The nationwide cohort study by Oh and Song examined the incidence and predictors of post-intensive care syndrome after critical illness 1. By leveraging a large national claims database and not including patients with previous post-intensive care syndrome-related diagnoses, the authors provide an important population-level estimate of morbidity. Their work is timely and clinically relevant. However, several methodological issues may affect interpretation of the reported associations. The cohort was restricted to patients who survived for at least 12 months following hospital discharge. Although this design captures post-intensive care syndrome among longer-term survivors, it may also introduce survivor selection bias, as death is a competing event rather than an independent censoring mechanism. Patients with ICU exposures with greater illness severity may be less likely to survive long enough to enter the analytic cohort, potentially distorting subsequent associations with post-intensive care syndrome and partly explaining findings such as the apparently lower adjusted risk associated with extracorporeal membrane oxygenation or COVID-19. A different approach would be to retain all hospital discharge survivors, define time zero at discharge and follow patients until incidence of post-intensive care syndrome, death or 12 months. This would permit estimation of cumulative incidence with death treated explicitly as a competing event. Presenting cause-specific or distribution hazard models, together with a sensitivity analysis using a composite endpoint such as death or post-intensive care syndrome, would help readers judge whether the reported associations reflect lower post-intensive care syndrome risk itself or differential survival into the risk set. The cohort appears highly heterogeneous, which may limit clinical interpretability. A substantial proportion of included admissions were surgery-related, and the mean ICU stay was very short, suggesting that many patients may have undergone brief postoperative monitoring rather than prolonged critical illness. In this setting, the overall incidence and predictors of post-intensive care syndrome may represent an average across markedly different clinical trajectories rather than the phenotype typically associated with ICU survivorship syndromes 2. This concern could be addressed through prespecified stratified analyses according to medical vs. surgical admission, mechanical ventilation exposure and longer vs. shorter ICU stay. Such analyses would clarify whether the main findings are robust across clinically distinct subgroups or are driven primarily by short-stay postoperative patients. Although the authors adjusted for a broad range of demographic and clinical covariates, residual confounding remains plausible. Important determinants of post-intensive care syndrome, including delirium, sedative and analgesic exposure, frailty, physiologic illness severity and post-discharge rehabilitation intensity, were unavailable in the claims database. Quantitative sensitivity analyses would therefore be valuable. In particular, reporting E-values for key estimates could help readers assess how strong an unmeasured confounder would need to be to explain away the observed associations, especially those with modest effect sizes 3.
Chen et al. (Wed,) studied this question.