AI-powered Predictive Model for Early Detection of Stroke In Children with Sickle Cell Disease Using Clinical Data
Evan Adetoye1, Funmilola Banjo2, Comfort Akanni2
1Medicine and Surgery, Obafemi Awolowo College of Health Sciences, Olabisi Onabanjo University, 2Medicine, Olabisi Onabanjo University
Objective:
This study aims to develop a predictive model that accurately identifies children at high risk of stroke based on clinical data and identifies the most important clinical variables that contribute to the model’s predictive power.
Background:
Children with sickle cell disease (SCD) are highly susceptible to neurological complications with one of the most devastating complications being stroke, which can lead to long-term disability, cognitive impairment, and even death. Current methods for stroke detection in SCD patients rely on regular transcranial doppler (TCD) screenings and clinical evaluation, but these methods have limitations in terms of accessibility, cost, and accuracy.
Design/Methods:
A retrospective cohort design utilizing existing clinical data from SCD patients at a tertiary hospital in Nigeria. Relevant clinical features, including demographic data, laboratory results, medical history and medications were extracted. StrokeSCDRisk, a multimodal deep learning model, analyzedcomplex patterns and relationships in the clinical data of 150 SCD patients to identify high-risk features and predict stroke likelihood.
Results:
StrokeSCDRisk demonstrated strong predictive performance with stroke prediction accuracy of 87%, a low FPR (4%), low FNR (5.2%), F1 score of 0.85 and ROC-AUC of 0.88, an improvement on traditional models. Key predictors of stroke were history of transient ischemic attack, haemoglobin levels, and frequency of sickle cell crises.
Conclusions:
These results indicate that StrokeSCDRisk can accurately identify high risk SCD patients while minimizing misclassification, highlighting its potential to support clinical decision making and improve patient outcomes.
Disclaimer: Abstracts were not reviewed by Neurology® and do not reflect the views of Neurology® editors or staff.