Cohort studies were found to be frequently biased by missing disease information due to death

Abstract

Objectives: In epidemiologic cohort studies with missing disease information due to death (MDID), conventional analyses right-censoring death cases at the last observation or at death may yield significant bias in relative risk and hazard ratio estimates. The aim of this study was to investigate susceptibility to this bias and assess its potential direction and magnitude.

Study design and setting: Literature review of selected epidemiologic, geriatric, and environmental journals in 2011-2012 and simulation study of various conventional approaches to handling missing disease data. A study was considered susceptible to MDID bias if disease information was collected at follow-up visits only, and a conventional analysis was performed on the data.

Results: Of 125 identified studies, 58 (46.4%, 95% confidence interval [CI]: 37.7-55.1%) were classified as susceptible to MDID bias, of which six (10.3%, 95% CI: 2.5-18.2%) attempted to address this in sensitivity analyses. The simulation revealed that depending on the analytic strategy for handling missing disease data, the potential exists for significant under- or over-estimation of risk factor effect estimates.

Conclusion: Awareness of MDID bias is important as more adequate analysis methods exist permitting an unbiased analysis. Recommendations for better reporting and analysis of MDID are provided.

Keywords: Cohort studies; Epidemiological biases; Illness-death model; Missing disease information due to death; Regression models; Time-to-event.