Tobacco Documents Online

320 Fetterman et al. TAJ~LE XIIIo Model R2 Values for All Possible Single-Variable Regressions of Summary Statistics on CARCIN, for Each Potency Measure* Potency MAX MEAN MEDIAN MAX MEAN MEDIAN measure overall overall overall over reps over reps over reps bM 0.0190 0.0079 0.0001 0.0244 0.0061 0.0000 bs~ 0.0245 0.0045 0.0003 0.0183 0.0037 0.0001 bB_- 0.0311 0.0127 0.0004 0.0370 0.0122 0.0019 bx 0.0348 0.0155 0.0018 0.0523 0.0142 0.0011 bE 0.0438 0.0288 0.0039 0.0613 0.0334 0.0136 bt~ 0.0361 0.0295 0.0076 0.0516 0.0335 0.0093 bo2 0.0367 0.0250 0.0057 0.0467 0.0277 0.0067 bt.~ 0.0071 0.0048 0.0028 0.0169 0.0061 0.0002 *For each potency measure, the highest R-" value is bolded and the second highest R-' value is italicized to facilitate comparison between best and second best models. TABLE XIV. Models Chosen for Predicting Quantitative Carcinogenicity From 8 Potency Measures, for Combined Sample of 108 Chemicals Predictors in F-test Summary statistic Final Model R'- P-value MAX Overall bt~. bt,eo 0.09 0.0080 MEAN Overall bDt, bDz, bL~t~ 0.11 0.0071 MEDIAN Overall intercept-only MAX over reps bR. bLED 0.09 0.0079 MEAN over reps bot. bD2. bLEI~ 0.12 0.0043 MEDIAN over reps intercept-only the final model chosen again depends on the potency mea- sure being considered, and all R2 values are ~0.15. Next, the six methods of summarizing all the experi- ments for a given chemical (described in Table VIII) were examined. The question of interest here, for each of the eight potency measures separately, is which of the six possible summary statistics is the best predictor of the, TD50? The results of this analysis for the combined sam- ple are shown in Table XII. No single summary statistic stands alone in predicting TD50, although MAX over the nine strain/activation combinations contributes to predic- tion for five of the eight potency measures. Table XIII provides the model R2 for all possible single- variable regressions for each potency measure. This table allows us to answer two questions: If we can choose only one summary statistic for predicting TD50 for a given potency measure, which summary statistic would it be? And how much better is the best model compared to other possible models? For 7 of the 8 potency measures, MAX over reps is the best summary statistic for predicting TD50. However, the R'- values for the "best" and the "'second-best'" models are not very different, and all the R" values are low (<0.07). For each of the six methods of summarizing experi- ments for a chemical, the eight potency measures were examined to determine which best predicts TD50. The TABLE XV. Models Chosen for Predicting Quantitative Carcinogenicity From Strain/Activation Combinations, Overall MAX, and Qualitative MUTA for the Combined Sample of 108 Chemicals Potency Predictors in F-test measure final model R-" P-value bM MUTA, h1535" 0.10 0.0046 bet hi00, rl00 0.13 0.0097 bE_, MUTA 0.07 0.0054 bx MUTA 0.07 0.0054 bR r1535, MUTA 0.13 0.1301 I bD~ r1535, MUTA 0.13 0.0010 bo., MUTA, r1535 0.11 0.0025 bLED h1535, MUTA 0.13 0.0013 *Strain/activation combination. The first letter represents the activation (none, _hamster liver, rat liver) and the numbers represent the Salmonella strain. results of this analysis for the combined sample are pro- vided in Table XIV. The results depend on the summary statistic being considered, but in general bLED most often appears in the final model. Conclusion III In all cases, the R~ values for the "best" and the "sec- ond-best" models are not very different, and all the R2 values are low (<0.07), indicating that the relationship between mutagenic potency, regardless of how it is mea- sured, and quantitative carcinogenicity is weak. All of the models discussed above were reexamined with qualitative mutagenicity (MUTA) included in the predictor lists. In the prediction of CARCIN from the nine strain/activation combinations, MUTA was included in the final model for 7 of the 8 potency measures (see Table XV). However. the inclusion of MUTA improved the model R-~ by only 0.05 or less. In the comparison of summary statistics for each potency measure, and the comparison of potency measures for each summary statis-

tic, the addition of MUTA to the predictor list yielded final models of "intercept + MUTA" only. Again, the model R" improved by <0.05. Conclusion IV In almost all cases, the inclusion of qualitative mutage- nicity improves prediction of quantitative carcinogenicity from quantitative mutagenicity. However, the improve- ments in the model R-" values are very small (-~0.05). DISCUSSION This study examined whether the potency of the SAL mutagenicity response, measured by a number of different parameters, could be used to improve the predictivity of carcinogenicity, either qualitatively or quantitatively. Qualitative carcinogenicity was determine.d by the posi- tive/negative NTP decision on each chemical, whereas quantitative carcinogenicity was determined by the avail- able TD50s of the chemicals under study. The study being reported is not the first to investigate carcinogenic pre- dictivity from SAL, but it is the first to use and compare different measures of SAL potency. When predicting qualitative carcinogenicity within the NTP dataset of chemicals, only qualitative mutagenicity is useful: none of the quantitative measures improves the carcinogenicity prediction. When predicting quantitative carcinogenicity, MAX over replicates has the highest R2 value for 7 of 8 potency measures. However, all R2 values are less than 0.07. Of the eight different potency measures considered, bLED was most often one of the best predictors of quantitative carcinogenicity. In all cases, however, the relationship between mutagenic potency predictors and quantitative carcinogenicity is very weak, supporting the low correlations between mutagenic and carcinogenic po- tency reported previously (see Table I). Piegorsch and Hoel [ 1988] demonstrated that the corre- lation between mutagenic and carcinogenic potencies could be affected by the chemical structural subsets ana- lyzed. According to their results, although the correlation for the entire set of chemicals was 0.44, the correlations for the various structural subsets (congeneric sets) ranged from 0.25 to 0.99. Similarly, Hatch [1992] showed a higher correlation when the calculations were limited to specific structural classes