Document w6Kok86v0NemzRXBK5rD3kND

Standardized Mortality Ratios and the "Healthy Worker Effect": Scratching Beneath the Surface A. J. McMichael, M.D., Ph.D. J age-standardized mortality ratio (SMR) is a frequently ferent jobs. However, there may also be major differences in used summary index of mortality in occupational epidemiolog mortality experience between subgroups differentiated by ic studies. By expressing the observed mortality' experience of criteria other than work exposure. That is, the healthy worker the occupational study population relative to the mortality that effect may rtol apply equally throughout the study population. would otherwise by expected (in the absence of any mortality Therefore, if one attempts to improve the mpaningfulnp nf an hazard within that occupational environment), ar, evaluative SMR by adjusting for the healthy worker effect, allowance t measure of the force of mortality is obtained. must be made for variation in this effect between different age Certain basic limitations of the SMR have been discussed groups, different races, different work-status factive. inactive. elsewhere. Firstly, since the SMR computation involves an in retired, etc) groups, different periods of observation and dif direct age-adjustment procedure,1 hence SMR's calculated for ferent causes of death. Thus, the application of a constant ad two or more-study populations, although-us7ng'the same slan- justment factor (say .90, as recently suggested by Goldsmith*) Har3~popUtatibn 'dea~th~rates~ for calculating expected deaths, will, in some circumstances, produce misleading results. cannot themselves be directly compared to one another.? This article presents illustrative examples from a variety of Secondly, the acual numeric magnitude of the SMR depends occupational epidemiologic sludies, to demonstrate the directly on the choree ot standard population. Many factors variation in healthy worker effect among different subgroups other than the occupational environment influence mortality, of workers. In each of the foliowing examples, the standard and yet only one such factorjaibeil a major oneffscontrolled population, with whose mortality the mortality of the oc form the computation of the SMR, namely, aee. This situation cupational population is compared, is the national United is exemplified by the somewhat inappropriate, but convenient, States population. comparison of an industrial wording population with the general population. This latter population, which includes sick Discussion and disabled persons, is usually at greater mortality risk than a When a population of active and retired workers, embracing population of active workers healthy enough to have been a targe age-range, is followed for a number of years, the SMR is (and to remain) employable. If working in a safe environment, usually higher in the older age-bands. Figure 1 illustrates this such a population of active workers has been variously trend, showing the SMR's in four successive age-bands within estimated to have a mortality risk 60%-90% that of the general a population of 6678 male rubber workers, aged 40 or more population! This difference in mortality risk, due to selection |aa 1,1964 and followed for ten years. (The composition and forces, has been described as the "healthy worker effect".1 An follow-up of this study population have been described epidemiologic comparison of a working population with the elsewhere.*) general community must take account of this effect. The SMR for the full age-range studied, 40-84, is 98. For the A third limitation of the SMR is that it is a measure of the aclive employment age-range, 40-64, the SMR is 87; whereas average relative mortality of what is usually a heterogeneous for the older, post-retirement age-range, 65-84, it is T03. Within study population. As a summary index of population mortality the 40-64 age-range, the SMR increases from 81 (ages 40-54) to it necessarily glosses over differences in the mortality ex 89 (ages S5-64). Likewise, in the post-relirement years, the perience of various sub-groups within the population. Now, a SMR continues to increase, from 95 (ages 65-74) to 113 (ages primary purpose of occupational epidemiologic studies is to 75-84). compare the mortality experience of subgroups of workers ex The explanation for this gradient has several facets, partly posed to different physico-chemical agents, or working in dif- over-lapping. In terms of pathologic processes, the diseases O. McMtcturt h xwocijred with the Orprnmtvrt o# fpid^miotogy. and Occu- pjiwu! H?jhh $iude* Croup. School of Publ< Healih, of Koflh Qioitnj, Chapel Hill. NC 27S14. Piesemed ai the ftOh Annual Meting of ile American Occupational NWd<al A'^ociaiion. American Industrial Health Conference, San francnco. April 14-17, 197$. causing excess mortality in this population are mostly long term or chronic (e.g. cancers, chronic lung disease) occurring only after a sufficient period of exposure and/or time since first exposure and therefore most evident at older ages. In terms of the healthy worker effect, younger workers are Journal of Occupational Medicine/Vol. 18, No. 3/March 1976 752463 0037 165 SUR-98 L Ratio or g, wives lor aduarfally-Otilrwif population c< twr-disabW industrial retirees. 