New prognostic score for mortality in critically ill patients. Development and validation

Abstract

Objective: To develop and validate a novel prognostic model to predict mortality in critically ill patients admitted to the ICU. Unlike APACHE II the new model incorporates risk factors assessed throughout the entire ICU stay, allowing for a more comprehensive evaluation and a better understanding of how the probability of mortality changes. Design: Post-hoc analysis of multicenter, prospective data from 167 Spanish hospitals (193 ICUs) collected over 7 years. Patients: Adults (>18 years). Interventions: Demographic and clinical variables were analyzed. The model was developed using multivariable logistic regression in an estimation group and validated using a separate cohort. Variables of Interest: The primary outcome was ICU mortality, which was clearly defined and analyzed in both the model development and validation phases. The model incorporated APACHE II and eight additional factors evaluated across the entire ICU stay: prior antibiotic use (48 h pre-ICU), hospitalization days before ICU, hematologic diagnoses, invasive mechanical ventilation, parenteral nutrition, shunt ventricular, renal clearance techniques, and infections associated with invasive devices leading to sepsis. Results: Of the 137,666 patients, 91,777 were assigned to the estimation group and 45,889 to validation. Mortality was 10.8 %, strongly associated with APACHE II severity. The new model demonstrated superior discriminatory ability (AUROC = 0.872) compared to APACHE II alone (AUROC = 0.826) and it improved reclassification by 52 % over APACHE II: 19.4 % of survivors and 32.75 % of non-survivors. This improvement, though numerically modest, has clinical relevance by enhancing risk stratification and guiding interventions. Conclusions: A validated NMP model was developed using nine additional risk factors alongside APACHE II.