An interpretable approach using hybrid graph networks and explainable AI for intelligent diagnosis recommendations in chronic disease care

With the rapid advancement of modern medical technology and the increasing demand for a higher quality of life there is an emergent requirement for personalized healthcare services. This is particularly pertinent in the sphere of pharmacological recommendations, where the necessity to provide patients with optimal and efficacious medication regimens is paramount. Traditional methodologies in this domain are increasingly seen as insufficient for the needs of contemporary medicine, prompting a shift towards more sophisticated technologies and algorithms. In this study, we addressed this pressing need by developing GCF++ i.e., two graph-based collaborative filtering methods, GCFYA (with attention) and GCFNA (without attention). These methods hold significant promise in revolutionizing how drug recommendations are made, ensuring that patients receive precise and trustworthy medication suggestions tailored to their unique needs and scenarios. To evaluate and compare these algorithms, we introduced three robust metrics: Precision, RMSE (Root Mean Square Error), and Recall. Precision value for GCF-YA is 88 % for hospital dataset, while 85 % for public dataset, similarly, GCF-NA is 77 % for hospital dataset while 78 % for public dataset which is much higher than other traditional methods. Furthermore, as algorithm models become increasingly intricate, transparency and interpretability have gained paramount importance. In response, we incorporated two model interpretation tools, SHAP and LIME, to demystify the decision-making processes behind these algorithms. These tools not only provide clear insights into the basis of recommendation results for both users and developers but also enhance patients' trust and satisfaction with the recommendation system. This study represents a significant step forward in the pursuit of personalized, transparent, and effective healthcare solutions.