex. L-140-50) concluded that it was impossible to identify a no-effect level for the renal effects of cadmium with certainty. Reasons for this included limited sample sizes of the studies, methodological differences between the studies, and imprecision of the exposure data. Thun et al. concluded that "The overall data suggest that the PEL for cadmium should not exceed 5 ug/m(3) to protect workers from kidney dysfunction and lung cancer over a working lifetime." This conclusion supports the independent analyses conducted by OSHA. Indeed, OSHA's analyses indicate that the risk of kidney dysfunction from 45 years of occupational exposure to a PEL of 5 ug/m(3) may be well in excess of one case per 1000 workers.

Discussion of Issues Related to Risk Assessment for Kidney Dysfunction

Two-phase Model Used by Mason et al.

Mason et al. (Ex. 8-669A) used a two-phase linear regression to model the relationship between the logarithm of the ratio of RBP levels in exposed subjects and unexposed matched referents [Ln(R(e)/R(r)), where R(e) is the RBP level in an exposed subject, R(r) is the RBP level in the matched unexposed referent] and the logarithm, Ln(X), of the cumulative cadmium exposure of the exposed subject. This model indicated that a change of slope in the relationship between RBP and cumulative cadmium exposure occurred at a cumulative exposure of 1108 ug-yr/m(3). A larger slope was indicated for cumulative exposures larger than 1108 ug-yr/m(3), and a smaller, but still positive, slope was indicated for cumulative exposure values smaller than 1108 ug-yr/m(3) (Table 6 and Figure 3 in Mason et al.).

In its proposed rule (Ex. 18, p.4083), OSHA interpreted the Mason et al. analysis as indicating that "the excess risk threshold for cadmium was approximately 1000 ug/m(3)-years." However, upon reevaluation, OSHA notes that, since the slope was still positive even at doses below 1108 ug-yr/m(3), Mason et al.'s model predicts an increasing response at all levels of cumulative cadmium exposure. Thus, it does not predict a threshold for the effect of cadmium at 1,100 ug-yr/m(3).

OSHA also notes that, whether a two-phase model is suggested by the data may depend heavily upon whether cadmium exposures are log-transformed before plotting. Figure VI-2 is a plot of the Mason data using log-transformed exposures and Figure VI-3 is a plot of the same data using untransformed exposures. Figure VI-2 appears perhaps to be suggestive of a two-phase relationship whereas Figure VI-3 does not. OSHA notes that, more generally, a wide variety of dose responses (including linear, no-threshold dose responses) can be made to appear two-phase or threshold-like by using log-transformed exposures when plotting. (This is because the log-transform has the effect of placing the exposure origin at minus infinity on the log scale and thereby exerts a "stretching" effect upon the dose response at low doses. This stretching effect can be see by comparing Figures VI-2 and VI-3.) OSHA also notes that the two-phase linear model as applied by Mason et al. is not meaningful at low exposures because it predicts that the RBP levels in cadmium-exposed subjects will be smaller than the RBP levels in unexposed subjects by arbitrary large factors at small exposures (e.g., the model predicts that the ratio R(e)/R(r) approaches zero at low exposures, whereas in actuality this ratio must approach one.) Consequently, OSHA does not consider that this model is reasonable for predicting the kidney response from low cadmium exposures. On the other hand, the models used by OSHA in fitting Mason et al. continuous data assume correctly that the ratio R(e)/R(r) approaches one at low exposures.

