Prioritizing attributes of approaches to analyzing patient-centered outcomes that are truncated due to death in...

A key challenge for many critical care clinical trials is that some patients will die before their outcome is fully measured. This is referred to as "truncation due to death" and must be accounted for in both the treatment effect definition (i.e. the estimand), as well as the statistical analysis approach. It is unknown which analytic approaches to this challenge are most relevant to stakeholders.

Using a modified Delphi process, we sought to identify critical attributes of analytic methods used to account for truncation due to death in critical care clinical trials. The Delphi panel included stakeholders with diverse professional or personal experience in critical care-focused clinical trials. The research team generated an initial list of attributes and associated definitions. The attribute list and definitions were refined through two Delphi rounds. Panelists ranked and scored attributes and provided open-ended rationales for responses. A consensus threshold was set as ≥ 70% of respondents rating an attribute as "Critical" (i.e., score ≥ 7 on a 9-point Likert scale) and ≤ 15% of respondents rating the measure as "Not Important" (i.e., a score of ≤ 3).

Thirty-one (91%) of 34 invited individuals participated in one or both rounds. The response rate was 82% in Round 1 and 85% in Round 2. Participants included eight (26%) personal experience experts and 26 (84%) professional experience experts. After two Delphi rounds, four attributes met the criteria for consensus: accuracy (the approach will identify effects if they exist, but will not if they do not), interpretability (the approach enables a straightforward interpretation of the intervention's effect), clinical relevance (the approach can directly inform patient care), and patient-centeredness (the approach is relevant to patients and/or their families). Attributes that did not meet the consensus threshold included sensitivity, comparability, familiarity, mechanistic plausibility, and statistical simplicity.

We found that methods used to account for truncation due to death in the treatment effect definition and statistical approach in critical care trials should meet at least four defined criteria: accuracy, interpretability, clinical relevance, and patient-centeredness. Future work is needed to derive objective criteria to quantify how well existing estimands and analytic approaches encompass these attributes.