Evaluation of Nonresponse Bias in a Case–Control Study of Pleural Mesothelioma

Abstract

Nonparticipation limits the power of epidemiological studies, and can cause bias. In a case–control study on pleural malignant mesothelioma (MM), we found low participation in interviews (63%) among controls. Our goal was to characterize nonresponder controls and assess nonresponse bias in our study. We selected all nonresponder controls (204) and a random sample of responder controls (174). Data were obtained linking hospital admissions and town registrars, and concordance between sources was assessed. Nonresponse bias was evaluated using a logistic regression model applying the inverse probability weighting approach. The odds ratio (OR) for the status of the respondents was 0.61 (95% confidence interval (CI): 0.33–1.16) for controls aged 61–70, 0.37 (CI: 0.20–0.66) for those aged 71–80, and 0.40 (CI: 0.20–0.80) for those aged above 80 (reference group: ≤60 years). Controls with low education level had lower OR (0.47; CI: 0.26–0.84). After adjustment, the ORs for MM by categories of cumulative exposure to asbestos were similar to the unadjusted results, ranging from 4.6 (CI: 1.8–11.7) for cumulative exposures between 0.1 and 1 f/mL-y to 57.5 (CI: 20.2–163.9) above 10 f/mL-y. Responder controls were younger and had higher education level. Nevertheless, there was little evidence of bias from nonresponse in the risk estimates of MM.