From data to diagnosis: Revolutionising cardiovascular risk assessment in Sri Lanka with artificial intelligence

There are no cardiovascular risk prediction models designed explicitly for Sri Lankans. As a result, we are relying on risk prediction models developed for white Caucasians or the Southeast Asia region (SEAR) to risk-stratify Sri Lankans. Our research has shown that risk predictions from different models can vary significantly. None of the risk prediction models has been validated for the Sri Lankan population, a gap we addressed in 2019. The findings demonstrated that the pre-dictions of the World Health Organisation (WHO-SEAR) risk charts were effective for low-risk individuals but less sensitive for high-risk individuals and women.
Therefore, we aimed to develop a risk prediction model specific to Sri Lankans, using data from a cohort followed up for 10 years as part of the Ragama Health Study. We applied machine learning (ML) to the dataset and found that the predictions generated by artificial intelligence are more accurate than those provided by the WHO risk charts. Subsequently, we developed a new ML-based risk prediction model for Sri Lankans, named the SLCVD score, along with an online risk calculator for practical use, even in rural areas of Sri Lanka. Furthermore, this new model was validated in an external cohort of hospital-based patients, demonstrating that it is more effective at identifying high-risk individuals in Sri Lanka than the WHO model. We further refined the model for generalizability using nationally representative data from the STEPwise approach to surveillance (STEPS) survey conducted in 2021. This risk calcu-lator will be launched through the Ministry of Health for free usage once we secure the patent rights.
Accurate risk prediction is crucial for implementing primary preventive measures in a “High-Risk strategy” to ensure they are cost-effective. This approach is especially important in resource-limited settings like Sri Lanka.