Generative AI and Strategic Prompt Engineering in Emergency Care: A Multi-Center Randomized Controlled Trial with Natural Language Processing Validation in Indian Healthcare Settings

Background: Indian emergency medicine is now under severe shortage of physicians, face a surge of patients and broad language barriers which require new AI methods that respond in the local health environments. Study methods: In the study, we implemented a multi-center randomized controlled trial in several leading academic medical centers in India between 01 January 2023 and 31 December 2023. We did the study on 1,000 adult emergency department patients and assigned them to AI-assisted care (ChatGPT-4 with culturally-adapted prompt engineering) vs. standard care. Length of stay, adverse events at 30 days, and diagnostics accuracy were used as primary endpoints. The quality of AI-produced clinical summaries was measured with ROUGE, BLEU, LSA metrics and compared to the way it was documented by the physician. There was blindness to treatment in all the outcome assessors. Findings: Of 1,000 randomized individuals (500 AI-assisted, 500 standard care) non-inferiority of AI-assisted care was shown in diagnostic accuracy (AI-assisted care 94.8%; standard care 94.2%; difference 0.6%, 95% CI: -2.1 to 3.3), and AI-assisted care had a superior performance in length of stay (AI-assisted care median 3.1; standard care median 4.3 hours; difference -1.2 hours). The Natural language processing evaluation showed high agreement, ROUGE-L scores 0.862±0.11, ROUGE-2F scores 0.804±0.14, and 689/1,000 (68.9%) cases scored 0.85 or above. The use of AI-assisted care saved physicians 38 percent of documentation time (P<0.001), raised clinical guidelines compliance by 23 percent (P<0.001) and raised patient satisfaction ratings (8.6 vs. 7.8; P<0.001). The cost-effectiveness analysis displayed savings of 2,847 Indian rupees per patient. Conclusions: It was accomplished with excellent clinical results and outstanding financial cost savings, relative to the clinical outcomes, cultural adaptations of prompt engineering and AI-aided emergency care. These results confirm the use of AI nationwide in Indian emergency medicine.