Explainable AI-Driven Clinical Decision Support Systems for Real-Time Cardiovascular Risk Stratification in Resource-Limited Settings

Abstract

Cardiovascular disease remains the leading cause of global mortality, disproportionately afflicting populations in low- and middle-income countries where diagnostic infrastructure and specialist expertise are critically scarce. Conventional risk stratification frameworks, such as the Framingham Risk Score and pooled cohort equations, exhibit limited applicability in heterogeneous, resource-constrained clinical environments due to their reliance on complete laboratory panels and population-specific calibration. This paper presents a novel Explainable Artificial Intelligence--driven Clinical Decision Support System (XAI-CDSS) designed to enable real-time cardiovascular risk stratification under conditions of data sparsity, computational constraint, and limited clinical oversight.

The proposed framework integrates gradient-boosted ensemble learning with post-hoc interpretability mechanisms, specifically SHapley Additive exPlanations (SHAP) and Layer-wise Relevance Propagation (LRP), to produce transparent, clinician-interpretable risk assessments from routinely available clinical parameters. A lightweight model architecture is optimized for deployment on low-power edge devices, ensuring operability in settings lacking continuous internet connectivity or high-performance computing resources. The system is validated on a multi-site retrospective cohort comprising over 18,000 patients across four sub-Saharan African and South Asian clinical centers.

Experimental results demonstrate that the XAI-CDSS achieves an area under the receiver operating characteristic curve (AUROC) of $0.891 \pm 0.014$ on held-out test sets, outperforming conventional scoring instruments by a statistically significant margin ($p < 0.001$). Calibration analyses confirm superior reliability, with a mean Brier score of $0.087$, while SHAP explanations exhibit high fidelity to domain clinical knowledge.

This work advances the translational potential of responsible AI in global cardiovascular medicine, offering a scalable, trustworthy, and resource-aware decision support paradigm with direct implications for equitable healthcare delivery.