There is substantial variation and suboptimal methodological choices in how contemporary ICU trials handle and report patient-important outcomes other than mortality.
OBJECTIVES: Randomized clinical trials (RCTs) conducted in adult ICU patients increasingly include patient-important outcomes other than mortality. This comes with challenges regarding outcome choices/definitions, handling of deceased patients and missing data in analyses, and choices of effect measures and statistical methods due to complex distributions. This scoping review aimed to characterize how these challenges are handled in relevant contemporary RCTs. DATA SOURCES: We systematically searched 10 selected journals for RCTs conducted primarily in adult ICU patients published between 1 January 2018 and 5 May 2022 reporting at least one patient-important outcome other than mortality, including "days alive without"…-type outcomes, functional/cognitive/neurologic outcomes, health-related quality of life (HRQoL) outcomes, and ordinal/other outcomes. STUDY SELECTION: Abstracts and full-texts were assessed independently and in duplicate by two reviewers. DATA EXTRACTION: Data were extracted independently and in duplicate by two reviewers using predefined and pilot-tested extraction forms and subsequently categorized to facilitate analysis. DATA SYNTHESIS: We included 687 outcomes from 167 RCTs, with 32% of RCTs using a patient-important outcome other than mortality as a (co-)primary outcome, most frequently "days alive without"…-type outcomes. Many different functional/cognitive/neurologic (103) and HRQoL (29) outcomes were reported. Handling of deceased patients varied, with analyses frequently restricted to survivors only for functional/cognitive/neurologic (62%) and HRQoL (89%) outcomes. Follow-up was generally longer and missing data proportions higher for functional/cognitive/neurologic and HRQoL outcomes. Most outcomes were analyzed using nonparametric tests (31%), linear regression/ t tests (27%), chi-square-like tests (12%), and proportional odds logistic regression (9%), often without presentation of actual treatment effects estimates (38%). CONCLUSIONS: In this sample of RCTs, substantial variation in practice and suboptimal methodological choices were observed. This calls for increased focus on standardizing outcome choices and definitions, adequate handling of missing data and deceased patients in analyses, and use of statistical methods quantifying effect sizes.
Granholm et al. (Wed,) studied this question.