Document 6BKQDxreg88K7Kn1OoZQ5kknE

September 22, 1987 VALIDATION OF WORK HISTORIES OBTAINED FROM INTERVIEWS GREGORY G. BOND1 KENNETH M. BODNER1 WENDY SOBEL1 RITA J. SHELLENBERGER2 GEORGE H. FLORES2 *The Dow Chemical Company, Epidemiology, H&ES 1803 Building, Midland, MI 48674 2 Dow Chemical U.S.A., Texas Operations Building B-101, Freeport, TX 77541 Address reprint requests to Dr. G.G. Bond, The Dow Chemical Company, Epidemiology, H&ES, 1803 Building, Midland, MI 48674 The authors gratefully acknowledge support for this study from local and corporate management of The Dow Chemical Company. In particular, we would like to recognize Drs. R.R. Cook, W.A. Fishbeck and J.H. Saunders. We thank the following indivi duals for technical assistance: J.B. Cartmill, G.A. Chateauneuf, A.Dang, D. Diaz, A.A. Grzegorczyk and B.J. Lasich. Running head: "VALIDATION OF INTERVIEW-BASED WORK HISTORIES STOI 64 922 ST0I64923 ABSTRACT Bond, G. G. (The Dow Chemical Company, Midland, MI 48674), K. M. Bodner, W. Sobel, R. J. Shellenberger, G. H. Flores. Validation of work histories obtained from interviews. There have been few attempts to assess the validity of occupa tional histories obtained from interviews. A nested case-control study of lung cancer conducted in 1985 among a cohort of 19,608 male workers at a Texas chemical production facility provided an opportunity to do that. Telephone interviews were attempted with 308 lung cancer patients and 588 matched controls or their next of kin in order to gather information on tobacco use, diet, places of residence and occupations. Interview data from 734 respondents (143 subjects and 591 next cf kin) were then compared with the records of work area assignments maintained by the company and with exposure profiles developed by industrial hygienists. Respondents recalled 48.4 per cent of all documented work area assignments, but fewer than 3 per cent of the chemical agents judged as likely exposures. Recall of usual work area assignment was 70.8 per cent. Among the factors found to have most influenced recall were the number and duration of assignments, a subjective assessment of the quality of the interview, and the relationship of the interviewee to the subject. The elapsed time from job assignment to interview was found also to affect recall of a subject's longest assignment. Keywords: epidemiologic methods, occupational diseases, question naire, recall, retrospective studies, sensitivity Bond 1. Questionnaires are often the sole source of occupational exposure data for epidemiologic studies, particularly for many of the large institution- or population-based case-control studies conducted of cancer. The validity of underlying data depends on accurate respondent recall. Misclassification of subjects can lead to over- or under estimates of risk, depending on whether cases are impacted differently than controls (1,2). There have been few attempts to validate work histories obtained by interview. Although a number of investigators have tried to confirm occupational data by contacting reported employers (3-8), such studies can estimate only the degree of overreporting of jobs, and cannot evaluate biases due to underreporting. Baumgarten et al (9) recently compared employer names obtained from interviews with those recorded in a Canadian government data bank. While the authors found good agreement between the two sources (82 per cent), only a recent 13-year time period was examined. Furthermore, the authors noted the limitations of using employer name as a surrogate measure of occupa tional exposure data. A nested case-control study of lung cancer conducted among a large cohort of chemical workers (10,11) provided the opportunity to assess the validity of occupational histories obtained by interview. The present report examines the frequency of recall errors and their rela tion to specific characteristics of the study design and interview respondents. One advantage of the present study over previous investigations was the opportunity to assess recall accuracy of jobs held as long ago as 40 years. In addition, the dependent variables were specific work area assignments and chemical or physical exposure agents: measures particularly applicable to case-control studies designed to identify occupational hazards. i2 6 (]9l01S Bond 2. MATERIALS AND METHODS Lung cancer study A nested case-control study was recently conducted among a cohort of 19,608 male chemical workers with one or more years of service at a Texas production facility, followed from 1940 through 1980 (10,11). These workers had been identified as possibly being at increased risk of lung cancer (12), for which an occupational etiology was considered. Cases were defined as those subjects who met the criteria for inclusion in the cohort (11,12), who died before 1981, and whose death certificates listed primary lung cancer anywhere on the document. Two control grou 'S, one a decedent and one a "living" series, were chosen from among all other cohort members without lung cancer. Controls were individually matched to cases on race, sex, year of birth and year of hire. Employees' documented work history records maintained by the company served as the starting point for categorizing workplace exposures. The records supplied demographic data and identified details of each job and department assignment at the