Development and validation a machine learning model based on clinical factors to predict short-term prognosis of ICU intracerebral hemorrhage patients: a retrospective study

Objective This study aims to construct and validate a short-term prognostic nomogram model for Intracerebral hemorrhage (ICH) based on information immediately upon admission, providing support for early risk identification and clinical decision-making in the intensive care unit (ICU). Methods This study retrospectively analysised 135 patients with intracerebral hemorrhage (ICH) admitted to the ICU of Neijiang Hospital of Traditional Chinese Medicine in Sichuan Province from January 2021 to December 2025, and divided them sequentially into a training set (n = 97) and a validation set (n = 38). Univariate and multivariate Cox regression analyses were employed to identify independent predictive factors, and nomogram were constructed to predict survival probabilities. Model discrimination, calibration, clinical utility, and stratification capability were evaluated using receiver operating characteristic curves, calibration curves, decision curves and Kaplan-Meier survival curves. Results Multivariate analysis showed that age ≥ 65 years, history of heart disease, admission pulse ≥ 82 beats/min, admission GCS ≥ 9 points, decreased pupil reflex to light, abnormal muscle tone, physiological reflex loss, and white blood cell count ≥ 10 × 10 ⁹/L were independent predictive factors for ICU admission mortality. The column chart constructed based on the above indicators had AUCs of 0.878, 0.824, and 0.863 on the 3/5/7 days in the training set, and AUCs of 0.727, 0.772, and 0.761 in the validation set. The Kaplan Meier curve for risk stratification clearly distinguishes between high, intermediate, and low-risk groups ( P < 0.0001). Conclusion The short-term survival prediction nomogram for ICU intracerebral hemorrhage patients constructed in this study has good predictive performance, robust validation results, and high clinical usability. This model is based on readily available clinical information at the bedside and can achieve early risk stratification for severe ICH patients, providing reliable basis for clinical treatment decision-making, resource allocation, and communication with family members. In the future, multi center prospective studies are needed to further validate its generalization ability.