Prospective assessment of AI-enabled cost savings, continuous support, and psychosocial enhancement (AI New Era Study).

Abstract

e13615 Background: Cancer patients (pts) need extensive primary care across their treatment journey, yet the world's primary health systems struggle with patient volume and resource scarcity. This study aims to assess the effectiveness of Artificial Intelligence (AI) medical service model in primary care and clinical service pre-screening, as well as their potential impact on improving the quality of medical services and reducing medical costs. Methods: A total of 17,856 medical consultation inquiries from 7,522 pts with various types of cancer have been collected and categorized. The study employed a novel AI-driven medical service model (based on large language model), amalgamating clinical guidelines with a local oncology database, to prospectively field questions from 454 cancer-afflicted pts. Clinicians evaluated the accuracy and comprehensiveness of the responses to basic medical issues, while pts assessed the intelligibility and satisfaction. Results: Treatment-related inquiries constituted 34.44% (n = 6,149) of the total, with 81.53% of pts asking such questions. Basic medical issues were addressed in 65.56% (n = 11,707) of inquiries, raised by 59.78% of pts. Specific concerns within basic medical issues included test report explanations (20.45%), adverse reaction consultations (19.82%), and medication information requests (17.66%). In light of the prevalence of basic medical issues, the study prospectively explores the potential of AI technology to revolutionize medical models, driving the advent of a new era in healthcare research. As result, the AI model exhibited high performance across several metrics: accuracy (96.87 ± 10.29), comprehensiveness (97.40 ± 9.91), intelligibility (95.72 ± 12.62), and satisfaction (96.34 ± 11.03). Notably, the model demonstrated a proficiency in identifying queries necessitating physician expertise, with a 95.60% accuracy rate upon review. A survey of 50 outpatients indicated an enhancement in service speed (7.78%) and patient satisfaction (14.29%). This increase in satisfaction was primarily ascribed to pts' heightened sense of control and safety. When benchmarked against GPT-4, the AI model's overall performance was significantly superior (95.08 ± 10.05 vs 89.70 ± 12.85, p<0.01), particularly in domestic drug knowledge and gene report analysis. Conclusions: Largely fueled by the advanced capabilities of large language model, AI has the potential to profoundly reshape the entire patient care journey. Our AI medical service model efficiently meets key healthcare needs, addressing over half of clinical issues and cutting costs. It also guides pts to the right specialists for timely care. Early surveys indicate better primary care and improved doctor-patient relations. Further research is needed to understand AI's effects on fairness, access, and outcomes across varied demographic and socioeconomic strata.