facility, including dates of job changes. Since these records were updated at the time of assignment changes, they present accurate accounts of job mobility suitable for validation purposes. The two measures used to characterize employee exposures, work area, and chemical/physical agent profiles (13-15), were developed independently. For the first, fifty homogeneous work area classifications were condensed from more than 400 departmental titles STOI 64 9 25 Bond 3. used by the personnel office by grouping departments with similar exposure conditions. For the second, chemical and physical agent profiles were developed for each subject by one of the authors, a certified industrial hygienist. Work history records for each individual were reviewed "blind" to his case or control status, and all agents he was likely to have encountered on any job were listed. Every attempt was made to include all agents that would have presented a reasonable potential for exposure, considering such factors as amount handled, handling procedures, and vapor pressure of the material. In the nested study, telephone interviews were attempted with 308 cases and 588 matched controls or their next of kin in order to gather data on potential confounders. Among other items, the questionnaire included a series of open-ended questions designed to elicit a chronological listing of all jobs, department assignments, and agent exposures pertinent to the subject's career at this facility. In order to confirm the identify of the subject and to solicit cooperation, the respondent was reminded that the subject had worked for the company, and the dates of first and last employment were supplied, prior to beginning the interview. Respondents were prompted to list chemical exposures; however, at no time did the interviewers suggest job titles, departments, or specific agents. (A copy of the questionnaire may be obtained from the authors upon request.) An attempt was made to inter view the person expected to be most knowledgeable about the subject during his adult lifetime. The subject himself was preferred, or, if deceased or otherwise incapacitated, a spouse, an offspring, a sibling or other relative, or a friend was contacted, in that order. Most of the interviews (89 per cent) were completed by one of three g z e 'is io is Bond 4. interviewers, with the remainder conducted by two supervisors. The interview sessions ranged in length from 5 minutes for partial interviews to 94 minutes, averaging about 28 minutes. There was essentially no difference in means between the case (27.3 minutes) and the control (decedents = 26.3; "living" = 31.3) interviews. Validation study Interviews were successfully completed with 743 respondents (82 per cent of the total) between March 13 and August 27 of 1984. Failure to locate the subject or a suitable proxy was the major reason for a less than complete response, since participation among those contacted for interview was excellent (96 per cent). In comparing the participants to non-participants, it was found that the former were more likely to have been longer term employees and more recently hired than the latter. Other factors examined (age at hire, year of death, educational level, job category) revealed only negligible differences. Work histories obtained by interview were coded according to the same scheme used for documented company work histories, and were input to the computer using a display management system, incorporating a number of quality assurance steps and edit checks (11). Respondents listed 248 separate department titles, 242 different jobs and 96 chemical or physical agents as exposures for the 734 subjects. When these are compared with the 417 department titles and 542 job titles listed in company records and the 171 different agents catalogued by the industrial hygienist for those same subjects, it was STOI 64927 Bond 5. obvious that the respondents were not able to provide the same level of detail. Recognizing that such detail may not be required to correctly classify exposures, department titles listed by the respondents were grouped into the 50 work areas described earlier. Whether a subject had been classified as having worked in each of the 50 areas was then checked against the company record, using that as a standard. An employee may have contributed observations to multiple work areas, but was counted only once for each area. Accuracy of respondent recall was measured by two dependent variables: recall of all work area assignments listed in the company record, and recall of chemical or physical agent exposures. A variation of the former, recall of "usual" work area assignments (defined as the one work area to which the subject had been assigned for the longest continual period of time) was also used. False positives, wherein the respondent reported assignments or agents not derived from the record, were exceedingly rare; thus specificity of the interview data can be regarded as nearly 100 per cent. The focus of the analysis was therefore the sensitivity rate, or the degree to which respondents correctly recalled assignments or agent exposures as they appeared in the documented company record. Study results were tabulated and analyzed by packaged procedures in the Statistical Analysis System (16). Since it was suspected that respondent recall might be influenced by a variety of