The prognostic and predictive modelling of mortality among acute stroke patients in peninsular Malaysia
Background: The rapid evolution of digital technology and artificial intelligence has revolutionized the application of machine learning in predicting stroke outcomes. The increasing burden of stroke in Malaysia, characterized by its impact on mortality and morbidity, underscores the need for accura...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2024
|
| Subjects: | |
| Online Access: | http://eprints.usm.my/61092/1/Che%20Muhammad%20Nur%20Hidayat%20Che%20Nawi-E.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Background: The rapid evolution of digital technology and artificial intelligence has revolutionized the application of machine learning in predicting stroke outcomes. The increasing burden of stroke in Malaysia, characterized by its impact on mortality and morbidity, underscores the need for accurate mortality prediction models. This need is heightened by the challenges in clinical decision-making and prognostic management, driving the development of various prognostic models and tools.
Objective: This study aimed to analyse trends in publications related to the application of machine learning in stroke outcome modelling, identify prognostic factors, perform predictive modelling for mortality among acute stroke patients, and develop a web-based application for stroke mortality prediction using data sourced from multiple stroke centres in Malaysia spanning the years 2016 to 2021.
Methodology: Our methodology spans a multifaceted approach: starting with a bibliometric analysis using Scopus and Web of Science data, followed by a retrospective cohort analysis of 950 stroke patients across five hospitals in peninsular Malaysia. We utilized survival analyses and an array of predictive modelling techniques, including Cox regression, Support Vector Machine (SVM), and Random Survival Forest (RSF). The development of Malaysian Ischemic Stroke Mortality Prediction Tool (MIST) involved a rigorous process of content and face validation with domain experts and users.
Results: The bibliometric analysis delineated a robust trend in machine learning research in the realm of stroke, punctuated by significant global collaborations. The retrospective study revealed a mean stroke onset age of 63.15 (13.09) years, with a male (n=552, 58.1%) and Malay ethnicity (n=771, 81.7%) predominance and a higher predictive precision of the National Institute of Health Stroke Score (NIHSS) scale (higher statistical significance, lower Akaike Information Criterion (AIC) values, a higher C-Index, and a more gradual decline in Kaplan-Meier survival curves) over Glasgow Coma Scale (GCS) for stroke-related mortality. Notably, the SVM model demonstrated superior predictive accuracy, evidenced by 3-month, 1-year, and 3-year time-dependent Area Under the Curve (AUC) values of 0.842, 0.846, and 0.791, a D-index of 5.31 (95% CI: 3.86, 7.30), a C-index of 0.803 (95% CI: 0.758, 0.847), and Brier scores ranging from 0.103 to 0.220. MIST, following comprehensive validation, was highly acclaimed by experts and users for its predictive accuracy and user-friendliness with Scale-Level Content Validity Index (S-CVI/Ave) and Scale-Level Face Validity Index (S-FVI/Ave) of 0.99 and 0.98, respectively.
Conclusion: Machine learning techniques are increasingly adopted in stroke research, facilitated by global collaborations and advancements in computational science. The study's findings highlight the need for effective predictive models in Malaysia, with SVM showing superior performance in mortality prediction. MIST, as a validated online tool, offers significant potential for enhancing stroke care and public health through accurate mortality risk estimation. |
|---|