Design and conduct of occupational epidemiology studies: IV. The analysis of case-control data

Our plain-English summary of Design and conduct of occupational epidemiology studies: IV. The analysis of case-control data, by N. Pearce, H. Checkoway, J. Dement (American Journal of Industrial Medicine, 1989).

The study

• Authors: N. Pearce, H. Checkoway, J. Dement
• Journal: American Journal of Industrial Medicine
• Published: 1989;15(4):403-16
• Type of article: Review
• Cited by: 11 articles, as counted by NIH iCite
• Read it: PubMed · Publisher

Abstract

This paper reviews the basic methods of analysis of data from case-control studies. The standard analytic methods are outlined first for a single stratum. The discussion is then extended to stratified analysis, multiple exposure levels, and analyses allowing for disease induction and latency periods. Finally, logistic regression is discussed as an extension of the more basic forms of analysis. The methods are illustrated with data from a study of lung cancer among asbestos textile plant workers.

In plain English

This paper was a methods review, not a new study of asbestos. The authors described methods for analysing case-control studies, from Mantel-Haenszel techniques to logistic regression. To illustrate them, they reanalysed data on 1,261 white male asbestos textile plant workers (Dement et al., 1983), comparing 35 lung cancer deaths with 140 age-matched controls. Twenty of 35 cases and 48 of 140 controls had cumulative exposure of at least 10,000 fibers/cc × days. The crude odds ratio was 2.56, the age-adjusted 2.38 and the matched 2.60. The text gives 95% limits of 1.25–5.40, 1.19–4.74 and 1.23–5.59; Table VII prints slightly different limits. The authors said this suggested little confounding by age. Compared with the lowest exposure group, crude odds ratios for the four higher groups were 2.28 (0.73–7.08), 2.45 (0.72–8.29), 6.74 (2.20–20.68) and 4.90 (0.81–29.72). The authors called the higher-category findings unstable due to small numbers. Logistic regression gave similar estimates except in the highest category. The authors concluded that most case-control studies can be analysed with Mantel-Haenszel methods. They added that logistic regression may be indicated when there are more than two confounders and problems with small numbers.

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: Methods review of case-control data analysis, illustrated with a nested case-control reanalysis of data from an earlier cohort study
• Who was studied: Illustrative example: a cohort of 1,261 white male asbestos workers from one textile manufacturing plant, employed at least one month in textile operations between 1 January 1940 and 31 December 1975. The case-control set was 35 lung cancer deaths and 140 controls (4 per case), matched by incidence density on age at risk. Data were derived from Dement et al. (1983).
• Exposure: Cumulative asbestos exposure in fibers/cc × days, taken from job history and industrial hygiene records. It was analysed first as a two-level split (above or below 10,000 fibers/cc × days) and then in five levels: under 1, 1–9, 10–39, 40–99 and 100 or more thousand fibers/cc × days. The article text names the fiber only as asbestos; the cited source study's title refers to chrysotile asbestos workers.
• Main results:
  • Two-level analysis (10,000 fibers/cc × days or more versus less): 20 of 35 cases and 48 of 140 controls were in the higher group. The crude odds ratio was 2.56, with approximate 95% limits of 1.25 and 5.40 given in the text (Table VII prints 1.21–5.40).
  • The age-stratified (adjusted) Mantel-Haenszel odds ratio was 2.38, with approximate 95% limits of 1.19–4.74 given in the text (Table VII prints 1.20–4.14).
  • The matched analysis gave an odds ratio of 2.60, with a test-based confidence interval of 1.23–5.59 given in the text (Table VII prints 1.22–5.55). The authors said the crude, age-adjusted and matched estimates were very similar, suggesting little confounding by age.
  • Multiple exposure levels, compared with under 1,000 fibers/cc × days (Table X), gave these crude odds ratios: 1–9 thousand, 2.28 (95% limits 0.73–7.08); 10–39 thousand, 2.45 (0.72–8.29); 40–99 thousand, 6.74 (2.20–20.68); 100 thousand or more, 4.90 (0.81–29.72).
  • Conditional and unconditional logistic regression gave estimates very similar to the crude and stratified analyses, except in the highest exposure category, which had very small numbers.
  • The authors stated that the case-control approach (175 subjects) gave effect estimates very similar to those of a full cohort analysis of 1,261 subjects.
• The authors' conclusion: The authors concluded that most case-control studies can be analysed in depth with forms of the Mantel-Haenszel procedure and Miettinen's test-based confidence limits. They added that logistic regression may be indicated when there are more than two confounders and problems with small numbers arise.
• Limitations the authors state:
  • Case-control findings may be less stable than a full cohort analysis for particular subgroups.
  • Findings for the higher exposure categories were unstable because of small numbers.
  • The case-control analysis was done for illustration only; exposure data had already been assembled for the whole cohort.

Topics

Medical terms in this study, explained in our Lexicon: Epidemiology, Latency, Lung Cancer.

Related studies

• The potential impact of epidemiology on the prevention of occupational disease — D.H. Wegman (1992)
• Hierarchical regression for multiple comparisons in a case-control study of occupational risks for lung cancer — M. Corbin, L. Richiardi, et al. (2012)
• Applying quality criteria to exposure in asbestos epidemiology increases the estimated risk — A. Burdorf, D. Heederik (2011)
• Lagging exposure information in cumulative exposure-response analyses — D.B. Richardson, S.R. Cole, et al. (2011)
• Prospective Registry On Mesothelioma Peritonei Treatment (PROMPT): study design and rationale — C. Cao, T.D. Yan, et al. (2012)
• Extrapolation from occupational studies: a substitute for environmental epidemiology — P.E. Enterline (1981)

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