[Risk factors analysis for severe acute kidney injury in septic patients and establishment and validation of an hour-specific prediction model].

Objective: To explore the risk factors of severe acute kidney injury (AKI) in septic patients, and to establish an hour-specific prediction model.

Methods: Based on the information of septic patients in the Medical Information Mart for Intensive Care- IV (MIMIC- IV) database, general information, comorbidities, vital signs, severity scoring system, laboratory indicators, invasive operations and medication use were recorded. The enrolled patients were randomized into a training set and a validation set according to a ratio of 7 : 3. AKI was diagnosed according to the guidelines of Kidney Disease: Improving Global Outcome (KDIGO). Based on Lasso regression and Cox regression, the risk factors of severe AKI (AKI stage 2 and stage 3) in septic patients were analyzed and hour-specific prediction model were established. Consistency index (C-index), area under the receiver operator characteristic curve (AUC) and calibration curve were used to assess the predictive efficacy of the model.

Results: A total of 20 551 septic patients were enrolled, including 14 385 patients in the training set and 6 166 patients in the validation set. Multivariate Cox regression analysis showed that atrial fibrillation [hazard ratio (HR) = 1.266, 95% confidence interval (95%CI) was 1.150-1.393], heart failure (HR = 1.348, 95%CI was 1.217-1.493), respiratory failure (HR = 1.565, 95%CI was 1.428-1.715), heart rate (HR = 1.004, 95%CI was 1.002-1.007), mean arterial pressure (HR = 1.245, 95%CI was 1.126-1.377), lactic acid (HR = 1.051, 95%CI was 1.025-1.077), simplified acute physiology score II (SAPS II, HR = 1.019, 95%CI was 1.016-1.023), serum creatinine (HR = 1.171, 95%CI was 1.127-1.216), anion gap (HR = 1.024, 95%CI was 1.010-1.038), serum potassium (HR = 1.155, 95%CI was 1.079-1.236), white blood cell count (HR = 1.006, 95%CI was 1.003-1.009) and furosemide use (HR = 0.414, 95%CI was 0.368-0.467) were independently associated with severe AKI in septic patients (all P < 0.01). The above predictors were applied to construct an hour-specific prediction model for the occurrence of severe AKI in septic patients. The C-index of the prediction model was 0.723 and 0.735 in the training and validation sets, respectively. The AUC for the occurrence of severe AKI at 12, 24, and 48 hours were 0.795 (95%CI was 0.782-0.808), 0.792 (95%CI was 0.780-0.805), and 0.775 (95%CI was 0.762-0.788) in the training set, and the AUC were 0.803 (95%CI was 0.784-0.823), 0.791 (95%CI was 0.772-0.810), and 0.773 (95%CI was 0.752-0.793) in the validation set, respectively. The calibration curves of the two cohorts were in good agreement.

Conclusions: The hour-specific prediction model effectively identifies high-risk septic patients for developing severe AKI within 48 hours, aiding clinicians in stratifying patients for early therapeutic interventions to improve outcomes.