of chemicals. An interesting alternative was presented by Bogen [1995], who studied only those chemicals that were positive in both the SAL and rodent bioassays. He found an improved prediction of carcinogenic potencies fron~ mutagenic potencies for chemicals positive in rodents and SAL. In contrast, we analyzed the entire diverse set of ~hemicals, rather than subdivide it into chemical or response classes, because there were too few mutagenic chemicals within each subset. Predicting Carcinogenicity From Mutagenic Potency 321 The goal here was to determine whether the results obtained were dependent on a particular potency measure and/or a way of summarizing the data. Our study firmly establishes that the predictive relationship between quan- titative SAL mutagenicity and carcinogenicity is, at best, weak, regardless of the potency measure used. The lack of a strong predictive relationship between SAL mutage- nicity and rodent carcinogenicity hardly seems surprising. Mutagenesis may be only the first step in some pathways that lead to cancer. The inference drawn here, however, is that thi~ DNA damage as measured by mutations in SAL is not the rate-limiting consideration in the ultimate development of cancer in rodents. Currently, similar studies are examining the predictive relationship between mutagenicity and rodent carcinoge- nicity using several other STTs (mouse lymphoma test, sister chromatid exchange, and chromosome aberrations). The purpose of these studies is to determine whether po- tency of the ST'[' responses, alone or in combination, could be used to improve the qualitative or quantitative prediction of carcinogenicity. ACKNOWLEDGMENTS We thank Dr. Lois Gold, University of California, Berkeley, for providing us with the carcinogen TD50 cal- culations used in these analyses, and Dr. Leslie Bernstein for supplying the computer code for analysis of potency. Portions of this research performed at UNC-CH were supported by contract 273-90-1-0005 from the National Institute of Environmental Health Sciences. Dr. Kim's research was partially supported by the Korea Research Foundation through its 1995 Overseas Research Program for University Professors. REFERENCES Ames BN, Durston WE. Yamasaki E, Lee FD (1973): Carcinogens are mutagens: A simple test system combining liver homogenates for activation and bacteria for detection. Proc Natl Acad Sci USA 70:2281-2285. Ames BN, McCann J, Yamasaki E ( 1975): Methods for detecting carcin- ogens and mutagens with the Salmonella/mammalian-microsome mutagenicity test. Mutat Res 31:347-364. Bemstein L, Kaldor J, McCann J, Pike MC ( 1982): An empirical ap- proach to the statistical analysis of mutagenesis data from the Salmonella test. Mutat Res 97: 267-281. Bogen, KT (1995): Improved prediction of carcinogenic potencies from mutagenic potencies for chemicals positive in rodents and the Ames test. Environ Mol Mutagen 25:37-49. Gold LS, Sawyer CB, McGaw R, Buckman GM, deVeciana M, Levin- son R, Hooper NK, Havender WR, Bernstein L. Peto R, Pike MC, Ames BN (1984): A carcinogenic potency database of the standardized results of animal bioassays. Environ Health Perspect 58:9 -22. Haseman JK. Zeiger E, Shelby MD, Margolin BH. Tennant RW f 1990): Predicting rodent carcinogenicity from four in vitro genetic toxic-

322 Fetterman et al. ity assays: An evaluation of 114 chemicals studied by the N~- tional Toxicology Program. J Am Stat Assoc 85:964-971. Hatch FT, Knize MG, Moore DH II, Felton JS (1992): Quantitative correlation of mutagenic and carcinogenic potencies for betero- cyclic amines from cooked foods and additional aromatic amines. Murat Res 271:269-287. Margolin BH, Kaplan N, Zeiger E (1981): Statistical analysis of the Ames Salmonella/microsome test. Proc Nat[ Acad Sci USA 78:3779 -3783. Margolin BH, Kim BS, Risko KJ (1989): The Ames Salmonella~micro- some mutagenicity assay: Issues of inference and validation. J Am Stat Assoc 84:651-661. Margolin .BH, Kim BS, Smith MG, Fetterman BA, Piegorsch WW, Zeiger E (1994): Some comments on potency measures in muta- genicity research. Environ Health Perspect 102(Suppt 1):91-94. Margolin BH, Risko ICI (1988): The statistical analysis of in vivo geno- toxicity data: Case studies of the rat hepatocyte UDS and mouse bone marrow micronucleus assays. In Ashby J, de Serres FJ. Shelby MD, Margolin BH, Ishidate M Jr, Becking, GC (eds): "Evaluation of Short-Term Tests for Carcinogens: Report of the International Programme on Chemical Safety's Collaborative Study on In Vivo Assays." Cambridge University Press, pp 1.31 - 1.42. McCann J, Choi E, Yamasaki E, Ames BN ( 1975): Detection of carcino- gens as mutagens in Salmonella microsome test: Assay of 300 chemicals. Proe Natl Acad Sci USA 72:5135-5139. McCann J, Gold LS, Horn L, McGill R, Graedel TE, Kaldor J (1988): Statistical analysis of Salmonella test data and comparison to results of animal cancer tests. Murat Res 205:183-195. Meselson M, Russell K (1977): Comparisons of carcinogenic and muta- genie potency. In Hiatt HH, Watson JD, Winsten JA (eds): "'Ori- gins of Human Cancer. Book C: Human Risk Assessment." Cold Spring Harbor, NY: Cold Spring Harbor Press, pp 1473-1481. Parodi S, Taningher M, Romano P, Grilli S, Santi L (1990): Mutagenic and carcinogenic potency indices and their correlation. Teratogen Carcinogen Mutagenen 10:177-197. Piegorsch WW, Hoel DG (19881: Exploring relationships between mum- genie and carcinogenic potencies. Mutat Res 196:161 - 175. Purchase IFH, Longstaff E, Ashby J, Styles JA, Anderson D, Lefevre PA, Westwood FR (I 976~: Evaluation of six short term tests for detecting organic chemical carcinogens and recommendations for their use. Nature 264:624-627. Sugimura T, Sato S, Nagao M. Yahagi T, Matsushima T, Seino Y, Takeuchi M, Kawachi T t 1976): Overlapping of carcinogens and mutagens. In Magee. PN, Takayama S, Sugimura, T, Matsu- shima, T (eds): "Fundamentals of Cancer Prevention." Balti- more: University Park Press, pp 191-215. Tennant RW, Margolin BH. Shelby MD, Zeiger E, Haseman JK, Spal- ding J, Caspary W, Resnick M, Stasiewicz S, Anderson B, Minor R (1987): Prediction of chemical carcinogenicity in rodents from in vitro genetic toxicit), assays. Science 236:933-941. Zeiger E (1987): Carcinogenicity of mutagens: Predictive capability of the Salmonella mutagenesis assay for rodent carcinogenicity. Cancer Res 47:1287-1296. Zeiger E, Haseman JK, Shelby MD, Margolin BH, Tennant RW ( t990): Evaluation of four in ~itro genetic toxicity tests for predicting rodent carcinogenicity: Confirmation of earlier results with 41 additional chemicals. Environ Mol Mutagen 16(Suppl 18): 1 - 14. Zeiger E, Risko KJ, Margolin BH (1985): Strategies to reduce the cost of mutagenicity screening with the Salmonella assay. Environ Mutagen 7:901-911. Zeiger E, Tennant RW (1986): Mutagenesis, clastogenesis, carcinogene- sis: Expectations, correlations, and relations. In Ramel C, Lam- bert B, Magnusson J (eds): "Genetic Toxicology of Environmen- tal Chemicals. Part B: Genetic Effects and Applied Mutagene- sis." New York: Wile.~. pp 75-84. Accepted K. R. Tindall

