Eligible participants were aged 20 years or older. The observation window during which death was ascertained began immediately after individuals were surveyed in the in-home interview. Out of 43 793 respondents who participated in the in-home interview and physical examinations, we excluded 354 persons who died within 1 year after recruitment to reduce bias arising from benzodiazepine prescriptions for participants with terminal illness.17 We further excluded individuals who lacked data on any prescription medications (n = 24), mobile examinations (n = 2134), and study covariates (n = 3896) (Figure 1). We also excluded those without prescriptions for selective serotonin reuptake inhibitors (SSRIs), benzodiazepines, or narcotics (n = 32 173). Our final analytic sample of 5212 individuals was linked to the NDI, which contains data on date and underlying cause for more than 99% of all US deaths,26,27,28 with a sensitivity and specificity both estimated to be greater than 97%.29 Linked mortality data extend from January 1, 1999, until time of death or censorship at December 31, 2015, whichever occurred first. If no match was found with the NDI, participants were assumed to be alive as of that date.
