Attributable risk: advantages of a broad definition of exposure

Our plain-English summary of Attributable risk: advantages of a broad definition of exposure, by S. Wacholder, J. Benichou, E.F. Heineman, et al (American Journal of Epidemiology, 1994).

The study

• Authors: S. Wacholder, J. Benichou, E.F. Heineman, P. Hartge, R.N. Hoover
• Journal: American Journal of Epidemiology
• Published: 1994;140(4):303-9
• Cited by: 146 articles, as counted by NIH iCite
• Read it: PubMed · Publisher

Abstract

Classification of exposure into two levels--one consisting exclusively of unexposed individuals and the other consisting of exposed and perhaps unexposed ones--yields an unbiased estimate of attributable risk when misclassification is nondifferential. The authors advocate, therefore, the use of a broad definition of exposure when estimating attributable risk. Based on this idea, they justify a simple and robust method for estimating the overall attributable risk from several exposures that is based on a division of subjects into two groups, a baseline consisting of those unexposed to all exposures and everyone else.

In plain English

This was a methods commentary, not a new asbestos study. The authors examined how defining exposure affects attributable risk, the fraction of cases avoidable had no one been exposed. They used a subset of data from an unpublished mesothelioma case-control study by Spirtas and colleagues. Exposure likelihood fell into four categories, mainly from work history; the moderately low category also included household contacts, people living near an asbestos mine or mill, and people claiming exposure. Counting the three upper categories as exposed gave an odds ratio of 8.2 and attributable risk of 0.82. Narrower definitions lowered it to 0.63 or 0.57. Moving 20 to 60 percent of the lowest group into the exposed group, equally for cases and controls, kept it at 0.82 but raised its standard error. Adjusting for family history of cancer changed estimates little. The authors concluded that the broadest definition gives the best estimate, provided all exposed people count as exposed, the unexposed group holds no one above baseline risk, and misclassification is equal for cases and controls. They noted cases' next-of-kin may have exaggerated exposure.

We wrote this plain-English summary from the full article with the help of AI, and a second AI model checked each sentence against the article. It is not the authors' wording; the abstract and the article are their own account.

What the study did and found

• Design: Methodological review and commentary, illustrated with a subset of data from an unpublished case-control study of mesothelioma
• Who was studied: A subset of cases and controls from an unpublished case-control study of malignant mesothelioma by R. Spirtas, E. F. Heineman, L. Bernstein and others. Table 1 lists 79 cases and 143 controls with a family history of cancer, and 104 cases and 280 controls without. Most participants were employed in the 1940s through the 1960s. The location is not stated. The authors said the data were used only to make a methodological point.
• Exposure: Likelihood of asbestos exposure, assessed mainly from complete work histories compared with the National Occupational Hazard Survey. There were four categories: extremely low (no known exposure); moderately low; medium; and high (insulators, shipyard workers, furnace or boiler installers and repairers). The moderately low category also included people who lived with asbestos-exposed workers, lived near an asbestos mine or mill, or claimed exposure, even if their job history gave no evidence of exposure. Fiber types were not specified. Family history of cancer was also examined.
• Main results:
  • Compared with the extremely low category, the crude odds ratios were 3.3 (moderately low), 6.3 (medium) and 13.8 (high). The level-specific attributable risks were 0.13, 0.06 and 0.64.
  • Counting categories 1 to 3 as exposed against category 0 gave an attributable risk of 0.82 (odds ratio 8.2). This equals the sum of the three level-specific attributable risks.
  • Narrower definitions lowered the estimates. Counting only categories 2 and 3 as exposed gave an attributable risk of 0.63 (odds ratio 6.0). Counting only category 3 as exposed gave 0.57 (odds ratio 5.7).
  • The authors moved 20, 40 and 60 percent of the baseline group into the exposed group, equally for cases and controls. The odds ratio fell from 8.2 to 7.5, 6.9 and 6.5. The attributable risk stayed at 0.82. Its standard error rose from 0.053 to 0.059, 0.069 and 0.085.
  • Adjusting for family history of cancer in a model without interactions gave a summed level-specific attributable risk of 81.91 percent, and 82.09 percent for exposure counted as yes or no. With interactions included, both were 82.33 percent.
  • For asbestos exposure or family history of cancer combined, compared with people who had neither, the odds ratio was 7.5 and the joint attributable risk was 82.9 percent. The adjusted attributable risk for occupational history alone was 82.1 percent, and for family history it was 12.3 percent.
• The authors' conclusion: The authors concluded that the broadest definition of exposure gives the best estimate of attributable risk. This holds when all exposed people are classified as exposed, everyone in the unexposed group has no more risk than the truly unexposed, and misclassification of unexposed people is the same for cases and controls. Wrongly counting unexposed people as exposed does not change the estimate but reduces its precision. Leaving truly exposed people out of the exposed group lowers the estimate.
• Limitations the authors state:
  • Counting unexposed subjects as exposed reduces the precision of the attributable risk estimate.
  • When misclassification rates differ greatly between cases and controls, neither a broad nor a restrictive definition of exposure solves the problem.
  • Misclassification may have differed between cases and controls in the Spirtas study, because next-of-kin of some cases may have exaggerated exposure. That would bias the attributable risk and the relative risk upward.
  • In models adjusted for other risk factors without interaction terms, the property behind the argument may not hold exactly.
  • The joint attributable risk from several common exposures may be imprecise if the baseline group is small.
  • Attributable risk depends on how common exposure is in the population studied, so it is likely to vary between communities.

Topics

The National Library of Medicine indexes this study under subjects our site covers: Mesothelioma.

Medical terms in this study, explained in our Lexicon: Attributable Risk.

Related studies

• Malignant mesothelioma: attributable risk of asbestos exposure — R. Spirtas, E.F. Heineman, et al. (1994)
• Occupational risk factors have to be considered in the definition of high-risk lung cancer populations — P. Wild, M. Gonzalez, et al. (2012)
• Occupational and non-occupational attributable risk of asbestos exposure for malignant pleural mesothelioma — A. Lacourt, C. Gramond, et al. (2014)
• Attributable risk in men in two French case-control studies on mesothelioma and asbestos — A. Lacourt, P. Rolland, et al. (2010)
• Hierarchical regression for multiple comparisons in a case-control study of occupational risks for lung cancer — M. Corbin, L. Richiardi, et al. (2012)
• Apportioning causes, targeting populations and predicting risks: population attributable fractions — N. Chatterjee, P. Hartge (2003)

Browse more research in American Journal of Epidemiology, or return to medical research.