characteristics, sensitivity rates were examined for various subsets of the sample of subjects. The relative strength with which these factors may have affected recall was further evaluated by logistic regression, using STOI 64928 Bond 6. stepwise model-building procedures and likelihood ratio statistical tests. The seven factors that were regarded as potential determinants of recall are outlined in table 1. These included characteristics of the work history (elapsed time from the end of an assignment to the interview, cumulative duration of the assignment, number of assignments), the subject (group membership), and the interview (respondent's relationship to the subject, quality of the interview, principal interviewer). Interview quality was a subjective appraisal made by the interviewer regarding the respondent's willingness and ability to furnish information about the subject. RESULTS Overall, respondents recalled 48.4 per cent of all work area assignments and 70.8 per cent of usual assignments, but only 2.6 per cent of the chemical or physical agents listed as exposures by the industrial hygienists. Sensitivity rates for recall of work area assignments are shown in table 1, for the sample as a whole, and for subsets of subjects classified by various potential recall determinants. It is seen that correct responses diminished with the number of jobs and with elapsed time, but increased with longer duration jobs. Better recall was obtained in higher quality interviews, i.e., those in which the respondent was mere cooperative and more familiar with the employee's career. 626*191018 ST0I6493Q Bond 7. In the case of usual work assignments, there appeared to be small differences in recall associated with the relationship of the respondent to the subject. These differences were not found in considering all work area assignments. Table 1 also presents recall sensitivity separately for the cases and the two control series. Fewer than three percentage points difference were seen among the groups for recall of either all or of usual work assignments, providing little evidence for any degree of bias based on disease state or vital status of the study subjects in this investigation. The relative importance of the factors determining recall and the dependencies among them were investigated by building prediction models. Moderately complex logistic models in the seven main effects listed in table 1 and in first-order interactions were considered. Model factors were defined somewhat differently than in the crude tabulations. Elapsed time to interview, duration of assignment, and number of assignments were treated as continuous variables. Based on statistical evaluation of preliminary modeling results, the number of levels of some factors were reduced. Thus, the respondent's relationship was found to be adequately described by the dichotomy, "self" versus "other". Interview quality was redefined as "high or adequate" versus "questionable or unsatisfactory", and, for physical/chemical agent exposures, the interviewer choice was regrouped as "interviewer A" versus "others". Bond 8. Some examples of model estimates for work area recall are given in table 2. The procedure identified as important (or unimportant) many of the same determinants noted in table 1. The number of assignments and interview quality were shown to be critical to recall accuracy of both usual work area assignment and all work assignments. No evidence of effects due to interview differences or to subjects' group status was seen. A few differences between the model and the crude tabulations were noted. The respondent's relationship to the subject had a stronger effect on recall in the all work areas model than in the usual work area model, the reverse of what is apparent from table 1. Although elapsed time to interview appeared to be a strong factor in table 1, its influence in the all work areas recall model was overshadowed by another factor with which it was highly correlated, duration of assignment. However, for usual assignments, presumably all of considerable and more uniform duration, the elapsed time effect was seen to emerge. A substantial degree of interaction can be seen in table 2, all owing to the influence of the number of assignments factor on the other independent variables. Only interview quality in the usual work areas model was unaffected. In both work area models, an increase in assignments was associated with poorer recall. It also acted to magnify the effects of elapsed time to interview, the respondent's relationship to the subject, and duration of assignment, but diminished the advantage of high quality interviews. STOI 6493 I Bond 9. It was noted that recall of some work areas was better than others. Table 3 shows that recall sensitivity varied by about 40 percentage points for the most frequently occurring work area assignments in the company records for the study sample, and as much as 60 points when considering only usual assignments. To determine how much of this variation was due to the factors already examined, work area was included as the last term in each of the above models. Even after adjusting for the factors in table 2, both models showed significant differences among the five most common work areas: magnesium, maintenance, pipe shop, administration, paint shop and "all others". Recall rates for