Other Recommended Models for Kidney Data

Several commenters (Tr. 6/8/90) (Ex. 19-43G) recommended modelling the dose response for kidney dysfunction using the probit distribution. These recommendations were based generally on the evidence that urinary cadmium or protein levels appear to follow a log-normal distribution. For example, ENVIRON Corp. (19-43G) cited evidence that the distribution of kidney cadmium concentrations among exposed individuals is often log-normal and concluded that the probit is a "log-normal tolerance distribution model" and consequently should be used for low-dose extrapolation of cadmium renal effects. Dr. Starr (Ex. 38) also considered the probit model to be a better choice for low-dose extrapolation purposes than the logistic model used by OSHA. In support of this position, Dr. Starr offered several pieces of evidence including the following: (1) A finding by Elinder et al. (1976, ref. in Ex. 38), based on measurements taken at Swedish autopsies, that the frequency distribution of kidney cadmium concentrations was log-normal; (2) the fact that Kjellstrom (1986b, ref. in Ex. L-140-50) used a log-normal distribution for individual critical kidney cortex cadmium concentrations; and (3) a finding by Kjellstrom et al. (Ex. 8-233) that

"The individual variation in effect at the same exposure duration is great, but in each dose group a log-normal distribution of B(2)-microglobulin excretion was found."

OSHA notes that all of the evidence presented by Dr. Starr addresses the distribution of cadmium or low molecular weight proteins (e.g., B(2)-microglobulin) in urine within various populations. For example, the quote by Kjellstrom et al. (Ex. 8-233) refers to a log-normal distribution for concentrations within dose groups, and consequently among individuals having similar cadmium exposures. None of this evidence addresses the issue of what is an appropriate dose-response model for relating cumulative airborne concentrations of cadmium to the probability of kidney dysfunction. Contrary to what Dr. Starr implies, a log-normal distribution for protein or cadmium in the kidney does not imply a probit model for the dose response. The model fit by OSHA to the continuous cadmium data of Mason et al. (Tables VI-20 and VI-21) assumes a log-normal distribution for the distribution of RBP among persons with the same cadmium exposures and consequently is consistent with the evidence cited by Dr. Starr. However, the corresponding dose response function is not a probit model(8).


__________
  Footnote(8) The probit dose response is defined as
P(X) = c + (1 -c)(*)N[a = b(*)Ln(X)] where X is cumulative dose, N
indicates the normal distribution function, Ln indicates natural
logarithm, and a, b, and c are parmeters estimated from data.  On the
other hand, the dose response oabtrained by OSHA from its model applied to
the unmatched Mason et al. data 9and shich assumed a log-normal
distsribution of RBP in subjects with the same cumulative cadmium
exposure) was:


P(X) = 1 - N{ILn(338) - delta - alpha(*)(X - X(0))(t))/sigma},

where 338 ug/g creatinine is the RBP level used to define kidney disfunction, and alpha, delta, X(o), t, and sigma are parameters. These two dose responses, while superficially somewhat similar in appearance, are not the same mathematically and may provide substantially different numerical values.

A log-normal distribution for kidney concentration after a fixed cadmium exposure does not place any restsriction upon the form of the dose response, because it can be shown that any dose response model is consistent with a log-normal distribution for kidney concentrations9i.e., given, a priori, any dose response model for the likelihood of kidney disfunction, an expression can be found for the distribution of the kidney concentration as a function of cadmium exposure that predicts the a priori dose response model as well as a log-normal distribution for kidney cadmium concentrations at fixed cadmium exposures).

OSHA concludes that, even if cadmium and protein levels in urine follow log-normal distributions, this in no way implies that a probit model is appropriate to model the dose response of kidney dysfunction from cadmium exposure. Dr. Lemen from NIOSH (Tr. 8-194) supported this conclusion, commenting that it doesn't matter whether the underlying B(2)-microglobulin data are or are not normally distributed because they are treated dichotomously by the model.

