Validación externa y actualización de un modelo predictivo de lesión renal aguda en pacientes adultos hospitalizados en cuidados intensivos

Abstract

AimTo analyze the predictive performance of the risk model for acute kidney injury (AKI) by Malhotra et al., in adults hospitalized in the intensive care unit (ICU) of Clínica Somer (Colombia), 2017 and 2021. MethodsRetrospective follow-up study of a cohort based on medical records. Critically ill patients over 18 years of age admitted to the ICU were included. Patients with end-stage chronic kidney disease or on renal replacement therapy, AKI at admission were excluded. AKI in the first week in the ICU was defined as the outcome variable, and those of the Malhotra model and additional clinical variables as predictor variables. The predictive performance was analyzed for the original (m0), calibrated (m1–m2) and updated (m3–m6) models using the C statistic, the Hosmer–Lemeshow (HL) test, and calibration maps. ResultsA total of 1235 patients were included, median age 60 years (IQR 45–60), 39.9% women. The incidence of AKI was 30.6%. The original model (m0) showed fit problems (p-HL value <0.05) overestimating the risk of AKI and obtained a C statistic of 0.68 (95% CI: 0.65–0.71). Calibration using intercept and global factor (m2) solved the adjustment problem. The update with ten additional variables (m6) only increased the C statistic to 0.72 (95% CI: 0.69–0.75). ConclusionThe performance of the original model in the external validation cohort is not perfect due to differences in the epidemiology of AKI in the ICU. The model updated improves the fit but not the discrimination, yet it still holds some degree of utility in clinical practice.