physical/chemical agents were considerably lower than those for work assignments, and modeling procedures identified a different set of determinants. As was found for the work area analyses, duration of assignment and the relationship of the respondent affected recall; however, some bias due to inter-interviewer differences and to the subject's study group membership was evident. The interpretation of these two factors is complicated by a significant interaction found between them. It was also found that recall of some agents was considerably better than others, although none approached the same levels obtained for work areas. Among the eight most common agents derived from the study sample's company records, the highest recall rates were found for chlorine (1G.7 per cent) and asbestos (8.6 per cent), and the lowest for sulfur dioxide (0.7 per cent) and heat (0.5 per cent). Significant differences among the agents were also noted in the logistic model, even after adjusting for the other factors named above. Thus, as with work STOI 64932 Bond 10. areas, there were apparently some factors operating to affect recall that were not specifically studies. DISCUSSION The results from this validation study indicate that a fair amount of underreporting of work histories can be expected when data are obtained from interview, but that some job characteristics are associated with better recall than others. Not surprisingly, recall of usual assignments was better than recall of all assignments in general. A job is more likely to be considered a usual assignment as it's duration increases. It was found that, when only assignments of at least five years duration were considered, recall of all work areas improved from 45.4 to 65.7 per cent. This figure approached the overall 70.8 per cent recall of usual assignments, which averaged about 5.9 years of duration. Recall of usual work areas increased only modestly to 73.2 per cent when assignments of less than five years were ignored. Failure to recall assignments of short duration is probably not critical to many study designs, if the operation of a dose-response mechanism is expected. In fact, it may be easier to detect an occupational effect among workers exposed for longer periods of time, without the diluting influence of inconsequential exposures. Multiplicity of jobs remains a definite concern to occupational epidemiology studies based on interview data. It can be expected that many more exposures will be overlooked by the respondent as the number of assignments increases. The finding of this study agrees with results STOI 6U933 Bond 11. obtained by Baumgarten et aj_ (9) who found significantly better agreement between report and record among respondents reporting fewer than two jobs compared with those reporting two or more jobs. Elapsed time between the end of an assignment and the interview was seen to have a substantial effect on recall, especially as the number of assignments increased. This is an important consideration to cancer studies in particular, since the job assignments of greatest relevancy to disease incidence are often ones that were held sufficiently long ago for an adequate latency period to have been expressed. The fact that agreement on work areas was found to be almost identical for cases and controls argues against the presence of any differential misclassification in these data. Non-differential misclassification, such as that due to exposure underreporting and as well to some study design features (2), has been shown not to change the direction of the risk estimate, but rather to bias it toward the null value (1). It can be demonstrated that the bias is particularly severe when the exposure rate among the controls is high. Since investigators tend to select study populations with inherently high rates of exposure, low recall sensitivity should be perceived as a definite problem. Investigators often make an effort to match dead controls with dead cases in order to achieve similar interview circumstances, even though this requires that interview data be obtained from proxies. Gordis (17) has argued against the use of dead controls, contending that proxies are ti 6 'l9 IO iS Bond 12. often able to provide data of comparable quality to that supplied by the subjects themselves. The results obtained from validation of work area recall in this study tend to support that contention. Only for more complicated work histories was the respondent's relationship to the subject seen to be a factor. Overall, free recall of specific physical or chemical agent exposures was abysmal (less than 3 per cent). This occurred in spite of the fact that employees at this facility have been appraised repeatedly of the materials they work with, as part of their safety training. This program, which has been evolving since the early 196G`s, involves identification of the materials in the employee's work environment and discussions of chemical toxicity, safe handling practices, and emergency response procedures. It should be noted that some considerations may have mediated the effect of the training program on recall. The majority (81 per cent) of the subjects' exposures in this study predate the formal inception of that program. Moreover, most of the interviews (82 per cent) were conducted with a surrogate respondent instead of the subject. The results of