[Development of acute kidney injury prognostic model for critically ill patients based on MIMIC-III database].

OBJECTIVE To investigate the risk factors affecting the prognosis of patients with acute kidney injury (AKI) in the intensive care unit (ICU) based on the Medical Information Mart for Intensive Care III (MIMIC-III) database, and to establish a prognostic model for AKI. METHODS Patients (aged ≥ 18 years) with acute renal failure, admitted to the ICU for the first time, and had complete hospital records (the RIFLE diagnostic criteria were used in the database, and the diagnosis was expressed as AKI in this article) were screened from MIMIC-III database according to diagnostic codes. Patients were divided into two groups based on survival state at discharge, and the general information, underlying diseases, injury factors, vital signs and laboratory indicators within 24 hours after AKI, related intervention and prognostic indicators were analyzed. Univariate and multivariate Logistic regression analysis were used to determine the risk factors affecting mortality in patients with AKI and established a prediction model. The receiver operator characteristic curve (ROC curve) was used to analyze the predictive value of the prediction model for the prognosis of AKI patients. RESULTS There were 4 554 patients with AKI included and 862 died, with mortality of 18.93%. Univariate Logistic regression analysis was performed for factors that might be associated with death in AKI patients, and the results showed that age, hypertension, lymphoma, metastatic carcinoma, vancomycin, aspirin, coagulation abnormalities, cardiac arrest, sepsis or septic shock, invasive mechanical ventilation, white blood cell count (WBC), platelet count (PLT), K+, blood urea nitrogen (BUN), total bilirubin (TBil), renal replacement therapy (RRT) and length of stay (LOS) were independent risk factors [odds ratio (OR) and 95% confidence interval (95%CI) were 1.002 (1.001-1.003), 0.764 (0.618-0.819), 1.749 (1.112-2.752), 2.606 (1.968-3.451), 1.779 (1.529-2.071), 0.689 (0.563-0.842), 1.871 (1.590-2.201), 2.468 (1.209-5.036), 2.610 (2.226-3.060), 2.154 (1.853-2.505), 1.105 (1.009-1.021), 0.998 (0.997-0.998), 1.132 (1.057-1.212), 1.008 (1.006-1.011), 1.061 (1.049-1.073), 2.142 (1.793-2.997), 0.805 (0.778-1.113), all P < 0.05]. Further binary Logistic regression analysis showed that lymphoma, metastatic cancer, vancomycin, cardiac arrest, sepsis or septic shock, coagulation dysfunction, invasive mechanical ventilation, increased BUN, increased TBil, increased or decreased blood K+ and increased WBC were independent risk factors for death [β values were 0.636, 1.005, 0.207, 0.894, 0.787, 0.346, 0.686, 0.006, 0.051, 0.085, and 0.009; OR and 95%CI were 1.889 (1.177-3.031), 2.733 (2.027-3.683), 1.229 (1.040-1.453), 2.445 (1.165-5.133), 2.197 (1.850-2.610), 1.413 (1.183-1.689), 1.987 (1.688-2.338), 1.006 (1.003-1.009), 1.052 (1.039-1.065), 1.089 (1.008-1.176), and 1.009 (1.004-1.015), respectively, all P < 0.05]. The Hosmer-Lemeshow test showed that the AKI prognostic model was able to fit the observed data well (P = 0.604). ROC curve analysis showed that the area under ROC curve (AUC) of the AKI prognostic model was 0.716 (95%CI was 0.697-0.735), when the cut-off value was 0.320, the sensitivity was 71.9%, the specificity was 60.1%, the positive likelihood ratio was 1.80, and the negative likelihood ratio was 0.47. CONCLUSIONS The prognostic prediction model of AKI in critically ill patients established and based on the MIMIC-III database may have practical significance for prognostic risk assessment of AKI and later intervention.