2063633499

Workplace Conditions, Socioeconomi ' 475" Status, and the Risk of Mortality and Acute Myocardial Infarction: The Kuopio Ischemic Heart Disease Risk Factor Study Objectives. This study investi- gated whether the association be- tween workplace conditions and the risk of all-cause and cardiovascular mortality and acute myocardial infarc- tion differed by socioeconomic sta- Methods. Prospective data were used to examine these associations in 2297 Finnish men, with adjustment for prevalent diseases and biological, behavioral, and psychosocial covari- ares, and stratified by employment smam and workplace social support. Results. Elevated age-adjusted relative hazards for all-cause mortal- ity were found for men who reported high demands, low resources, and low income; high demands, high resouro~ and low income; and low demand~ high resources, and low income. Similar patterns were found for cardiovascular mortality. In con- trast, elevated age-adjusted t~iatj_'ve lmzar~ for acute myocardial infarc- tion were observed only in men who reported high demands, low resources, and low income. These result~ did not differ by level of workplace social support or employ- ment status. Conc/us/ons. The negative ef- feet~ of workplace conditions ~ mortality and of myocardial infarc- tion risk depended on income level and were largely mediated by known risk factors. (Am J Public Health~ 1997;87:617-622) John Lynch, PhD, MPH, Niklas Krause, MD, PhD, George A. Kaplan, PhD, Jaakko Tuomilehto, MD, PhD, and Jukka T. Salonen, MD, PhD Introduction Researchers' understanding of how organizational and psychosocial features of work affect morbidity and mortality has been greatly influenced by the idea that poor health outcomes may be associated with work that is psychologically demand- ing but offers few opportunities for control.t-3 This notion has been opemtion- alized in a variety of ways and has received empirical support in a large number of cross-sectional and case- control studies,4 but when studied prospec- tively, the evidence has been more mixed.~-6 In addition, relatively little is known about the pathways through which job characteristics might influence disease risk.7 In their review of these studies, Schnall and Landsbergis~ suggest the need to expand the basic demand/control formulation to include other important workplace characteristics such as social support, physical exertion, job security, and hazardous exposures. They also argue that it is important to adjust the associa- tion between job conditions and disease risk to control for potential confounding by socioeconomic Status (SES). Previous studies have generally adopted this line of reasoning and treated SES as a con- founder of the association between, job characteristics and health outcomes in an attempt to find the "independent" effect of workplace factors on health,s-t° In contrast, we believe that statisti- cally partitioning the independent effects of SES and job conditions on disease risk ignores important structural connections between social class and work.)1 Further- conditions on health. We investigated the association between workplace demands and resources and the risk of all-cause mortality, cardiovascular mortality, and incident acute myocardial infarction at different levels of SES, as measured by economic reward. These associations were examined prospectively in a popula- tion-based sample of Finnish men, with adjustment for prevalent diseases and biological, behavioral, and psychosocial covariates, and in subsamples stratified by employment status and workplace social support. Methods Study Population The subjects were participants in the Kuopio Ischemic Heart Disease Risk Factor Study, which was designed to investigate previously unestablished risk factors for ischemic heart disease, carodd atherosclerosis, and other related out- comes in a population-based sample of men in eastern Finland.t2 Of 3433 eligi- ble men aged 42, 48, 54, or 60 years resident in the town of Kunpio or its surrounding communities, 198 could not be included because of death, serious disease, or migration away from the area; of the remainder, 2682 (82.9%) agreed to John Lynch and Niklas Krause are with the Western Consortium for Pubiic Health. and George A. Kaplan is with the California Depart- merit of Health Services, at the Human Popula- tion Laboratory, Berkeley, Calif. Jaakko Tu- omilehto is with the Nationa/ Public Health Institute, Helsinki, F'mland. Jukka T. Salonen is with the Research Institute of Public Health. University of Kuopio, Kuopio, F'mland. more, it is possible that having high levels Requests for reprints should be sent to John of income or education may provide Lynch. PhD, MPH, Human Population Labora- cognitive and tangible resources that tory, 2151Berkeley Way, Annex2, Berkeley, CA 94704. could reduce the effects of poor working This paper was accepted October ~. 1996. APrif I~3T- rV°f- ST,N'o-4 American Journni of Public Health 617

