WCN25-606 Comparing Clinical Trial Pre-Screening “AI vs Nephrologist”

Methods: A WhatsApp-based interactive platform allowed patients to converse using text and voice in multiple languages.The AI system involves:-A curated knowledge base of detailed information on managing symptoms, patient education, dialysis-friendly recipes, and patients' clinical profiles.-A Retrieval Augmented Generation framework leveraging state-ofthe-art language models to provide personalized responses.-Implemented guardrails to eliminate AI hallucinations, ensuring reliability.-A model to classify food images providing real-time nutrient information and personalized recommendations based on patients' unique clinical profile.-A clinical module to answer basic questions about drug interactions and side effects.The tool was made available to End Stage Kidney Disease (ESKD) patients undergoing hemodialysis across 360 centers in India.User engagement and conversation data were analyzed over an 8-week period (June 25 -August 19, 2024) to categorize interaction levels and identify usage patterns.Text embeddings and BIRCH clustering were used to identify patterns in patient conversations.Results: Over the study period, 2,142 unique users from 360 dialysis centres engaged with the platform, generating 4,528 sessions and 9,734 messages.35% of users were responsible for 76% of all the messages, indicating high engagement among a core user group.They demonstrated progression from initial interaction to deep, sustained usage.Temporal analysis of the top 10 most engaged users showed consistent interaction.Engagement was in English (79.2%) and Hindi (16.76%), with w4% in other 7 regional languages.Patients from 26 Indian states engaged, indicating widespread adoption.Topic clustering revealed 4 clusters: Administrative (11%), Medical (22%), Diet-related (34%), and Others (32%), highlighting versatility.Around 219 food images were uploaded by 158 users and analyzed to identify food items and give personalized recommendations.Independent manual validation of the food image analysis found >98% accuracy.A sample image analysis is attached.Conclusions: The tool has proven to be a highly effective AI-powered health coach, successfully engaging ESKD patients across India.Ubiquitous availability on WhatsApp and its multilingual nature confer easy access to patients across the socio-economic spectrum by offering continuous guidance at minimal cost.By addressing varied needs and providing highly accurate food image analysis, it demonstrates reliability in providing personalized support.This tool has significant potential in improving health outcomes for ESKD patients and reducing burden on healthcare systems, particularly in resource-constrained environments.I have no potential conflict of interest to disclose.I used generative AI and AI-assisted technologies in the writing process.