Overlap-weighted retrospective cohort study of early versus delayed renal replacement therapy in sepsis-associated AKI

The optimal timing of continuous renal replacement therapy (CRRT) initiation in sepsis-associated acute kidney injury (SA-AKI) remains controversial. Landmark randomized controlled trials have yielded neutral results, but whether their findings extend to specific SA-AKI subpopulations warrants further investigation. This retrospective cohort study used the MIMIC-IV database (v3.1), informed by target trial design principles. Adult SA-AKI patients (KDIGO stage 2/3) who received CRRT were classified into early (≤12 h) or delayed (>12 h) groups based on time from AKI diagnosis to CRRT initiation. Propensity score overlap weighting was applied to adjust for baseline confounding. The primary outcome was 60-day all-cause mortality, analyzed using weighted Cox proportional hazards models. Secondary analyses included multivariable Cox regression, restricted mean survival time (RMST) analysis with 60-day truncation, and exploratory subgroup analyses. Among 629 SA-AKI patients who received CRRT (115 early, 514 delayed), overlap-weighted Cox regression showed that early CRRT was significantly associated with increased 60-day mortality (HR 1.62, 95% CI 1.17–2.25, p = 0.004). The association persisted in landmark analysis at 12 h (HR 1.48, 95% CI 1.06–2.06, p = 0.023) and was directionally consistent across sensitivity analyses. Restricted mean survival time difference at 60 days was − 11.2 days (95% CI -14.6 to -7.7, bootstrap p < 0.001). In exploratory subgroup analyses, the association was more pronounced in patients with thrombocytopenia (platelet < 100 × 10⁹/L; HR 3.06, 95% CI 1.68–5.57). In this overlap-weighted retrospective cohort study, early CRRT initiation was associated with increased 60-day mortality compared with a delayed strategy in SA-AKI patients who ultimately received CRRT. These findings, while requiring cautious interpretation given residual confounding, generate hypotheses regarding phenotype-specific CRRT timing and warrant prospective validation.