Explainable AI Framework for Clinical Decision Support in Rural Healthcare Settings

Abstract

AI systems offer significant potential to enhance health care in rural areas with limited resources through a clinical decision
support system (CDS). The lack of transparency of predictive models used in these contexts, however, introduces some
important questions about clinician trust, regulatory compliance and algorithmic accountability. This paper brings together
recent studies on explainable AI (XAI) frameworks tailored to the rural healthcare setting for clinical decision support
(CDS). This includes a thorough examination of model architectures, explanation methods and deployment issues based
on empirical studies in value-based primary care [1], systematic reviews in human-computer interaction [8] and field
studies of deploying AI-based CDS in rural clinics [13]. XAI-based CDS systems outperformed logistic regression baselines
by 37 percentage points for the clinician acceptance rate and by 0.18 points for the area under the receiver operating
characteristic curve (AUC-ROC) values when explanations were given with predictions. The proposed framework combines
the SHAP (SHapley Additive exPlanations) [17] and LIME (Local Interpretable Model-agnostic Explanations) [9] algorithms
with the attention-based visualization techniques in a modular design that is suitable for the low bandwidth of rural
infrastructure. This study systematically identifies digital literacy issues, connectivity issues, data governance issues in the
deployment of the system, and proposes strategies for iterative and user-centered system design [10]. All the results show
that explainability is not a side-effect, but a definite need for using AI in rural health.