this investigation would argue that, in at least some instances, free recall is not a viable tool upon which to classify chemical exposures. It would be preferable for investigators conducting interview-based occupational health studies to consult with industrial hygienists in developing a specific chemical agent exposure classification scheme through the use of job titles and work duties. Some important considerations gleaned from the modeling process are that the determinants of recall are numerous, that they act and interact in complex ways, and that they are to some degree dependent on study S 6 *19 I 01S ST0I6U936 Bond 13. design. The large differences in recall among common work areas and agents, even after adjusting for known factors, hints that many important factors have not, at least in this study, been evaluated. This is further indicated in noting that the per cent variation explained by the models was in all cases very low (all work areas: R = 2 13 percent; usual work areas R =8 per cent; physical/chemical agents: o R = 7 per cent), although it is unlikely that more complex model structures would have helped. It should be realized that for the response variables of all work area assignment recall and physical/chemical agent recall, almost all subjects contributed multiple responses to the analysis, whereas the statistical design ignored these potential intra-subject dependencies. (We note, however, that this situation is not dissimilar to the prevalent use of person-year as the sampling unit in other types of epidemiologic research). A completely acceptable framework for such an analysis was not apparent. On the other hand, the analysis of usual work area recall was not burdened by such dependencies, since by definition, there is only one usual assignment per subject. The similarities in outcome for usual assignments and all assignments lessens any concern for violations of independence assumptions in this study. Some unique features of this study may limit the inferences that can be drawn from it. First, the validation of work histories was limited to one employer. (Although interviews furnished data on other exposures and occupations, these could not be validated using company resources.) Work conditions or demographic characteristics of employees STO I 61(937 Bond 14. at other facilities might differ sufficiently to render different results, or contribute substantial additional variability to inter-industry studies. The comparisons made were limited to specific work area assignments as well as to chemical and physical agents, as opposed to job titles. This approach was taken because work areas and agent exposures have been the primary variables upon which case-control studies of workplace hazards at this facility have focused (13-15). To some extent, recall or lack or recall may be attributed to the particular classification schemes used in this study, and may differ under another approach. However, many of the work areas imply job titles; e.g., machine shop, paint shop, electric shop, pipe shop. Furthermore, even chemical production work areas imply job titles, since nearly all employees assigned to such areas can be classified as operators. The interviews were administered by telephone. Face-to-face interviews were not feasible owing to substantial migration of former employees and their next-of-kin from the area. By necessity, telephone interviews are brief, and it may not be possible to obtain the respondent's complete attention. The respondent is less able to consult with records or other sources to stimulate his or her memory. It might be expected that better agreement may have resulted from face-to-face interviews, although this is a matter yet to be confirmed by investigation. While this questionnaire employed open-ended questions to elicit complete responses, it is recognized that the use of closed-ended ST01G4938 Bond 15. questions, such as a list of occupations or chemical agents of interest, might have stimulated recall and improved the performance of the respondents. On the other hand, such an instrument may introduce significant recall bias. This too, is a matter to be resolved by further study. Finally, there were few demographic data collected on the respondents (as opposed to subjects). Although one might expect recall to be influenced by such factors as the respondent's age, education, or income level, Baumgarten et al^ (9) did not find these factors to have affected agreement in their study. It is clear from this investigation that a real potential for recall bias and exposure misclassification exists in interview-based occupational case-control studies, and that the effects on risk estimation can be serious. While some factors that influence recall can be identified and adjusted during analysis, others are best handled by careful decision-making during the design stage of a study. Because so few of the determinants of recall have yet been adequately examined, there is good reason for using caution in the interpretation of such interview-based research. TABLE 1 Percentage of recorded work assignments reported in interviews by selected characteristics, study of lung cancer among chemical workers in a Texas production facility, 1940-1980 ST0 I 64939 Character!stic Number of Work Area Assignments Company Respondents Records (%) Respondent's Relationship