The Office of Budget and Management (OMB) (Ex. 17-D) suggested that, since the logistic regression analyses of kidney dysfunction involved "thresholds for classification purposes" (meaning that a cutoff was derived based on urinary concentrations of B2-microglobulin or other low molecular weight proteins and persons with urine concentrations above the cutoff were assumed to have kidney dysfunction), this "implies that kidney dysfunction is in fact a threshold-related health effect." OMB then raises the question, "To what extent does this argue against the use of a non-threshold probability model to estimate risk?" In this comment OMB is using two unrelated notions of a threshold. On one hand, they are identifying the cutoff for the urinary concentration of low molecular weight protein used to define kidney dysfunction as a "threshold" for urinary protein. On the other hand, they are suggesting that this implies the existence of a threshold of exposure to cadmium (i.e., a cadmium threshold) below which the risk of kidney dysfunction would be not be increased. Thus, in one case, they are referring to a "protein threshold" and in the other case a "cadmium threshold." These two concepts are, in fact, essentially unrelated. Use of a cutoff for urinary protein levels to define kidney dysfunction is unreleated to the use of a threshold or non-threshold model for evaluating the relationship between cadmium exposure and proximal tubular dysfunction. Since there is B(2) microglobulin excreted in the general population and cadmium augments that mechanism, there is more support for the use of a non-threshold model as compared to a threshold model.

Summary of Quantitative Estimates of Kidney Dysfunction from Cadmium Exposure

OSHA has made nine sets of estimates of the risk of kidney dysfunction (Tables VI-19 and VI-21), based on data from five different studies. In addition, OSHA has reviewed estimates based on a sixth study, that of Thun et al. (Ex. 19-43B). As indicated by Table VI-15, these six studies differ in many ways. The cohorts studied come from Sweden, the United Kingdom, Michigan, and Colorado. They were exposed to cadmium in smelters and several different types of manufacturing facilities. Exposures were to different forms of cadmium, including fume, dust, welder fume, and CdO dust. The number of cadmium-exposed subjects in the studies ranged from 33 to 440, representing a difference of more than an order of magnitude. Three of the studies included non-exposed controls, and one of these matched controls to exposed subjects by age. The cadmium exposure data were of variable type and quality and different procedures were used to quantify individual exposures.

There were also considerable differences in the procedures used in collecting the urine samples. Some of the studies involve spot urine samples. Mason et al. collected 3-hour samples, Elinder et al. collected morning samples, and Falck et al. collected spot samples, but confirmed findings in subjects with kidney dysfunction using 24-hour samples. Some of the studies were based on historical urine samples and others were collected specifically in conjunction with the associated study. Elinder et al. insured that urine pH would be acceptably high by administering sodium bicarbonate prior to the sampling. Some of the remaining studies adjusted pH after collection of the sample, and others make no mention of any adjustment for pH.

There were also differences in the type of protein and the level of protein in urine used to define kidney dysfunction. Mason et al. defined kidney dysfunction in terms of RBP in urine. Ellis et al. defined kidney dysfunction using a combination of B(2)-microglobulin and total protein. The remaining four studies defined kidney dysfunction in terms of B(2)-microglobulin. However, these four studies use different amounts of B(2)-microglobulin in urine to define kidney dysfunction; these cutoffs range from 300 ug/g creatinine (Ex. L-140-45) to 629 ug/g creatinine (Ex. 4-28).

Despite these differences, each of these studies demonstrated a relationship between exposure to cadmium and kidney dysfunction. In addition to the differences among the underlying studies, two different modelling approaches were applied. The logistic model, modified to include the possibility of background response, was applied to data from six studies. A model for continuous data was applied to the continuous data from the Mason et al. study. Versions of this latter model were applied that incorporated the possibility of a threshold.

All of these differences could contribute to disparities in quantitative results obtained from dose-response modelling. Given these many differences, one would not expect necessarily to get close agreement among the studies in quantitative estimates. For example, since different studies used different definitions of kidney dysfunction, different studies are actually estimating somewhat different endpoints.

