Analysis of 60,294 adults across 12 African countries reveals that machine learning models significantly outperform conventional cardiovascular risk scores like Framingham in African populations. The XGBoost algorithm achieved the highest discrimination with an AUC of 0.769, while identifying hypertension and alcohol-related harm as the strongest predictors among the 5% CVD prevalence population. This finding addresses a critical gap in global health equity. Traditional risk prediction tools developed primarily in high-income populations often miscalibrate when applied to African contexts, potentially leading to inappropriate clinical decisions. The superior performance of tree-based algorithms suggests complex, non-linear relationships between risk factors that linear models miss entirely. For cardiovascular disease prevention—the leading cause of death globally—this represents meaningful progress toward personalized medicine in resource-limited settings. However, this preprint awaits peer review, and several limitations temper enthusiasm. The 5% self-reported CVD prevalence likely underestimates true disease burden due to underdiagnosis in African healthcare systems. Additionally, cross-sectional survey data cannot establish causation, and the models require prospective validation before clinical implementation. While promising for population health screening, these tools need real-world testing.
Machine Learning Improves Cardiovascular Risk Prediction Across 12 African Countries
📄 Based on research published in medRxiv preprint
Read the original research →⚠️ This is a preprint — it has not yet been peer-reviewed. Results should be interpreted with caution and may change following peer review.
For informational, non-clinical use. Synthesized analysis of published research — may contain errors. Not medical advice. Consult original sources and your physician.