To explain why standardized HAP/VAP response criteria do not, by themselves, define the treatment effect estimated in randomized trials, particularly when death occurs before response assessment.
Approach:
Treatment estimand: The authors discuss how death before day 10 and other events that prevent response assessment should be addressed when specifying the target population, outcome, and treatment-effect summary.
Proposed outcome frameworks: They describe incorporating response criteria into a global outcome or DOOR framework that includes mortality and other patient-centered outcomes.
Validation analyses: They distinguish prognostic associations among assessable patients from causal treatment comparisons in randomized trials.
Key Findings:
Of 1,603 pneumonia episodes, 302 had an indeterminate overall treatment outcome; 240 involved death. Classifying death as failure increased the failure proportion from 22.2% to 34.3%.
Comparing only patients alive and classifiable on day 10 can compare treatment-specific survivor groups; if treatment affects survival, this may not preserve the randomized causal contrast.
Associations between day-10 failure and later outcomes support prognostic discrimination among assessable patients, but do not establish the causal interpretation of survivor-conditioned treatment comparisons.
A review of 255 randomized trials found that the primary estimand could be determined in 46%, and intercurrent-event handling in 49%.
Interpretation:
The authors argue that trials using these response criteria should prespecify the estimand and how death and other events precluding assessment will be handled. They do not advocate one mandatory approach to death.
Limitations:
The commentary does not report new empirical analyses.
A principal-stratum estimand such as the survivor average causal effect is not identified by randomization alone without additional assumptions.
Some validation outcomes were assessed in selected post-baseline groups, including patients still in the index ICU or receiving mechanical ventilation on day 10.
Conclusion:
Protocols should specify the target population, treatment conditions, response variable, strategy for death and other assessment-precluding events, and treatment-effect summary measure.