However, despite differences in study protocols and modelling approaches, quantitative results are reasonably consistent. All of the results in Tables VI-19 and VI-21 predict a high risk of kidney dysfunction at a 45-year TWA cadmium exposure of 100 ug/m(3) (at least 242 cases per 1000 workers and, except for estimates based on the Jarup et al. study, at least 700 cases per 1000 workers). It was noted earlier that, with the exception of the Mason 1 analysis, the results in Table VI-19 provide consistent estimates of the extra risk of proteinuria at TWA cadmium exposures of 5 ug/m(3), in the sense that, with the exception of the Mason 1 analysis, all of the 90% confidence intervals in Table VI-19 for the extra risk of proteinuria at a TWA exposure of 5 ug/m(3) contain the range between 14 cases per 1000 workers and 23 cases per 1000 workers. (The upper 95% confidence limit from the Mason 1 analysis is 12 cases per 1000 workers, which is just barely below this range.) However, estimates of risk from exposure to 5 ug/m(3) range as high as 95 per 1000 workers (based on data from the Ellis et al. study).

The models applied to the continuous data of Mason et al. (Table VI-21) that do not predict a threshold predict extra risks at a 45-year TWA exposure to 5 ug/m(3) of between 28 and 91 cases per 1000 workers. These are within the range of the point estimates in Table VI-19. The models that predicted a threshold only did so in conjunction with a supralinear dose response, which is of questionable biological plausibility. These threshold models predicted a no-effect threshold at a cumulative exposure equivalent to a 45-year exposure to a TWA of between 20 and 50 ug/m(3) (matched analysis), or between 5 and 10 ug/m(3) (unmatched analysis). However, both of these threshold models predict that the risk of kidney dysfunction rises rapidly for exposures slightly above the threshold. For example, the matched model predicts that the extra risk of kidney dysfunction rises from 0 per 1000 workers to 469 per 1000 workers as 45-year TWA exposures increase from 20 to 50 ug/m(3), and the unmatched model predicts that the risk rises from 2.3 per 1000 workers to 204 per 1000 workers as 45-year TWA exposures increase from 10 to 20 ug/m(3).

OSHA concludes that it should give greater weight to the non-threshold models than to the models that predict a threshold, because (1) there are plausible biological arguments that a threshold may not exist for cadmium effects upon the kidney; (2) a threshold is only estimated in conjunction with a supralinear dose response which is of questionable biological plausibility; and (3) even if the estimated threshold is genuine, slight errors in estimation of the threshold could result in significant risk at the estimated threshold. OSHA further notes that, even if the threshold estimates are taken at face value, the PEL would need to be below 10 ug/m(3) in order to prevent kidney dysfunction from a 45 year occupation exposure at the PEL.

OSHA's preferred estimates of the extra risk of kidney dysfunction is in the range of between 14 to 23 cases per 1000 worker exposed to a TWA of 5 ug/m(3) for a 45 year working lifetime. This range is consistent with the majority of the analyses conducted by OSHA, although there are individual estimates both above and below this range. OSHA notes that this risk range is considerably in excess of one per thousand.

Overall Summary of Risk Assessment for Lung Cancer and Kidney Dysfunction

OSHA has developed estimates of the risk of lung cancer from occupational exposure to cadmium using several different types of analyses based on data from animal studies of Takenaka et al. (Ex. 4-67), Oldiges et al. (Ex. 8-694D) and Glaser et al. (Ex. 8-694B) and using human data from the cohort of workers exposed to cadmium at a cadmium smelter (Thun et al., Ex. 4-68; Stayner et al., Ex. L-140-20). These animal and human data indicate an increased risk of lung cancer from occupational exposure to cadmium.

OSHA has also developed estimates of the risk of kidney dysfunction from occupational exposure to cadmium using data from five different epidemiological studies of the effect of occupational exposure to cadmium upon kidney dysfunction (Falck et al., Ex. 4-28; Ellis et al., Ex. 4-27; Elinder et al., Ex. L-140(45); Mason et al., Ex. 8-669A; Jarup et al., Ex. 8-661). All of these studies, as well as the study of Thun et al. (Ex. 19-43B), indicate an increase in proteinuria among workers exposed to cadmium.