participate in the study. Baseline examina- tions were conducted between March 1984 and December 1989. No marked sociodemographic differences have been found between participants and nonpartici- pants.13 Complete information on work- place demands, resources, economic re- ward, and all covadates was available for 2297 men for the mortality analyses. There were 289, 315, 1387, and 306 men in the 42-, 48-, 54-, and 60-year-old age groups, respectively. A total of 570 of these men were excluded from the acute myocardial infarction incidence analyses (n = 1727) because of a prior history of acute myocardial infarction, angina pecto- ris, nitroglycerine use, or positive findings of angina from the London School of Hygiene Cardiovascular Questionnaire.~'~ Assessment of Workplace Demands, Resources, .ayd Economic Reward At the baseline examinations partici- pants completed detailed questionnaires including items on aspects of their work environment, income, and education. Items that conformed to important theoretical domains discussed in the literature were considered for inclusion in the measure- ment of workplace demands.4 In accor- dance with suggestions made in this literature, items on risk of unemployment, accidents, and physical exertion were included to supplement the questions about psychological demands. Partici- pants were asked to rate on a Likert-type scale (0-4) how much mental strain or stress the following things caused them at work: excessive supervision of time sched- ules, troublesome supervisors, trouble- some fellow workers, job responsibility, poorly defined tasks and responsibilities, risk of accidents, risk of unemployment. irregular work schedules, and the mental strenuousness of work. They were also asked how often they had work deadlines, how much stress this caused them, and the physical strenuousness of their work. Scores for the demands scale were im- puted on the basis of nonmissing values for men who had no more than 2 missing items. Men who had more than 2 missing items were excluded from the analyses. The I1 individual items were dichoto- mized at the midpoint of the rating scale, so that only when men reported that the particular aspect of work caused them more .than "average" sla-ain were their responses considered positive. The di- chotomized items were then summed to form the workplace demands scale, which had high internal consistency (Cronbach's alpha = .78). Resources were assessed with ques- tions asking participants to rate statements concerning the degree to which their work was interesting, allowed them to use their skills and capabilities, allowed them to feel composed and competent, was enjoy- able, and was meaningful. Imputation of items and scoring of the resources scale were done in the same way as for demands (Cronbach's alpha = .77). Eco- nomic reward was assessed by self- reported income, dichotomized so that the lowest 40% of income earners were' considered low. Previous analyses had shown that the bottom two quintiles of the income distribution were at significantly elevated risk of mortality and acute myocardial infarction,ts The distributions of scores for demands and resources were dichotomized at the median, producing eight possible combinations of high and low demands, resources, and economic reward. Assessment of Follow-Up Events Participants were followed until the end of December 1994 for the mortality analyses, with a median follow-up of 8.1 years (range: 5.0-10.8). For the acute myocardial infarction analyses men were followed until the end of December I992, for a median of 6.1 years (range: 3.0-8.8). All-cause and cardiovascular mortality were ascertained by linkage to the Na- tional Death Registry, which is main- tained for all Finnish citizens. Classifica- tion of death was based on the underlying cause, reviewed at the National Center of Statistics of Finland. Cardiovascular deaths were classified according to the ninth revision of the International Classifica- tion of Diseases (ICD) for ICD codes 390-459. Of the 189 deaths, 93 were from cardiovascular causes. First-event, nonfatal acute myocar- dial infarctions and coronary deaths were ascertained by linkage to an acute myocar- dial infarction register established under the World Health Organization's MONICA (Monitoring of Trends and Determinants of Cardiovascular Diseases) project,t6 There were 89 fatal or nonfatal incident acute myocardial infarctions recorded in this group of men. Assessment of Covariates As part of the baseline examinations, extensive information was collected on biological, behavioral, and psychosocial covariates. In addition, the prevalence of diseases was assessed by detailed medical histories. All covariates included in these analyses have been shown to be associ- ated with mortality and acute myocardial infarction,is Biological covariates. Biological co- vadates included plasma fibrinogen, high- density lipoprotein, serum apolipoprotein B (APO B), serum triglyceddes, blood hemoglobin and leukocyte count, serum ferritin and copper, hair mercury, systolic blood pressure, body mass index, height, and eardiorespiratory fitness. The meth- ods of assessment for each of these factors have been previously described.tSAT-22 Behavioral covariates. Alcohol con- sumption, measured in grams per week, was assessed by dietary recording for a 4-day period and also for the previous 12 months, by self-administered question- naire.23 Smoking was measured by ques- tionnaire and classified for this analysis as "never smoked, .... former smoker," and "current smoker" (measured in pack- years). The total duration (minutes per week) of conditioning physical activity was assessed from a 12-month leisure- time history,z~ Psychosocial covariates. Depression was assessed from a shortened 180-item version of the Minnesota Multiphasic Personality Inventory that had previously been used in Finnish populations. Hope- lessness was assessed with two question- naire items, scored on a five-point Likert scale.24 Marital status was assessed by questionnaire and categorized as "mar- ried," "single," or "divorced/widowed." Prevalent diseases. Prevalent dis- eases were ascertained from detailed medical histories, medication records, and examinations at baseline. Indicator vari- ables were used to represent a history of cardiovascular disease (symptomatic, asymptomatic, claudication or. cardiomy- opathy, and other), hypertension, stroke, diabetes, respiratory disease, and cancer. Statistical Analysis Associations between workplace de- mands, resources, and economic reward and all-cause mortality, cardiovascular mortality, and acute myocardial infarction were assessed with Cox proportional hazard models.2~ The analyses were con- ducted with the PHREG procedure in SAS version 6.09 on a Sun Spare Station II.26 To assess the impact of covariate adjustment on the age-adjusted relative hazards (RHs), we calculated the propor- tion of excess relative risk (hazard) accounted for by covariate adjustm.ent as [RH~,~j~a~- 1] 618 Atnerican Journal of Public Health April 1997. Vol. 87. No. 4

