Prediction model for successful liberation from continuous renal replacement therapy in patients with acute kidney injury: development and temporal validation.

BACKGROUND

The optimal timing of liberation from continuous renal replacement therapy (CRRT) in adults with acute kidney injury (AKI) remains uncertain. This study developed and temporally validated a model for predicting successful 7-day liberation after a planned CRRT discontinuation attempt and compared its performance with urine output and serum creatinine (SCr)-based indices.

METHODS

This single-center retrospective cohort included 447 adults with AKI who underwent a first planned CRRT discontinuation attempt between 2019 and 2024. Successful liberation was defined as survival without kidney replacement therapy (KRT) for 7 consecutive days after CRRT discontinuation. Patients treated in 2019-2022 formed the development cohort (n = 304), and those treated in 2023-2024 formed the temporal validation cohort (n = 143). Six prespecified predictors-Sequential Organ Failure Assessment (SOFA) score, sepsis-associated AKI, preceding 24-h urine output, relative SCr decline over the preceding 24 h, time-weighted mean arterial pressure (MAP), and vasoactive drug use-were entered into a multivariable logistic regression model. Multiple imputation, bootstrap optimism correction, calibration assessment, Brier score, decision curve analysis, and sensitivity analyses were performed.

RESULTS

Overall, 241 patients (53.91%) achieved successful 7-day liberation. The optimism-corrected area under the receiver operating characteristic curve (AUC) in the development cohort was 0.860 (95% confidence interval [CI], 0.817-0.900), and the temporal validation AUC was 0.853 (95% CI, 0.785-0.913). In temporal validation, the Brier score was 0.156 (95% CI, 0.123-0.192), calibration intercept was -0.225 (95% CI, -0.648 to 0.198), and calibration slope was 0.984 (95% CI, 0.654-1.313). At the fixed development-derived threshold of 0.529, sensitivity was 84.21% and specificity was 71.64%. The full model outperformed urine output alone and SCr decline alone, whereas its AUC did not differ significantly from the combined urine output plus SCr model.

CONCLUSION

This model provides individualized 7-day liberation risk estimates after standard CRRT discontinuation criteria and checklist completion. It is intended as an adjunctive risk-stratification tool for experienced clinicians and should not replace clinical judgment. Prospective multicenter external validation, local recalibration when needed, and clinical impact evaluation are warranted before implementation.