Balanced crystalloids versus saline for initial resuscitation in sepsis and 30-day major adverse kidney events: A target trial emulation across two critical-care databases

Background Whether balanced crystalloids improve kidney outcomes in sepsis remains uncertain, and observational comparisons are vulnerable to time-zero misalignment, post-treatment selection, confounding by indication, and limited treatment overlap. We evaluated whether the first recorded balanced crystalloid, compared with first recorded saline, was associated with 30-day major adverse kidney events (MAKE30). Methods We emulated a point-intervention target trial in the Medical Information Mart for Intensive Care IV (MIMIC-IV) and performed a prespecified secondary multicentre comparison in the eICU Collaborative Research Database (eICU-CRD). Time zero was the start of the first recorded eligible intensive-care-unit crystalloid. The primary exposure was first recorded balanced crystalloid versus first recorded saline. Baseline-only propensity-score models included variables measured no later than time zero; the eICU-CRD analysis was restricted to hospitals with at least five patients in each treatment group and incorporated hospital effects and hospital-clustered uncertainty. Alternative 6-h and 24-h exposure definitions, post-treatment covariate adjustment, renal-outcome thresholds, and E-values were examined as sensitivity analyses. Results The primary cohorts included 5705 patients in MIMIC-IV (1003 balanced; 4702 saline) and 3549 patients in the eICU-CRD within-site-overlap sample (311 balanced; 3238 saline). The MAKE30 risk ratio (RR) was 0.67 (95% confidence interval [CI], 0.60–0.73) in MIMIC-IV and 1.00 (0.69–1.33) in eICU-CRD; the random-effects synthesis was imprecise and heterogeneous (RR, 0.79; 0.54–1.16; I² = 80.6%). The corresponding mortality hazard ratios were 0.65 (0.57–0.75) and 1.10 (0.68–1.79), with a random-effects estimate of 0.80 (0.49–1.33). The pooled creatinine-defined acute kidney injury RR was 0.88 (0.78–0.99). Estimates varied materially across exposure definitions, which represented different treatment strategies and study populations. Conclusions First recorded balanced crystalloid was associated with fewer MAKE30 events in MIMIC-IV, but this association was not reproduced in the eICU-CRD within-site comparison. The heterogeneous and imprecise cross-database synthesis, together with susceptibility to residual confounding, outcome misclassification, and database-specific selection, precludes a confirmatory causal interpretation.