Predicting ICU Mortality in Sepsis: A Retrospective Cohort Study with Nomogram Development

Background Sepsis remains a leading cause of death among critically ill patients, yet existing severity scores such as the APACHE II and the SOFA provide limited individualized prognostic accuracy. This study aimed to develop and internally validate a nomogram to predict ICU mortality among adult patients with sepsis. Methods We performed a retrospective cohort study including 505 adult patients with sepsis admitted to the ICUs of a tertiary medical center in Taiwan between 2017 and 2021. Clinical, laboratory, and infection-related data at ICU admission were extracted from electronic medical records. Univariate and multivariate logistic regression analyses were used to identify independent predictors of ICU mortality. Variables with statistical significance in the multivariate model were incorporated into a predictive nomogram. Model calibration and discrimination were evaluated using calibration plots and the area under the ROC curve. Results Among 505 patients, 225 (44.6%) died during ICU stay. Independent predictors of ICU mortality included male gender (adjusted odds ratio [AOR] 0.62, 95% confidence interval [CI] 0.39–0.99), higher body mass index (AOR 1.09 per kg/m², 95% CI 1.04–1.13), higher APACHE II score (AOR 1.07 per point, 95% CI 1.03–1.10), pneumonia as the primary infection source (AOR 2.45, 95% CI 1.50–3.99), lower hemoglobin level (AOR 0.96 per g/dL, 95% CI 0.92–0.99), and higher serum bilirubin (AOR 1.07 per mg/dL, 95% CI 1.01–1.14) and lactate (AOR 1.08 per mmol/L, 95% CI 1.01–1.16). The nomogram demonstrated good discrimination (area under the curve = 0.84) and satisfactory calibration between predicted and observed mortality rates. Conclusions This study developed an internally validated nomogram integrating demographic, physiologic, and biochemical parameters available at ICU admission to predict mortality in patients with sepsis. The model provides a practical, individualized bedside tool to assist early risk stratification, guide management decisions, and optimize resource allocation in critical care settings. External validation in independent cohorts is warranted. Trial registration Not applicable.