Immortal time bias in the association between retirement age and mortality

In their 2016 paper, Wu et al reported that each 1-year increase in retirement age was associated with lower all-cause mortality (HR 0.89, 95% CI 0.85 to 0.92), concluding that early retirement may be a risk factor for mortality.1 The published Cox specification classifies participants by a future retirement age while follow-up begins at the 1992 baseline, creating guarantee-time bias.^{2 3} A participant who retires at age 70 must have survived from baseline to that age, a span of 9–19 years depending on birth year. The Cox model counts this preretirement person-time as time at risk, mechanically biasing the HR downward. Even a 5-year contrast implies roughly 44% lower mortality (HR 0.89⁵=0.56), a large effect for a single behavioural choice in an observational study.

Using the same HRS initial cohort, I reconstructed a closely comparable sample from public RAND and HRS core files (N=3212 vs the…

Read the full article ›