Workplace Conditions and Mortality TABLE 1--Workplece Demands, Resources, and Economic Reward and Prevalence of Selected Sociodemographic Characteristics at Baseline among Men in Eastern Finland (n = 2297) Prevalent Level of Demands/ Age 55 or Blue- White- Not Ischemic Heart Low Social Completed Resources/ Older, % Farmers, % Collar, % Collar, % Employed, % Disease, % Support, % High School, % Income No. (%) (n = 346) (n = 341) (n = 984) (n = 944) (n = 96) (n = 570) (752) (n = 393) High/Low/Low 260 (11.3) 11.9 15.0 17.2 4.0 17.7 15.8 13.3 2.5 High/Low/High 353 (15.4) 12.1 5.6 13.8 20.1 17.7 13.9 20.6 19.1 Low/Low/Low 159 (6.9) 9.2 12.6 7.5 4.5 3.1 7.7 7.7 1.8 Low/Low/High 361 (15.7) 9.0 5.9 13.6 21.5 7.3 7.2 14.9 29.3 High/High/Low 261 (11.4) 17.1 19.1 15.4 4.6 18.8 20.5 9.4 1.3 High/High/High 244 (10.6) 11.0 6.7 10.6 12.3 14.6 12.1 10.1 12.2 Low/High/Low 243 (10.6) 12.4 24.1 10.5 5.8 11.5 11.9 10.4 1.8 Low/High/High 416 (18.1) 17.3 11.1 11.5 27.2 9.4 10.8 13.6 32.1 TABLE 2~Workplace Demands, Resources, and Economic Reward and the Relative Hazard (RH) of All-Cause Mortality among Men in Eastern Finland (n = 2297), A~usted forage Plus,.. Adjusted for Age Prevalent Behavioral Psychosocial Biological Level of Demands/ Disease= Covadatesb Covariatesc Covariatesd All Covadates Resources/Income RH (95% CI) RH (95% CI) RH (95% CI) RH (95% CI) RH (95% CI) RH (95% CI) High/Low/Low 3.00 (1.81, 4.98) 2.38 (1.42, 4.01) 2.58 (1.55, 4.31) 2.00 (1.16, 3.42) 2.33 (1.39, 3.89) 1.64 (0.94, 2.87) High/Low/High 0.94 (0.50, 1.76) 0.85 (0.45, 1.60) 0.87 (0.46, 1.64) 0.76 (0.40, 1.44) 0.90 (0.48, 1.70) 0.79 (0.41, 1.50) Low/Low/Low 1.05 (0.51, 2.16) 0.94 (0.45, 1.94) 1.04 (0.50, 2.14) 0.82 (0.39, 1.71) 0.86 (0.41, 1.79) 0.79 (0.38, 1.67) Low/Low/High 0.74 (0.37, 1.47) 0.76 (0.38, 1.51) 0.72 (0.36, 1.42) 0.69 (0.33, 1.33) 0.78 (0.39, 1.56) 0.77 (0.38, 1.55) High/High/Low 2.15 (1.26, 3.68) 1.61 (0.93, 2.80) 1.90 (1.11, 3.25) 1.58 (0.91,2.75) 1.48 (0.86, 2.56) 1.11 (0.62, 1.98) High/High/High 0.59 (0.26, 1.33) 0.53 (0.23, 1.18) 0.58 (0.26, 1.30) 0.52 (0.23, 1.18) 0.53 (0.23, 1.18) 0.47 (0.21, 1.08) Low/High/Low 2.30 (1.35, 3.92) 1.97 (1.15, 3.37) 1.99 (1.16, 3.41) 1.83 (1.06, 3.15) 1.73 (1.01, 2.97) 1.30 (0.74, 2.27) Low/High/High Reference Reference Reference Reference Reference Reference Note. CI = confidence interval. =Cardiovascular disease (symptomatic, asymptomatic, cardiomyopathy, claudication and other), hypertension, stroke, diabetes, respiratory disease, and cancer. bSmoking, alcohol consumption, and physical activity. ¢Hopelessness, depression, and marital status. ~Plasma fibdnogen, high-density lipoprotein, serum apolipoprotein B, serum tdglyceddes, blood hemoglobin and leukooytes, serum ferdtin and copper, hair memury, systolic blood pressure, body mass index, height, and cardiorespiratory f'dness. Results The 27 covariates were grouped into four categodes~prevalent diseases and biological, behavioral, and psychosocial covariates--and analyses conducted in two phases. First, we examined associa- tions with separate adjustment for each group of covariates and age. In the second stage, associations were adjusted for age and all 27 covariates simultaneously. In all cases hazards were relative to the low- demands, high-resources, high-income group. Table 1 shows sociodemographic characteristics for the eight combinations of demands, resources, and income. There were striking differences in the distribu- tion of job demands, resources, and income by age, education, white-collar employment, prevalent ischemic heart disease, and unemployment. Men who had jobs with low demands were almost twice as likely as men in work with high demands to have completed high school (65% vs 35%). Table 2 presents the relative hazards for all-cause mortality by combination of demands, resources, and income, adjusted for age, for age plus each covariate group separately, and for age plus all covariates simultaneously. Significantly elevated age- adjusted relative hazards for all-cause mortality were found for men who re- ported high demands, low resources, and low income (RH = 3.00; 95% confidence interval [CI] = 1.81, 4.98); high de- mands, high resources, and low income (RH = 2.15; 95% CI = 1.26, 3.68); and low demands, high resources, and low income (RH = 2.30; 95% CI = 1.35, 3.92). Separate adjustment for each covari- ate group attenuated the magnitude of the associations. For example, the excess rel- ative hazard for the high-demand, low- resource, low income group was reduced by 31% after adjustment for prevalent disease, by 21% after adjustment for behavioral covariates, by 50% after adjust- ment for psychosocial covariates, and by 34% after adjustment for biological covari- ates. Simultaneous adjustment for all covariates reduced the excess reladve hazard by 68%. Table 3 presents the relative hazards for cardiovascular mortality by combina- tion of demands, resources, and income, with the same adjustments by age and covariates. The pattern of findings was very similar to that for all-cause mortality. Significandy elevated age-adjusted rela- April 1997. Vol, 87. No. 4 American Journal of Public Health 619

TABLE 3---Workplace Demands, Resources, and Economic Reward and the Relative Hazard (RH) of Cardiovascular Mortality among Men in Eastern Finland (n --- 2297) Adjusted for Age Plus.. Adjusted Level of Demands/ for Age Prevalent Behavioral Psychosocial Biological All Resources/ Disease Covadates Covariates Covariates Covadates Income RH (95% CI) RH (95% CI) RH (95% CI) RH (95% CI) RH (95% CI) RH (95% CI) High/Low/Low 3.12 (1.48, 6.60) 2.05 (0.96, 4.40) 2.59 (1.22, 5.52) 1.94 (0.88, 4.29) 2.28 (1.07, 4.89) 1.54 (0.67, 3.54) High/Low/High 0.97 (0.38, 2.45) 0.80 (0.31,2.03) 0.91 (0.36, 2..32) 0.74 (0.29, 1.90) 0.88 (0.34, 2.24) 0.82 (0.31, 2.14) Low/Low/Low 1.49 (0.57, 3.93) 1.16 (0.44, 3.08) 1.43 (0.54, 3.78) 1.13 (0.42, 3.01) 1.03 (0.38, 2.75) 0.83 (0.30, 2.33) Low/Low/High 0.87 (0.33, 2.28) 0.89 (0.34, 2.35) 0.84 (0.32, 2.20) 0.76 (0.29, 2.01) 0.97 (0.36, 2.56) 0.94 (0.35, 2.55) High/High/Low 2.75 (1.28, 5.90) 1.53 (0.69, 3.37) 2.33 (1.08, 5.03) 1.95 (0.88, 4.29) 1.63 (0.74, 3.58) 1.12 (0.48, 2.61) High/High/High 0.49 (0.14, 1.78) 0.39 (0.11, 1.43) 0.47 (0.13, 1.72) 0.42 (0.11, 1.52) 0.39 (0.11, 1.43) 0.37 (0.10, 1.35) Low/High/Low 2.29 (1.03, 5.06) 1.72 (0.77, 3.82) 1.88 (0.84, 4.21) 1.84 (0.82, 4.13) 1.49 (0.66, 3.35) 0.99 (0.42, 