Potential bias in self-controlled case series by incident vs. prevalent exposures/outcomes
Addressing bias in incident- and prevalent-user designs within self-controlled case studies (SCCS) remains unexplored. This study explores potential bias and its impact on study outcomes by varying the definitions of exposures and outcomes in SCCS studies. Using the National Health Insurance Service (NHIS) database from 2002 to 2015, we conducted two SCCS studies to examine the risk of retinal detachment following fluoroquinolone and the risk of cardiovascular disease following ranibizumab. Incident and prevalent exposures/outcomes were defined with wash-out periods. Each study included four distinct exposure-outcome scenarios: 1) incident exposure and incident outcome, 2) incident exposure and prevalent outcome, 3) prevalent exposure and incident outcome, and 4) prevalent exposure and prevalent outcome. We used conditional Poisson regression to estimate the incident rate ratio (IRR) and 95% confidence intervals (CIs). Subgroup analyses were conducted by underlying characteristics, and sensitivity analyses were performed by varying wash-out periods and exposure. The first study revealed a significant association between fluoroquinolone use and retinal detachment in all scenarios. For the second study, prevalent exposure showed inconclusive results, while incident exposure to ranibizumab significantly decreased the risk. This emphasizes the importance of precisely defining exposures and outcomes in alignment with study objectives.
2024 Spring Convention