OSHA's preferred estimate of the excess lung cancer risk from 45 years of occupational exposure to cadmium ranges from 58 to 157 excess deaths per 1000 workers from exposure to a TWA of 100 ug/m(3), and estimates of the risk of kidney dysfunction from this exposure range above 900 cases per 1000 workers. OSHA's preferred estimates of risk from exposure to a TWA of 5 ug/m(3) range between three and nine excess lung cancer deaths based on the epidemiologic data and 15 excess cancer deaths based on the animal data and between 14 and 23 excess cases of kidney dysfunction per 1000 workers. Thus, risks of both lung cancer and kidney dysfunction are predicted to be in excess of one case per 1000 workers from 45 years of exposure to a TWA of 5 ug/m(3).


        Figure 1 - Fit of Three Models to Mason Matched Data

(For Figure VI-1, Click Here) Figure 2 - Fit of Three Models to Mason Unmatched Data

(For Figure VI-2, Click Here) Figure 3 - Fit of Three Models to Mason Unmatched Data

(For Figure VI-3, Click Here) Figure 4 - Prevetence of Tubular Proteinura by exposure to Cadmium In seven cross-sectional studies compared to Prediction by the OSHA risk assessment from the proposed rule, by the Kjellstrom metabolic model and by the modified logistic model applied to the Mason et al. data.

(For Figure VI-4, Click Here) References BEIR IV. 1988. Health Risks of Radon and Other Internally Deposited Alpha-Emitters. Committee on the Biological Effects of Ionizing Radiations. National Academy Press, Washington, DC. BEIR V. 1990. Health Effects of Exposure to Low Levels of Ionizing Radiation. Committee on the Biological Effects of Ionizing Radiations. National Academy Press, Washington, DC. Cox DR, Hinkley DV. 1974. Theoretical Statistics. Chapman and Hall, London. Crump K. 1984. An improved procedure for low-dose carcinogenic risk assessment from animal data. J Environ Pathol Toxicol 5:339-348. Crump K. 1985. Mechanisms leading to dose-response models. In: Principles of Health Risk Assessment. Ricci P, ed. Prentice Hall, Englewood Cliffs, NJ. pp. 321-372. Crump K, Howe R. 1984. The multistage model with a time-dependent dose pattern: applications to carcinogenic risk assessment. Risk Anal 4:163-176. Crump K, Guess H, Deal K. 1977. Confidence intervals and tests of hypotheses inferred from animal carcinogenicity data. Biomet 33:437-451. Krewski D, Crump K, Farmer J, Gaylor D, Howe R, Portier C, Salsburg D, Sielken R, Van Ryzin J. 1983. A comparison of statistical methods for low-dose extrapolation utilizing time-to-tumor data. Fund Appl Toxicol 3:140-160. National Academy of Sciences (NAS). 1983. Risk Assessment in the Federal Government: Managing the Process. National Academy Press, Washington, DC. National Toxicology Program (NTP). 1984. Report of the NTP Ad Hoc Panel on Chemical Carcinogenesis Testing and Evaluation. U.S. Department of Health and Human Services. Peto, R. 1978. Carcinogenic Effects of Chronic Exposures to very low levels of toxic substances. Environmental Health Perspectives 22:155-159. U.S. Department of Commerce (USDOC). 1980. Census of Population. Detailed Population Characteristics. Colorado. PC80-1-D7. U.S. Environmental Protection Agency (USEPA). 1989. Risk Assessment Guidance for Superfund. Volume I. Human Health Evaluation Manual. OSWER Directive 9285.7-01a. Office of Emergency and Remedial Response, Washington, DC. September 29, 1989. U.S. Environmental Protection Agency (USEPA). 1991. Health Effects and Dose-Response Assessment for Hydrogen Chloride Following Short-Term Exposure. Final Report. Prepared for Office of Air Quality Planning and Standards, Research Triangle Park, NC. Prepared by Clement International Corporation.

[57 FR 42102, Sept. 14, 1992; 58 FR 21778, April 23, 1993]

Regulations (Preambles to Final Rules) - Table of Contents Regulations (Preambles to Final Rules) - Table of Contents