Development and internal validation of an ICU mortality prediction model for patients with concurrent sepsis and heart failure: a retrospective cohort study based on eICU-CRD.

backgroundSepsis and heart failure are common critical conditions within the ICU, and their coexistence significantly increases the difficulty of patient management and mortality risk. This study aims to develop and validate a practical nomogram for predicting ICU mortality, thereby facilitating risk stratification and clinical decision-making for patients coded as having both sepsis and heart failure.

methodsThis study constitutes a retrospective cohort investigation. Patients meeting inclusion criteria for both sepsis and heart failure were identified via the eICU-CRD database, subsequently randomised in a 7:3 ratio to form training and validation cohorts. Variable selection employed the Least Absolute and Selective Operator (LASSO) regression, with nomogram analysis constructed. Model discriminatory ability was assessed via area under the receiver operating characteristic curve (AUC). Calibration was evaluated using calibration curves and the Hosmer-Lemeshow test. Decision curves and clinical impact curves were plotted to evaluate the model's net benefit and clinical applicability.

resultsA total of 1,394 patients were included and divided into a training set (975 cases) and a validation set (419 cases) at a ratio of 7:3. The final model integrated eight independent predictors: mechanical ventilation, lactate, respiratory rate, white blood cell count, age, platelet count, systolic blood pressure, and oxygen saturation. The model demonstrated acceptable to good discriminatory performance in both training and validation cohorts (AUC 0.826 [95% CI 0.789-0.863] and 0.798 [95% CI 0.732-0.864], respectively), with good calibration (Brier scores: training cohort 0.090, validation cohort 0.091; Hosmer-Lemeshow goodness-of-fit test P-values > 0.05), and demonstrated clear clinical net benefit.

conclusionsThis study developed and internally validated a predictive nomogram based on eight routine clinical variables for ICU mortality in patients coded for both sepsis and heart failure. The model demonstrated acceptable discrimination and calibration. However, given the broad, code-based case definition and the lack of heart failure-specific variables, this tool should be reframed as an internally validated ICU mortality prediction model for this high-risk population, rather than as a cardiology-specific or clinically actionable nomogram. External validation is required before any clinical application.