2.30) Low/High/High Reference Reference Reference Reference Reference Reference Note. CI = confidence interval. =Covariates as in Table 2. TABLE 4--Workplace Demands, Resources, and Economic Reward and the Relative Hazard (RH) of Incident Acute Myocardial Infarction among Men in Eastern Finland (n = 1727) Adjusted for Age Plus. Adjusted for Age Behavioral Psychosocial Biological All Level of Demands/ Covadates Covadates Covadates Covariates Resources/Income No. RH (95% CI) RH (95% CI) RH (95% CI) RH (95% CI) RH (95% CI) High/Low/Low 170 2.59 (1.36, 4.94) 2.30 (1.20, 4.41) 2.18 (1.11, 4.28) 1.94 (1.00, 3.76) 1.57 (0.78, 3.18) High/Low/High 274 0.67 (0.29, 1.57) 0.60 (0.26, 1.41) 0.61 (0.26, 1.48) 0.61 (0.26, 1.44) 0.50 (0.21, 1.20) Low/Low/Low 115 0.62 (0.21, 1.87) 0.60 (0.20, 1.81) 0.56 (0.18, 1.69) 0.54 (0.18, 1.62) 0.41 (0.13, 1.29) Low/Low/High 320 1.25 (0.63, 2.49) 1.22 (0.61, 2.41) 1.24 (0.62, 2.46) 1.30 (0.65, 2.58) 1.11 (0.55, 2.24) High/High/Low 144 1.04 (0.44, 2.44) 0.91 (0.39, 2.15) 0.86 (0.36, 2.07) 0.62 (0.25, 1.50) 0.55 (0.22, 1.35) High/High/High 175 0.63 (0.23, 1.71) 0.60 (0.22, 1.54) 0.59 (0.22, 1.62) 0.52 (0.19, 1.44) 0.43 (0.15, 1.22) Low/High/Low 175 0.93 (0.41,2.10) 0.83 (0.38, 1.89) 0.85 (0.37, 1.95) 0.70 (0.30, 1.60) 0.65 (0.28, 1.52) Low/High/High 354 Reference Reference Reference Reference Reference Note. CI = confidence interval. =Covadates as in Table 2. tive hazards for cardiovascular mortality were found in the same groups as for all-cause mortality. Separate adjustment for each covafiate group attenuated the magnitude of the associations. Simulta- neous adjustment for all covariates re- duced the excess relative hazard by 75%. Table 4 presents the relative hazards for incident cases of acute myocardial infarction by combination of demands, resources, and income, adjusted for age, for age plus each covariate group sepa- rately, and for age plus all covariates simultaneously. As 570 men with preva- lent ischemic heart disease had already been excluded from these analyses, there was no further adjustment for other prevalent diseases. In contrast to mortal- ity. significantly elevated age-adjusted relative hazards for acute myocardial infarction were observed only in men who reported high demands, low resources, and 10'~: iiicome (RH = 2.59; 95% CI = 1.36, 4.94). Simultaneous adjustment for behavioral, psychosocial, and biological covariates decreased the age-adjusted rela- tive hazard for men with high demands, low resources, and low incomes by 64% to 1.57 (95% CI = 0.78, 3.18). Discussion These results show that the effect of job conditions on mortality and acute myocardial infarction depends on the level of economic reward, and that these associations are largely mediated by known risk factors. Our findings are consistent with the effort-reward imbal- ance model proposed by Siegrist, which suggests that the imbalance between high job demands and high psychological immersion in work roles and low eco- nomic and psychosocial rewards is associ- ated with poor health outcomes.27 In addition, these findings are consistent with evidence from other studies, which found stronger associations between poor job conditions and health in less educated men and in blue-collar workers) How- ever, in stratified analyses (not shown), there was no evidence that the patterns of increased mortality and acute myocardial infarction risk differed by the level of workplace social support. Similar patterns of increased risk were found for both all-cause and carclio- vascular mortality. The highest mortality risks were found in men whose work was demanding with low resources and low economic reward, while men with the same levels of demand and economic reward but with high resources had 620 American Journal of Public Health April 1997, Vol. 87. No. 4

Workplace Conditions and Mortality somewhat lower mortality risks. Surpris- ingly, we found elevated mortality risks in men with low-demand, high-resource, low-income jobs (RH = 2.30). This might be explained as an effect of low income, but men with the same level of job demands and income but low resources were not at increased" risk. As the low- demand, high-resource, low-income group had the highest proportion of farm and forestry workers (31%), it is possible that the measures of demands and resources used in this study did not fully address specific negative job characteristics, such as close exposure to organic and chemical pollution, associated with work in these occupations.2s In addition, the fact that men in jobs with low demands, high resources, and low incomes were not at increased risk of acute myocardial infarc- tion suggests that other factors might be responsible for their increased mortality dsk. When the association between job conditions, income, and mortality was adjusted for covadates, biological risk factors reduced the magnitude of the associations by between 34% and 60%. In addition, psychosocial factors and preva- lent diseases reduced the associations by as much as 50%. However, as job conditions, income, psychosocial charac- teristics, and prevalent diseases were all assessed at the same point in time, it is impossible to disentangle their temporal sequencing. One interpretation of these results is that over time, the effects of poor working conditions and low economic reward lead to feelings of hopelessness and depression, poorer behavioral and biological risk factor profiles, and higher levels of morbidity, which contribute to increased mortality risk. As we have argued elsewhere, adjustment for factors that may be consequences of working in poor conditions with low economic re- wards would constitute overadjustment.29 The association between job condi- tions, economic reward, and incident acute myocardial infarction showed that men in high-demand, low-resource, low- income jobs had an age-adjusted risk of acute myocardial infarction that was more than 2.5 times that of men with low- demand, high-resource, high-income jobs. The magnitude of this association was reduced by more than 40% with adjust- ment for biological risk factors for acute myocardial infarction, and by over 60% with simultaneous adjustment for all covafiates. Several issues should be mentioned before conclusions are drawn from these results. First, the measure of workplace demands may have been subject to reporting bias because it was based on a