to Subject: Self Spouse Other 358 50.6 811 47.8 429 47.6 Elapsed Time From Assignment to Interview: <15 Years 15-29.9 Years 30+ Years 275 58.9 544 48.5 785 44.6 Duration of Assignment: <1 Year 1-4.9 Years 5+ Years 436 24.e 626 49.8 542 65.7 Number of Assignments: 1 281 80.4 2-3 808 46.3 4+ 515 34.2 Quality of Interview: High Adequate Questionable or Unsatisfactory 1021 49.6 467 48.8 112 35.7 Usual Work Area Assignment* Company Respondents Records (*) 143 76.2 361 72.0 228 65.4 151 82.1 269 68.4 314 67.5 38 44.7 296 70.9 400 73.2 281 80.4 343 66.8 110 59.1 446 77.1 229 65.9 57 40.4 Principal Interviewers: A B C Group Membership Cases Decedent Controls "Living" Controls 547 50.1 531 49.9 346 47.7 519 49.5 526 47.0 559 48.7 243 73.3 259 68.7 154 74.0 245 71.8 251 70.5 238 70.2 Defined as work area assignment of longest duration, according to company records. (NOTE: Total counts for some characteristics may vary owing to missing data) TABLE 2 Percent recall of recorded work assignments predicted by logistic models, study of lung cancer among chemical workers in a Texas production facility, 1940-1980 Quality of Interview High or adequate Model Parameters - All Work Area Assignments Respondent's Duration of Number of Assignments Relationship Assignment 1 2 3 Self 2 years 5 years 10 years 62.2 66.7 73.6 53.7 59.8 69.1 45.1 52.4 64.2 Other 2 years 5 years 10 years 65.4 69.7 76.2 51.9 58.0 67.5 38.1 45.2 57.4 Questionable or unsatisfactory Self Other 2 years 5 years 10 years 2 years 5 years 10 years 23.9 27.2 34.8 26.6 30.6 38.0 29.5 34.9 44.7 28.0 33.2 42.8 35.8 42.8 54.9 29.4 35.9 47.8 ST0 IGU9U0 Quality of Interview High or adequate * Model Parameters - Usual Work Area Assignments Elapsed time from Number of Assignments Assignment to interview 1 23 1 year 5 years 10 years 88.5 87.8 86.7 88.6 86.9 84.5 88.6 86.0 82.1 Questionable or unsatisfactory 1 year 5 years 10 years 64.7 62.9 60.7 64.8 61.1 56.4 64.9 59.3 52.1 * Defined as work area assignment of longest duration according to company records. TABLE 3 Percentage of recorded work assignments expected In Interviews by selected work areas, study of lung cancer among chemical workers in a Texas production facility, 1940-1980 ST0 I 6tx94 I Work Area Number of Work Area Assignments Company Respondents Records of Usual Work Area Assignment* Respondents Number % Magnesium Construction 223 + 36 69.1 30.6 122 + C u 82.8 40.0 Power 46 65.2 26 80.8 Styrene 29 48.3 11 12.1 Chlorine 56 46.4 25 60.0 Maintenance 143 28.7 42 45.2 Machine Shop 40 47.5 21 61.9 Pipe Shop 135 55.6 85 64.7 Administrative 204 59.3 112 79.5 Vinyl-Vinylidene Chloride 11 72.7 8 100.0 Paint Shop 97 55.7 52 82.7 Overal1 1604 48.4 734 70.8 ir Defined as work area assignment of longest duration, according to company records. tNurr.ber of subjects with work area assignment recorded in company work history. Number of subjects for whom this assignment was the usual assignment. ^Number of total work area assignments (sum of number of assignments held by each subject). STO164942 REFERENCES 1. Copeland KT, Checkoway H, McMichael AJ et al. Bias doe to misclassification in the estimation of relative risk. Am J Epidemiol 1977; 105:488-95. 2. Flegal KM, Brownie C, Haas JD. The effects of exposure misclassification on estimates of relative risk. Am J Epidemiol 1986; 123:736-751. 3. Keating E, Paterson DG, Stone CH. Validity of work histories obtained by interview. J Appl Psychol 1950; 34:6-11. 4. Mosel JN, Cozan LW. The accuracy of application blank work histories. J Appl Psychol 1952; 36:365-9. 5. Jonathan G, Moore F, Roberts L. A discussion of technique and an analysis of errors in taking industrial histories in coal miners. Br J Ind Med 1957; 14:135-6. 6. Weiss DJ, Davis RV. An objective validation of factual interview data. J Appl Psychol 1960; 44:381-5. e s n9 1oi s REFERENCES -2- 7. Weiss DJ, Dawis RV, England DW et al. Validity of work histories obtained by interview. Minnesota Studies in Vocational Rehabilitation 12, September, 1S61, Industrial Relations Center, bulletin 34. 8. Goldstein IL. The application blank: how honest are the responses? J Appl Psychol 1971; 55:491-2. 9. Baumgarten M, Siemiatycki J, Gibbs GW. Validity of work histories obtained by interview for epidemiologic purposes. An 0 Epidemiol 1S83; 118:583-91. 10. Bond GG, Flores GH, She!lenberger RJ et al. Nested case-control study of lung cancer among chemical workers. Am J Epidemiol 1986; 124:53-66. 11. Bond GG. Determinants of lung cancer among chemical workers. Ph.D. dissertation, University of Michigan, 1985 University Microfilms International, Ann Arbor, Michigan. 12. Bond GG, She!lenberger RJ, Fishbeck WA et al. Mortality among a large cohort of chemical manufacturing employees. JNCI 1985; 75:859-69. REFERENCES -3- 13. Bond GG, Cook RR, Wight PC, et al. A case-control study of brain tumor mortality at a Texas chemical plant. J Occup Med 1983; 25:377-86. 14. Bond GG, Flores GH. Exposure classification in nested case-control studies (Abstract). Am J Epidemiol 1984; 120:503. 15. Bond GG, Shellenberger RJ, Flores GH. A case-control study of renal cancer mortality at a Texas chemical plant. Am J Ind Med 1985; 7:123-139. 16. SAS Institute Inc. SAS User's Guide: Basics, Version 5 Edition. Cary, NC: SAS Institute Inc, 1985. 17. Gordis L. Should dead controls be matched to dead cases? Am J Epidemiol 1982; 115:1-5. *1*16191015