1968-72. to U. S. rales. 1968. L Erocdad doaths alculatolfrom 1968U.S. ralea9e-sptci(icdaath ntn within S-yairiqa-tandi. 2. Norralrrtirerant occurs at 09a 65. F|| 1 -- Fortier eWence of deditmy carry-mer of Uto kooltby worker effect la post retireoieot 0(e wife. 1 lit L -- Darfioa i* fcuKkt wortor offact with (ft, auk rubber werben. least distant from the time of initial selection (i.e. hiring on) and are therefore most likely to demonstrate the survival ad vantage resulting from their initial better-than-communityaverage health status. In addition, within the age-range 40-64, ' workers developing discomforting illnesses are most likely to voluntarily extend the selection process by leaving the in dustry, if still young enough to have the emotional.mental and financial independence necessary to learn a new job or trade. The SMR for the younger years, 40-S4, therefore remains low. However, older workers in this 40-64 age-range, if unhealthy, are more likely to either stay on in less demanding jobs (e.g. J janitoring) or take an early or disability retirement and thus) remain on the company's pension register. The SMR for ages/ 55-64, in Figure 1, is thus based on a mixture of active early or disabled retirees. Within the post-retirement age-rang^ 65-64, the "healthy worker" selection process no lot r operates actively. The residual survival advantage it cor s declines with age, as the retirees come to increasingly resem ble the general population in their health status characteristics. The healthy worker effect disappears by around age 75. Figure 2 presents the mortality experience of a population of actively employed male workers in the Bell System. Single-year age-specific mortality ratios are plotted for these active male employees, aged 40-64 (from data published previously!). The healthy worker effect is greatest at the youngest ages and declines steadily with age until eight years before normal ('retirement, when, presumably, the out-setection of the less ^healthy (via early retirement) causes the healthy worker effect (to increase among the remaining active workers. (This phenomenon of the increase in healthy worker effect in the later years of working, and the accompanying mortality ex perience of early and retired workers, is currently under in vestigation within the rubber industry.4) In similar vein. Figure 3 provides further evidence of the declining carry-over of the healthy worker effect in the postretiremem age-range. Life table (q) data for an actuarially l-ZSr c too 75 <E o 2 O -50 O lli aco. u o < 40 ____ g-ronCftlTKY WORKER EFFECT 45 50 55 AGE 60 64 l. Ratio at ti(-tibk qK vatun in 6*11 System rate emptoym. 19S9-61. to U. S. rates, 196a fit 1 -- Mortality tiprrirncr at < popalaUo* of wiplojed unit worltn. Fit < -- tkllthf worker efled it (reater for noovhitri tharr whiter, at all ajev 166 Standardized Mortality Ratios and the "Healthy Worker Effect'TMcMidiael fi( 7. -- Three different "directions'' that cm be foHotred in cakulalint SMR's. CraphtBMrfontabuhrrtatafrwaEhttrlitM, P.E. (I965L fit S----- Ottfiiw ia tht hulthy Mrittr effect with puu{i of time efter Mitral idem tifkitiM ef i cobart of active wmfien (nbestos products). defined population of nondisabled male, industrial retirees, 1968-72,7 indicate an initial mortality of only 55% of the general population. This effect declines steadily with age such that, at age 84, the mortality is %% of the general population. figure 4 shows that, within the same rubber worker population of figure 1, the healthy worker effect is greater for nonwhites than whites, at all ages. In particular, nonwhite male workers in the active employment age-range 40-84 are at a substantial survival advantage compared to the national nonwhite male population (many ofjwhom are unemployed, indigent, retired due to poor health, or working in hazardous occupations.) That is, the quality of the comparison differs for the two race groups. Qn the one hand, working-class whites are being compared to the national average "middle-class" white population; on the other hand, working-class nonwhites AGE Source: Graph tesed on tabular data from Milham. S. (19741. fit *-----Cofii^trabta diffrnnea can occur In thn healthy oorter effect for different ainea of death (carpenters and joiners!Journal of Occupational Meditine/Vol. 18. No. 3/March 1976 are being compared to the national average "lower-class** non while population. The assessment of race-specific healthy worker effects is therefore confounded by class differences. pearly, the greater the proportion of nonwhites in a working population, the lower the race-standardized 5MK's will tend to. be and the greater the risk of not perceiving an actual, but small or moderate, excess of jnpxtality. figure 5 illustrates the decline in the healthy worker effect with the passage of time after the identification of a cohort of already active workers. The graphs are based on tabular data from a cohort mortality study of asbestos products workers, by Enterline.