A Hybrid Bidirectional Deep Learning Model Using HRV for Prediction of ICU Mortality Risk in TBI Patients

Accurately predicting early mortality risk for traumatic brain injury (TBI) patients admitted to the intensive care unit (ICU) is crucial for optimizing patient care, allocating resources effectively, and reducing mortality rates. This study introduces an approach to predict mortality risk for TBI patients by analysing heart rate variability from the first 24 h of electrocardiogram (ECG) signals. A deep learning hybrid model was developed by integrating a weight predictor with a bidirectional long short-term memory (BiLSTM) unit. This hybrid architecture enhances predictive performance by weighting features and capturing patterns in HRV data. This study utilised TBI patient data from the Gold Coast University Hospital and Cerebral Haemodynamic Autoregulatory Information System (CHARIS) for model training and testing. The experimental results demonstrated that the proposed hybrid model achieved cross-validation metrics, including an accuracy of 0.933 (95% CI: 0.844–1.000), an area under the curve of the receiver operating characteristics (AUROC) of 0.995 (0.978–1.000), and an area under the precision‒recall curve (AUPRC) of 0.998 (0.99–1.000). With the hold-out test dataset, the model obtained a prediction accuracy of 0.917 (0.75–1.000), an AUROC of 0.926 (0.766–1.000), and an AUPRC of 1.0. Comparative analysis with conventional machine learning models confirmed that the proposed model significantly outperformed existing approaches. The results highlight the potential of the proposed model in helping critical care strategies by providing more accurate early predictions of mortality risk through HRV analysis. Since the proposed model relies exclusively on ICU monitoring ECG data, it facilitates straightforward implementation in clinical settings.