| Sumario: | This systematic review synthesizes evidence from 296 core Web of Science publications to evaluate ChatGPT's utility across modern medicine through a three-tiered framework: (1) Data Management (e.g., automating medical information extraction, enhancing electronic health record (EHR) integration, and optimizing clinical documentation workflows), (2) Clinical Decision Support (demonstrating capability in diagnostic accuracy and effective treatment), and (3) Therapeutic Support & Patient Services (facilitating personalized patient education via multilingual interaction and accelerating drug discovery). Additionally, dialogue-based AI frameworks, which prioritize communication efficacy over diagnostic performance, show particular promise in physiotherapy contexts. While ChatGPT demonstrates potential in enhancing operational efficiency, diagnostic speed, and personalizing patient management, key challenges still persist: data inaccuracies (e.g., hallucinated references), algorithmic biases (e.g., gender or racial disparities in treatment suggestions), insufficient clinical validation (overreliance on pattern recognition without physiological causality), communication barriers (misinterpretation of nuanced symptoms), ethical risks (privacy breaches in unsecured deployments) and other related issues. Successful implementation requires prioritizing five critical research avenues: (1) enhancing multimodal data analysis with diverse, domain-specific training datasets; (2) advancing explainable AI (XAI) techniques to improve algorithmic transparency/ clinician trust; (3) conducting multi-specialty validation through randomized controlled trials (RCTs) assessing safety and cost-effectiveness in real-world settings; (4) establishing adaptive ethical frameworks balancing innovation with patient rights, fairness, and harm prevention; (5) optimizing Traditional Chinese Medicine (TCM) applications via syndrome differentiation for complex comorbidities.
|
|---|