Comparative efficacy and safety of anticoagulation strategies in continuous renal replacement therapy: a real-world cohort study

BACKGROUND: The optimal anticoagulation strategy for continuous renal replacement therapy (CRRT) remains controversial. Real-world comparisons of contemporary agents-regional citrate anticoagulation (RCA), nafamostat mesylate (NM), and unfractionated heparin (UH)-are limited. METHODS: In this single-center retrospective cohort study, we analyzed 420 CRRT sessions (from 197 unique patients) conducted in four intensive care unit (ICU) wards at an academic medical center in China between September 2025 and November 2025. The primary efficacy outcome was filter lifespan, with Restricted Mean Survival Time (RMST) analysis adopted as the primary method (due to violation of the proportional hazards assumption). Propensity score matching (PSM) and mixed-effects Cox models were used as sensitivity analyses, adjusting for patient, treatment, ward, and machine factors. Multiplicity adjustment (Holm-Bonferroni) was applied for pairwise comparisons. RESULTS: Among 420 sessions, RCA was the most frequently used approach (59.0%), followed by NM (21.2%), UH (12.6%), and no anticoagulation (4.0%). Median filter lifespan was longest with RCA (65.5 h, IQR 48.0-72.0) compared to NM (32.0 h, IQR 15.5-52.5), UH (40.0 h, IQR 26.0-72.0), and no anticoagulation (12.5 h, IQR 6.0-19.5; P < 0.001). RMST analysis confirmed that RCA was associated with significantly longer mean filter survival time compared to NM (difference 18.2 h, 95% CI 10.5-25.9; adjusted P < 0.001) and UH (difference 14.6 h, 95% CI 5.8-23.4; adjusted P = 0.046). The NM vs. UH comparison showed no significant difference (difference - 3.6 h, 95% CI -12.1 to 4.9; adjusted P = 0.52). After PSM, results remained consistent. Major bleeding events were rare (n = 9, 2.1%), but the small number of events precludes meaningful between-group safety comparison. However, 20.9% of RCA sessions failed to achieve the protocol-specified post-filter ionized calcium target, reflecting a metabolic calcium monitoring challenge unique to citrate anticoagulation. Twenty-eight-day mortality (patient-level analysis, n = 197) did not differ significantly across groups after adjustment (overall 36.5%, 72/197). CONCLUSION: In this retrospective data analysis of real-world practice, RCA provided superior filter lifespan compared to NM and UH for CRRT, while NM showed intermediate efficacy. Although major bleeding events were uncommon, the small number of events precludes definitive safety conclusions. Vigilant monitoring remains necessary regardless of the agent employed. The choice of anticoagulation should balance filter efficacy, bleeding risk, metabolic monitoring capabilities, and institutional resources.