self-assessment of the extent of stress or strain associated with aspects of work, although mortality and acute myocardial infarction risks remained elevated even after adjustment for depression and hope- lessness. While the most accurate assess- ment of job demands and resources would be achieved by a combination of subjec- tive and objective measures, high correla- tions between subjective assessments and expert ratings of job conditions have been demonstrated,a° Furthermore, there is no rationale for how a bias in the self- reporting of job demands could explain the overall income-dependent pattern of our findings for mortality and acute myocardial infarction. Second, it is pos- sible that the measure of resources used in this study did not fully capture both the "skill discretion" and "decision author- ity" dimensions of workplace control that have been suggested as important modifi- ers of workplace demands.ao Third, our assessment of job de- mands, resources, and income was based on a single measurement and does not take into account changes in job expo- sures over time. Furthermore, structural alterations to the Finnish economy have seen large increases in unemployment and changes in the occupational structure of the region.3~ However, our results were no different in stratified analyses (not shown) that excluded men who reported any change in job title over the last 10 years or in other analyses that excluded men who were either unemployed or retired at baseline. Fourth, while our findings are based on a population of men in eastern F'miand, we believe these results may be applicable to similar populations beyond the immedi- ate confines of the region. Kuopio is the major provincial center in eastern F'miand and has an administrative, industrial, and service-based economy dominated by processing of farm, food, metal, and forest products. Most risk factors for mortality and acute myocardial infarction in Fin- land have been documented in other 32 populations. However, because this sam- ple is limited to middle-aged men, it is unclear whether these findings can be applied to the relationship between work- ing conditions and income and mortality and acute myocardial infarction in women. To our knowledge, this is the first study to show that an increased mortality and acute myocardial infarction risk asso- ciated with organizational, physical, psy- chological, and social aspects of work was concentrated in low-income groups. With respect to informing interventions, our findings could be interpreted in three contexts. First, while there are a myriad of health-related interventions that target the workplace, relatively few--with perhaps the exception of programs to reduce toxic exposures~ectly address the physical, organizational, psychosocial nature of work itself. The majority of so-called workplace programs are individually ori- ented psychosocial and behavioral modifi- cation interventions that use the work- place as the site of program delivery. In this context, our findings imply that these efforts will be most effective by attempt- ing to alter the risk factor profiles of low-income workers. Second, a similar interpretation of our results suggests that interventions that do focus on the actual task requirements and organizational characteristics of work should also focus on those low-income groups that bear the highest cardiovascu- lar disease and mortality burden. These interventions could focus on workplace design by reducing psychological and physical demands and increasing skill utilization, job satisfaction, and economic rewards. This approach would consider low income as an internal feature of the workplace, which, like other job demands and resources, could be modified. While efforts to improve the conditions and economic returns of work would be laudable, it is also important to remember that low income is representative of a whole set of life experiences that extend beyond work life into family, recreational, and social domains. Third, we have shown that jobs with higher demands are more prevalent in Iow-SES groups. In addition, Iow-SES groups have fewer educational and eco- nomic resources with which to gain better jobs over lime, and so may have greater exposure to poor working conditions over the lifecourse. In this way, social position structures both the likelihood and duration of exposure to work that is detrimental to health. Several investigators have argued that the effect of work conditions on health must be considered in the context of the powerful economic, political, and social forces that determine both the distribution of and changes in potentially pathogenic job characteristics across dif- ferent population groups.IL33-37 These broader structural features of society determine the types of jobs that are available for particular sectors of the population. c.rl April 1997. Vol. 87. No. 4 American Journal of Public Health 621

Lynch et ~1. Interventions that focus on the re- ward and organizational features of extant jobs witl not necessarily affect the power- ful economic, political, social, and tectmo- logical forces that generate and sustain both jobs with poor conditions of employ- ment and the system of social stratifica- tion that constrains employment opportu- nities for Iow-SES workers. Increased economic rewards, job enrichment, and work democratization are important, but they should exist within a broader context of life enrichment and social democratiza- tion for Iow-SES groups. If poor job conditions are just one of many deleteri- ous exposures for people of low SES, then we need to see the relationship between work conditions and health in the broader framework of a series of interacting circumstances, events, and behaviors that cascade over the lifecourse3829 and that ultimately place low-SES groups at higher risk of morbidity and mortality. [] Acknowledgments This work was supported in part by grant HL44199 from the National Heart, Lung, and Blood Institute and by grants from the Acad- emy of Finland and the Famish Ministry of Education. References 1. Karasek R, Tbeorell T. Healthy Work. New York, NY: Basic Books; 1990. 2. Ham M. Job strain and ischacmic heart disease: an epidemiologic study of metal workers. Ann Clin Res. 1988;20:143-145. 3. Johnson JV, Hall EM, Theorell T. Com- bined effects of job s~ain and social isolation on cardiovascular disease morbid- ity and mortality in a random sample of the Swedish male working population. ScandJ Work Environ Health. 