Real-World Effectiveness: Addressing Selection Bias in the Well Elderly 2 Study Strengthens Effect Estimates

Rationale: Intention-to-treat (ITT) analyses in randomized trials may underestimate effects due to poor adherence or follow-up loss. Objectives: To demonstrate methods to address internal validity threats in ITT using the Well Elderly 2 study as a case example. Methodology: This secondary analysis used 4-year data from 460 older adults. The original trial, by a separate team, included a crossover intervention with 6-month follow-ups. We addressed follow-up loss using inverse probability of censoring weights to report an adjusted ITT. Results: ITT showed improvement in 10 of 17 outcomes; inverse probability of censoring weight (IPCW)-adjusted ITT in 13. In SF-36v2, ITT improved mental (B = 0.54, 99% confidence interval (CI) [0.05, 0.60]) but not physical (B = 0.71, 99% CI [-0.21, 0.08]); IPCW-adjusted ITT improved both mental (B = 0.77, 99% CI [0.34, 1.19]) and physical composites (B = 2.28, 99% CI [0.14, 0.71]). Conclusion: IPCW-adjusted ITT showed stronger effects but may be biased, while ITT was more conservative yet prone to misclassification; reporting both provides a clearer interpretation.