Methods/Statistics
Selection bias due to selective survival before study enrollment on the association between relative age and suicide Yoshikazu Komura* Yoshikazu Komura Komura Komura Komura Komura Department of Social Epidemiology, Graduate School of Medicine and School of Public Health, Kyoto University, Kyoto, Japan
Relative age, the age difference between individuals within a school grade, is an early-life determinant of long-term mental health, including suicidality. Although research on early-life factors often suffers from selection bias due to selective survival before study enrollment, it remains unclear how this bias affects estimates. We quantified this selection bias on the association between relative age and suicide, using vital statistics in Japan. This cohort study included individuals born in 1974–1983. Relative age was defined by birth date (younger [born Oct 1–Apr 1] or older [Apr 2–Sep 30]). Death by suicide was identified from death records. Characteristics at birth, such as birth year and household occupation, were collected from birth records. Using the Aalen–Johansen estimator, we estimated the risk of suicide at age 40 among those with younger and older relative ages, and calculated the risk difference (RD) and risk ratio (RR). To quantify the selection bias, we estimated these effects by restricting analytical samples to survivors at four different enrollment points (age 0 [target estimate], 10, 20, and 30), and calculated % bias, defined as the deviation of each estimate from the target estimate. In a cohort of over 17 million individuals with 85 372 deaths by suicide, we found no difference in characteristics at birth by relative age. Individuals with younger relative age had a higher suicide risk at age 40 (RD per 10 000: 2.08 [95% CI, 1.47–2.68]; RR: 1.05 [95% CI, 1.04–1.07]. The RD was underestimated among survivors at study enrollment points (% bias: 1.2%, −16.8%, and −54.3% at ages 10, 20, and 30). The RR was also underestimated but the degree of the bias was much smaller (% bias: 0.0%, −0.6%, and −0.4% at ages 10, 20, and 30). Future research on suicide related to relative age should account for selection bias due to selective survival before study enrollment to avoid overlooking the impact of relative age, particularly when relying on difference scale.