1989;15:271-279. 4, Schnall PL, Landsbergis PA. Job strain and cardiovascular disease. Annu Rev Public Health. 1994;15:381-411. 5. Reed DM, Lacroix AZ, Karasek RA, Miller D, MacLean CA. Occupational strain and the incidence of coronary heart disease. Am J Epidemiol. 1989;129: 495-502. 6. Alterman T, Shekelle R.B, Vernon SW, Buran KD. Decision latitude, psychologic demand, job strain, and coronary heart disease in the Western Electric Study. Am J Epidemiol. 1994;139:620-627. 7. Greenlund KJ, Liu K, Knox S, McCreath H, Dyer AR, Gardin J. Psychosocial work characteristics and cardiovascular disease risk factors in young adults: the CARDIA study. Soc Sci Med. 1995;41:717-723. 8. Falk A, Hanson BS, lsacsson S-O, Oster- gren P-O. Job strain and mortality in elderly men: social network, support, and influence as buffers. Am J Public Health. 1992;82:1136-1139. 9. Karasek RA, Theorell T, Schwartz fiE, Schnall PL, Pieper CF, Michela JL. Job characteristics in reladon to the prevalence of myocardial infarction in the US Health Examination Survey (HES) and the Health and Nutrition Survey (HANES). Am J Public Health. 1988;78:910-918. 10. Tbeorell T, Hamsten A, de Faire U, Orth-Gomer K, Perski A. Psychosocial work conditions before myocardial infarc- tion in young men. lntJ CardioL 1987;15: 33-46. 11. Johnson .IV, Hall EM. Class, work and health. In: Amick BC, Levine S, Tarlov AlL Chapman Walsh D, eds. Society and Health. New York, NY: Oxford University Press; 1995;247-271. 12. Salouen .IT. Is there a continuing need for longitudinal epidemiologic research? The Kuopio Ischaemic Heart Disease Risk Factor Study. Ann Clin Res. 1988;20: 13. Lakka TA, Salouen Jr. Physical activity and serara lipids: a cross-sectional popula- fiou study in eastern Finnish men. Am J EpidemioL 1992;136:806-818. 14. Rose GA, Blackburn H, Gillum RF, et al. Cardiovascular Survey Methods. Geneva. Switzerland: World Health Organization; 1982. 15. Lynch JW, Kaplan GA, Salonen R, Cohen RD, Tuomilehto J, Salonen Jr. Do cardio- vascular risk factors explain the relation between socioeconomic status, risk of all-cause mortality, cardiovascular mortal- ity and acute myocardial infarction? Am J Epidemiol. 1996;144:934-942. 16. Tuomilehio J, A~tila M, Kaarsalo E, et al. Acute myocardial infarction (AMI) in Finland---baseline data from the FIN- MONICAAMI register in 1983-1985. Eur Heart J. 1992;13:577-587. 17. Wilson "IN¢, Kaplan GA, Kauhanen J, et al. The association between plasma fibrinogen concentration and five socioeconomic indi- ces in the Kuopio Ischemic Heart Disease Risk Factor Study. Am J Epidemiol. 1993;137:292-300. 18. Salouen Jr, Salonen R, Sepp~en K, et al. HDL, HDL2 and HDL3 cholesterol subfrac- tions and the risk of acute myocardial infarction. A prospective population study in eastern Finnish men. Circulation. 1991; 84:129--139. 19. Salonen JT, Nyyss6nen K, KorpelaA, et al. High stored iron levels are associated with excess risk of myocardial infarction in eastern Fiani~h men. Circulatimt 1992;86: 803-811. 20. Salouen JT, Salonen R, Korpela A, et al. Serum copper and the risk of acute myocardial infarction: a prospective study in men in eastern Finland. Am J EpidemioL 1991;134:268-276. 21. Salonen J'T, Seppgnen K, Nyyss6nen K, et al. Intake of mercury from fish, lipid peroxidation, and the risk of myocardial infarction, and coronary, cardiovascular, and any death in eastern F'trmish men. Circulation~ 1995;91:645-655. 22. Lakka T, Ven~il~inen JM, Rauramaa IL et al. Relation of leisure-time physical activ- ity and cardiorespiratory fitness to the risk . of acute myocardial infarction in men. N EnglJ Med~ 1994;330:1549-1554. 23. Ihanalnen M, Salonen R, Seppanen R, et al. Nutrition data collection in the Kuopio Ischaemi¢ Heart Disease Risk Factor Study: nu~ent intake of middle-aged Finnish men. Nutr Res. 1989;9:597-604. 24. Everson SA, Goldberg DE, Kaplan GA, et al. Hopelessness and risk of mortality, myocardial infarction and cancer inci- dence. Psychosom Med. 1996;58:113-121. 25. Cox DR, Oakes D. Analysis of Survival Data. New York, NY: Chapman Hall; 1984. 26. SAS User's Guide: Statistics. Version 6.09. Cary, NC: SAS Institute Inc; 1990. 27. Siegrist J. Adverse health effects of high effort/low reward conditions. J Occup Health Psych. 1996;1:27-41. 28. Notkola V J, Husman KR, Laukkanen VJ. Momdity among male farmers in Finland during 1979-1983. Scand J Work Environ Health. 1987;13:!24-128. 29. Lynch JW, Kaplan GA, Salonen R, Cohen RD, Salonen JT. Socioeconomic status and earodd atheroselerosis. Circulation. 1995; 92:1786-1792. 30. Karasek R, Baker D, Marxer F, Ahlbom A, Tbeorell T. Job decision latitude, job demands, and cardiovascular disease: a prospective study of Swedish men. Am J Public Health. 1981 ;71:694-705. 31. Hansen EJ, Ringen S, Uusitalo H, Erikson 1L eds. Welfare Trends in the Scandinavian Countries. New York, NY: M. E. Sharpe; 1993. 32. Stamler J. Established major coronary risk factors. In: Marmot M, Elliot P, eds. Coronary Heart Disease Epidemiology. London, England: Oxford University Press; 1992:35-66. 33. Kaplan GA. Job strain and cardiovascular health. Invited address, Academy of Behav- ioral Medicine Research; June 5. 1994; Coopersto.wn, NY. 34. Muntaner C, O'Campo PJ. A critical appraisal of the demand/control model of the psychosocial work environment: episte- mologieal, social, behavioural and class considerations. Soc Sci Med. 1993;36: 1509-1517. 35. Siegrist J, Peter R, Junge A, Cremer P, Seidel D. Low status control, high effort at work and ischemic heart disease: prospec- tive evidence from blue-collar men. Soc Sci Mea~ 1990;31:1127-1134. 36. Marmot M, Theorell T. Social class and cardiovascular disease: the contribution of work. lnt J Health Serv. 1988;18:659-689. 37. Fenwick R, Tausig M. The macroeconomic context of job stress. J Health Soc Behav. 1994;35:266-282. 38. Lynch JW, Kaplan GA, Salonen .IT. Why do poor people behave poorly? Variations in adult health behaviours and psychoso- cial characteristics by stage of the socioeco- nomic life course. Soc Sci Med. 1997;44: 809-820. 39. Lynch JW, Kaplan GA, Cohen RD, et al. Childhood and adult socioeconomic status as predictors of mortality in Finland. lancet. 1994;343:524-527. 622 American Journal of Public Health April 1997, Vol. 87. No. 4