NEW guidance details valuable approaches to analyze longitudinal patient-reported outcomes when many patients die during follow-up.
Why High Mortality Distorts Longitudinal Patient-Reported Outcomes
Analyzing longitudinal patient-reported outcomes can become misleading when a substantial proportion of participants die during follow-up. Many studies focus only on survivors, but that restriction can introduce selection bias, since the patients who remain alive may differ systematically from those who die. The result may be estimates that are harder to interpret clinically, particularly in older adult populations and serious illnesses where death is common.
The authors highlight that the same dataset can yield different, even conflicting, conclusions depending on how deaths are handled in analysis. This makes it essential to define the research question up front and select an approach that aligns with what clinicians and patients need to know.
Six Methods Compared in an Older Cancer Cohort
To illustrate the issue, the study draws on a prospective cohort of adults aged 70 years and older with cancer. Quality of life was measured using the EuroQoL 5-Dimension (EQ-5D) instrument at baseline, then again at 6 and 12 months after treatment initiation. Among 1,218 participants, 321 (26%) died within 12 months.
Across six analytic approaches, four methods estimated absolute changes in mean EQ-5D scores, while two methods focused on the proportion of patients whose quality of life changed. Approaches limited to the 897 survivors at 12 months produced results that the authors argue are difficult to interpret because they exclude a clinically meaningful outcome, death, in a high-risk population.
Choosing Death-Aware Estimands Clinicians Can Use
The paper also explains how linear mixed models may implicitly reconstruct EQ-5D scores for those who died, creating a hypothetical picture of quality of life as if nobody died. In contrast, alternative approaches such as a while-alive strategy, or composite endpoint strategies that treat death as the worst possible EQ-5D score, can provide a more complete view of the burden experienced during follow-up.
Ultimately, the authors conclude that longitudinal patient-reported outcomes are only as interpretable as the estimand they target, and that carefully matching the method to the clinical question is central to producing meaningful findings.
Reference: Baltussen JC et al. How to Analyze Longitudinal Patient-Reported Outcomes in Populations With High Mortality Rates. J Am Geriatr Soc. 2026;DOI: 10.1111/jgs.70327
