A Multilingual AI-Driven Conversational System for Cervical Cancer Awareness and Risk Assessment: Development and Evaluation Study
JMIR Cancer · 2026-11-06 · 12 · e103866
DOI: 10.2196/103866 ↗Abstract
Background Cervical cancer is the fourth most common cancer among women globally, with an estimated 660,000 new cases and 350,000 deaths in 2022. It imposes a disproportionate burden on underserved populations in low- and middle-income countries, where language barriers, cultural stigma, and limited access to health care hinder awareness and uptake of preventive measures. Objective This study aimed to design, develop, and evaluate a multilingual AI-driven dialogue system for cervical cancer awareness, personalized risk assessment, and health care facility navigation, targeting underserved populations through a culturally adaptive and privacy-preserving digital health platform. Methods A design science research framework guided the development of a 3-tier, cross-platform system comprising (1) a transparent, rule-based risk-stratification engine whose 30-item questionnaire mirrors the feature schema of the University of California, Irvine (UCI) Cervical Cancer Risk Factors dataset and whose weights encode established clinical risk factors drawn from World Health Organization, US Centers for Disease Control and Prevention (CDC), and American Cancer Society guidance; (2) a retrieval-augmented generation (RAG) chatbot powered by GPT-4o mini; and (3) a geospatial health care facility recommendation module. The system was implemented in TypeScript (NestJS) and Flutter (Google LLC) and deployed across Android, iOS, and web platforms with multilingual support for English, Spanish, Yoruba, and Swahili, of which English, Spanish, and Yoruba were included in the multilingual evaluation reported here. The risk engine was evaluated through clinical face-validity review of its factor weights; chatbot quality was assessed using retrieval precision, hallucination rate, large language model (LLM)–as-a-judge quality scoring, adversarial safety testing, and native-speaker rating of multilingual output. Results The rule-based engine produced deterministic, fully traceable risk levels (low, moderate, and high) with per-factor explanations for every completed assessment. The RAG chatbot achieved retrieval precision@3 of 0.88 and a 71.4% reduction in hallucination rate compared with a non-RAG baseline. Automated LLM-as-judge evaluation rated factual correctness at 4.3/5, and independent clinician review of 50 responses identified factual errors in 8% (4/50) of RAG-grounded outputs compared with 28% (14/50) for a non-RAG baseline (Cohen κ=0.84). The geospatial module achieved a specialty relevance precision of 0.87. Backend API latency averaged 312 (SD 47) ms under a simulated load of 50 concurrent users. Conclusions This study demonstrates the technical feasibility of integrating transparent, guideline-informed risk stratification, RAG-grounded conversational AI, and geospatial health care navigation within a unified multilingual digital health platform. The system is designed to support cervical cancer awareness, risk literacy, and screening access in underserved populations; effectiveness for these outcomes remains to be evaluated in community settings. Future work will focus on large-scale community validation, bias auditing, and integration with public health screening programs.
EUVIMED