Asian Journal of Medical Principles and Clinical Practice

Clinical decision support systems (CDSS) have evolved substantially over the past two decades, driven by advances in artificial intelligence (AI), machine learning (ML), and natural language processing (NLP). In internal medicine, AI-enhanced CDSS offer the potential to support diagnostic reasoning, improve risk stratification, and inform therapeutic decision-making through the analysis of large and complex clinical datasets. This structured narrative review critically examines the evidence base for AI-enhanced CDSS, with particular attention to their diagnostic performance across major internal medicine subspecialties, including cardiovascular medicine, respiratory medicine, endocrinology, oncology, infectious disease, and nephrology. The review finds that AI systems have shown strong performance in selected, well-defined clinical tasks, including arrhythmia detection, cardiovascular risk prediction, diabetic retinopathy screening, sepsis prediction, and early identification of occult malignancy; however, this performance is often derived from retrospective, highly curated, or proof-of-concept settings rather than routine real-world deployment. Important limitations remain, including algorithmic bias, limited explainability, variable data quality, regulatory uncertainty, workflow integration challenges, and persistent concerns regarding clinician trust and accountability. The review also considers key ethical issues, including transparency, patient autonomy, health equity, and data governance. Emerging developments in federated learning, multimodal AI, large language models, and precision medicine may further expand the role of AI-enhanced CDSS in internal medicine, but their responsible adoption will require rigorous prospective validation, careful implementation, and sustained interdisciplinary collaboration. Overall, AI-enhanced CDSS should be understood not as replacements for clinical judgement, but as potentially valuable tools whose clinical impact will depend on the quality of the evidence, the safety of deployment, and the context in which they are used.