* When follow-up is achieved of a total cohort, in cluding those that quit or retire early for health reasons, then the initial healthy worker effect associated with active em ployment declines with time, because of the absence of any continued selection process. For total mortality, the effect in this cohort disappeared after five years largely due to the in crease in cancer mortality in the second and third quinquen nial period of follow-up. figure 6 illustrates the considerable difference that can occur in the healthy worker effect for different causes of death. The graphs are based on tabular data from a cross-sectional study of mortality in carpenters and joiners, by Milham.* At all ages, there is an obvious mortality deficit of about 25% for ischemic heart disease, whereas for cancers the mortality ratio remains in the vicinity of 100 throughout Although part of this higher figure for cancer may reflect excess deaths from certain specific cancers for which Milham reports that these workers are at in creased risk of death, part of the explanation lies in the causespecific variation in the healihy worker effect. Where cancer is a "silent" disease, with long-deferred clinical manifestations, ischemic heart disease is a chronic and readily detectable con dition. Heart disease is therefore much more likely to be select ed against than is cancer, in the recruitment and retention of an active workforce. Finally, Figure 7 presents, diagrammatically, three different "directions" that can be followed in calculating SMR's. A "slab" of recorded mortality experience within an oc cupational population (see hatched area. Fig 7) may result in an overall SMR of, say, 85. However, by partitioning the total mortality experience in different directions, significant ex cursions above and below that figure may be detected. SMR's 752463 0039 167 can be calculated by age at death, by successive cohorts (year of birth, or year of hire cohorts), or by year of death. The choice of one or more of these approaches depends upon one's anticipations of the data. An example of variation in SMR's by age at death has been given in Figure 1. An example of mortality differences between year-of-hire cohorts is to be found in a recent study of bladder cancer deaths, during 1967-71, in a population of 16,035 British rubber workers.10 Workers joining the industry before 1950 had a bladder cancer SMR of 130, while for those joining in or after 1950 the SMR was 97. The major presumed bladder carcinogen, beta-naphthylamine, was removed from the British rubber industry in 1949. The third approach, using year-ofdeath SMR's (i.e. period mortality), would be useful in iden tifying temporary mortality excesses resulting from some short term exposure to an agent whose use was subsequently discontinued (for reasons, say, unrelated to health). For example, a chemical with potent respiratory sensitization or irritation properties might precipitate excess deaths among workers with chronic respiratory ailments. Mortality analysis by year-of-death SMR's would best identify this problem. Summary The age-standardized mortality ratio (SMR) is a relative in dex of mortality, expressing the mortality experience of the study population relative to that of a comparison ("standard") population. With the general population as the "standard", the SMR for an occupational population will underestimate the mortality experience of that latter population (since it com prises individuals necessarily healthy enough to be employable -- and whose mortality risk is therefore initially lower than the general population average). However, this "healthy worker effect" does not apply equally to all groups within the study population. Therefore, if one attempts to adjust (or this effect the summary nature of the SMR must be recognized, and allowance must be made for variation in the healthy worker ef fect between different age groups, different races, different work-status groups, different causes of death, and different elapsed-time periods of observation. References 1. Bradford Hill A: Principles of .Medical Statistics: 215-217. Oxford University Press. New York, 1971. 2 Miertineo OS: Standardization of risk ratios. Amer / Cpid 96:383- 388, 1972. 3. McMichael A|, Haynes SC. Tyroler HA: Observations on the evaluation of occupational mortality data. JOM 17:128-131, 1975. 4. Goldsmith F What we expect from an occupational cohort? KM 17:126-127, 1975 5. McMichael A). Spiftas R, Kupper ll: An epidemiologic study of mortality within a cohort of rubber workers. 1964-72. JO.V416:458-464 1974. 6 Andjelkovic DA et al. Unpublished data on differential mortality experience of active, early retirees and disability retirees. Paper in preparation. 7. Toussaint RC Driscoll FT: Note on a New Mortality Table for use in pension plans. Unpublished data from the George 8. Buck Consult ing Actuaries, 1974. 8. Enleriine PE: Mortality among asbestos products workers in the United States. Ann NY Acad Sciences 132:156-165, 1965. 9. Milham S: Mortality Experience of the AFl-CIO United Brother hood of Carpenters and Joiners of America, 1969-70. DHEW Publica tion No. (NIOSH) 74-129. 1974. 10. Fox At Lindars DC, Owen R: A survey of occupational cancer in the rubber and cablemaking industries: results of five-year analysis, 1967-71. Br / Indust Med 31:140-151, 1974. 168 Standardized Mortality Ratios and the "Healthy Worker EffecC/McMichael 752463 0040