Heterogeneous Treatment Effects of Renal Replacement Therapy in Critically Ill Patients: A Multicenter Observational Study Using a Japanese ICU Registry

OBJECTIVE: Randomized trials evaluating the timing of renal replacement therapy (RRT) have informed current practice toward more conservative initiation strategies because of the lack of mortality benefit from early initiation. However, the optimal indications for RRT to offer lifesaving benefits remain uncertain. We aimed to estimate heterogeneous treatment effect of RRT and to identify patient groups associated with benefit or harm. DESIGN: Multicenter observational study. SETTING: A nationwide intensive care registry in Japan, Japanese Intensive Care Patient Database. PATIENTS: Adult patients who were admitted to the ICUs from 2018 through 2023 were included, excluding those with end-stage kidney disease, early death, readmission, or normal renal function. INTERVENTIONS: None. MEASUREMENTS AND MAIN RESULTS: Exposure was RRT during the ICU stay. Hospital mortality was the primary outcome. After 1:1 propensity score matching to adjust for confounding, 18,794 patients (9,397 per group) were analyzed. On average, RRT was associated with higher hospital mortality (risk difference, 6.1% points [95% CI, 3.4-8.7]). However, our machine learning-based causal forest algorithm showed substantial heterogeneity of treatment effects heterogeneity of treatment effects (HTEs) (rank-weighted average treatment effect, -8.2 percentage points [95% CI, -9.7 to -6.7]). Patients characterized by elevated serum creatinine, reduced urine output, and elevated blood urea nitrogen levels showed potential mortality benefit from RRT, whereas relatively preserved renal function was more likely to be associated with harm. CONCLUSIONS: Although RRT was associated with an increased risk of hospital mortality overall, HTEs were identified in selected subgroups that may benefit from RRT. These findings support a data-driven approach to defining clinically applicable optimal RRT indications to guide more judicious use of RRT.