Influence of Incomplete Death Information on Cumulative Risk Estimates in United States Claims Data

Background: Administrative claims databases often do not capture date of death, so studies using these data may inappropriately treat death as a censoring event—equivalent to other withdrawal reasons —rather than a competing event. Understanding the influence of not distinguishing between disenrollment and death on cumulative risk estimates is necessary to inform real-world data (RWD) suitability.

Objectives: To investigate the influence of specifying death as a censoring event versus competing event on the validity of cumulative risk (CR) estimates.

Methods: We examined 1-, 3-, and 5-year CR of a composite outcome (myocardial infarction, stroke, and hospitalization for congestive heart failure) among initiators of antihypertensive medications telmisartan (exposure) and ramipril (referent) ages ≥55 in Optum claims (which reliably captures death) from 2003 to 2020. The cohort was created according to eligibility criteria of a published RWD emulation of a randomized clinical trial (NCT00153101). Inverse probability of treatment weighting was used to adjust for patient characteristics. Censoring occurred upon disenrollment or end of study. We compared CR estimates from models where death was treated as a censoring event (cause-specific risk [misspecified]) versus competing event (subdistribution risk [truth]). We examined whether error (difference between two estimates) depended on mortality rate.

Results: 34,197 patients were included (7,036 telmisartan, 27,161 ramipril). 4,664 events occurred over 86,616 person-years. Mortality rates per 1,000 person-years were 9.8 for ages 55-64, 25.3 for ages 65-74, and 79.1 for ages ≥75. CR error increased over time, as event and mortality risks increased. Results were similar for both exposure groups. For ramipril users (selected results), true and misspecified CR estimates, respectively, were 14.9 (95% CI 14.4, 15.5) and 15.3 (14.8, 15.9) at year 3 (error=0.4) and were 21.8 (21.1, 22.5) and 22.8 (22.0, 23.5) at year 5 (error=1.0). The increase in CR error over time was greatest for the oldest age group: among ramipril users, 5-year true and misspecified CR estimates, respectively, were 13.5 (12.5, 14.5) and 13.6 (12.6, 14.7) among ages 55-64 (error=0.1) and were 33.3 (31.8, 34.8) and 37.0 (35.3, 38.8) among ages ≥75 (error=3.7).

Conclusions: CR errors due to inappropriate censoring of death increased with greater follow-up and increasing age, where event and mortality risks were higher. We suggest researchers consider baseline cohort mortality risk associated with treatment indication when deciding whether RWD with incomplete death data can be used without concern. Future research will explore the influence of informative censoring models on these results.