Abstract Introduction Conventional endpoints in severe pneumonia trials are inefficient, prone to bias, and fail to capture the multidimensional experience of critically ill patients. Hierarchical composite endpoints (HCEs) have emerged as a more holistic alternative, yet their construction and analysis remain inconsistent, with limited validation and comparative assessment in real-world settings. This study aimed to derive novel pneumonia-specific HCEs grounded in clinician preferences and evaluate their validity and interpretability through secondary application to pneumonia trial data. Methods Using clinician preference data from the CHOIR international survey, we derived HCEs integrating death and key nonfatal outcomes—ventilation duration, discharge disposition, antibiotic toxicity, clinical cure, and pneumonia recurrence. We created two Desirability of Outcome Ranking (DOOR) scores: a standard DOOR summing unweighted nonfatal events on an ordinal scale, with higher scores indicating worse outcomes, and a Partial Credit DOOR (PC-DOOR) applying clinician-derived weights to reflect each outcome’s relative importance. Separately, we employed a Win Ratio framework comparing patient pairs across the same hierarchy, enabling inclusion of continuous and time-to-event endpoints. HCEs were applied in a secondary analysis of pneumonia patients from the OVERCOME randomized trial comparing colistin monotherapy versus colistin + carbapenem for extensively drug-resistant gram-negative infections. Novel HCEs were compared with the trial’s primary (28-day mortality) and secondary (clinical failure) endpoints using Van Elteren tests for DOOR and the WINS package for WR, stratified by APACHE score. Probabilistic indices (PIs) were derived for each HCE to enable standardized comparison. Results Among 298 subjects, 152 received colistin monotherapy and 146 combination therapy. Baseline characteristics were evenly distributed between treatment arms. 28-day mortality (45% vs 40%, p = 0.41) and clinical failure (73% vs 65%, p = 0.19) did not significantly differ between treatment arms. Mean (SD) DOOR scores were 3.97 (2.06) vs 3.71 (2.15) and PC-DOOR scores 0.73 (0.31) vs 0.69 (0.33) for monotherapy vs combination therapy (p = 0.25 and 0.29, respectively). In WR analysis, the WR was 1.26 (95% CI 1.01-1.58; p = 0.04), favoring the combination arm. Visualizations of DOOR distributions, WR proportions and HCE-derived PIs are shown in Figure 1. Conclusions DOOR visualizations suggested possible treatment differences between arms that did not reach statistical significance, whereas the Win Ratio detected a significant benefit for combination therapy. These findings suggest that HCEs may not only provide complementary insights to conventional endpoints but also change pneumonia trial interpretation. Future work will explore HCE performance and power under diverse trial scenarios. This abstract is funded by: NIH NIAID K23 AI177689
Albin et al. (Fri,) studied this question.