Machine Learning Models for Predicting Stroke Mortality in Malaysia: An Application and Comparative Analysis

Abstract

Background Stroke is a significant public health concern characterized by increasing mortality and morbidity. Accurate long-term outcome prediction for acute stroke patients, particularly stroke mortality, is vital for clinical decision-making and prognostic management. This study aimed to develop and compare various prognostic models for stroke mortality prediction. Methods In a retrospective cohort study from January 2016 to December 2021, we collected data from patients diagnosed with acute stroke from five selected hospitals. Data contained variables on demographics, comorbidities, and interventions retrieved from medical records. The cohort comprised 950 patients with 20 features. Outcomes (censored vs. death) were determined by linking data with the Malaysian National Mortality Registry. We employed three common survival modeling approaches, the Cox proportional hazard regression (Cox), support vector machine (SVM), and random survival forest (RSF), while enhancing the Cox model with Elastic Net (Cox-EN) for feature selection. Models were compared using the concordance index (C-index), time-dependent area under the curve (AUC), and discrimination index (D-index), with calibration assessed by the Brier score. Results The support vector machine (SVM) model excelled among the four, with three-month, one-year, and three-year time-dependent AUC values of 0.842, 0.846, and 0.791; a D-index of 5.31 (95% CI: 3.86, 7.30); and a C-index of 0.803 (95% CI: 0.758, 0.847). All models exhibited robust calibration, with three-month, one-year, and three-year Brier scores ranging from 0.103 to 0.220, all below 0.25. Conclusion The support vector machine (SVM) model demonstrated superior discriminative performance, suggesting its efficacy in developing prognostic models for stroke mortality. This study enhances stroke mortality prediction and supports clinical decision-making, emphasizing the utility of the support vector machine method.

Keywords: comparative analysis; machine learning; malaysia; prediction models; stroke mortality.

Figure 1. Graphical illustration of the study workflow

HCTM UKM: Hospital Canselor Tuanku Muhriz Universiti Kebangsaan Malaysia; HSJ: Hospital Seberang Jaya; HSNZ: Hospital Sultanah Nur Zahirah; HRPZ II: Hospital Raja Perempuan Zainab II; HUSM: Hospital Universiti Sains Malaysia; Cox: Cox proportional hazard regression; Cox-EN: Cox model with elastic net; C-index: Concordance index; D-index: Discrimination index.

Figure 2. Overall survival of acute stroke patients

Figure 3. The coefficient of features change for varying alphas for the Cox-EN model

nihss: National Institutes of Health Stroke Scale; dm: Diabetic status; age: Age in years; gcs_reduct: Glasgow Coma Scale reduction; ethnicity: Ethnicity; ss: Study site; sex: Gender of the patients.

Figure 4. The important coefficient of each feature corresponding to the optimal α by an elastic net

age: Age in years; nihss: National Institutes of Health Stroke Scale; dm: Diabetic status; sex: Gender of the patients; ethnicity: Ethnicity; gcs_reduct: Glasgow Coma Scale reduction; lipid: Hyperlipidemia; hf_ihd: Heart disease; married: Marital status; hpt: Hypertension status; ss: Study site; af: Atrial fibrillation.

Figure 5. The important coefficient of each feature by random survival forest

af: Atrial fibrillation; hpt: Hypertension status; ethnicity: Ethnicity; hf_ihd: Heart disease; married: Marital status; lipid: Hyperlipidemia; sex: Gender of the patients; ss: Study site; dm: Diabetic status; gcs_reduct: Glasgow Coma Scale reduction; nihss: National Institutes of Health Stroke Scale; age: Age in years.

Figure 6. Time-dependent receiver operating characteristic curves of models at three months, one year, and three years

Cox: Cox proportional hazard regression; Cox-EN: Cox model with elastic net; RSF: Random survival forest; SVM: Support vector machine.

Figure 7. Time-dependent AUC of models over time

AUC: Area under the curve; Cox: Cox proportional hazard regression; Cox-EN: Cox model with elastic net; RSF: Random survival forest; SVM: Support vector machine. The blue dotted line represents the mean area under the curve.

Figure 8. Survival curves of high-risk and low-risk groups divided according to the risk score from (A) Cox: Cox proportional hazard regression, (B) Cox-EN: Cox model with elastic net, (C) SVM: support vector machine, and (D) RSF: random survival forest

Figure 9. Performance of the SVM model across different alpha values during hyperparameter tuning