Akademik Veri Yönetim Sistemi
COMMUNICATIONS IN COMPUTER AND INFORMATION SCIENCE, cilt.2669, sa.1, ss.463-473, 2025 (Scopus)
Trauma remains a leading cause of mortality and morbidity worldwide, underscoring the need for rapid and accurate outcome prediction to guide intensive care interventions. This study evaluates the performance of multiple machine learning (ML) algorithms—Decision Tree, Logistic Regression, K-Nearest Neighbors, Artificial Neural Network, and Deep Learning models—in forecasting in-hospital mortality among trauma patients admitted to the ICU. Using a retrospective cohort from January 2021 to August 2023, we incorporated a broad spectrum of clinical variables, including demographics, comorbidities, admission trauma scores, resuscitation details, laboratory results (blood gas analyses, complete blood count, and biochemistry), and treatment data (blood product transfusions and ventilator usage). Data integrity was maintained through rigorous chart review and nearest-neighbor imputation for missing values. Our results highlight the superior and robust performance of ANN and deep learning models across multiple test sets, with key predictors of mortality identified as initiation of inotropic drugs, APACHE II score, a severity of illness score, initial creatinine level, days on ventilator, Glasgow Coma Scale (GCS), and BE (base excess) as an input feature, our ML framework bridges traditional scoring approaches with advanced analytics, offering a clinically interpretable decision-support tool. Incorporating traditional scoring systems like APACHE II as features enhanced the prognostic capabilities of ML models, supporting their potential use as robust decision support tools in critical care settings. This research advances trauma care by refining mortality prediction, informing early intervention strategies, and ultimately aiming to reduce in-hospital death among critically injured patients.