Diagnostic Support Platform for the Identification of Pediatric Genetic Neurological Diseases Through a Machine Learning-Based Recommendation System

This study aims to evaluate the DIAGEN-IA diagnostic support platform, developed to assist in the identification of pediatric neurological diseases with a genetic basis. The primary objective is to assess whether the platform reduces the proportion of inappropriate referrals to clinical geneticists. Secondary objectives include improving the completeness of initial diagnostic evaluations, enhancing the quality of referral requests, and evaluating user satisfaction with the platform.

The study will be conducted at Carlos Van Buren Hospital in Valparaíso, a high-complexity hospital serving over 486,000 individuals. Using a prospective before-and-after design, the study is divided into two phases. The initial 6-month phase will collect baseline data on referrals, their appropriateness, and the completeness of initial diagnostic evaluations. Interobserver variability among geneticists will also be analyzed. In the 6-month intervention phase, healthcare professionals will use DIAGEN-IA during consultations, and the same outcomes will be reassessed.

Participants include healthcare providers from primary and secondary care centers who manage pediatric patients and are responsible for referring cases to clinical geneticists. Eligible participants must be Spanish-speaking professionals with advanced proficiency, working with children under 18 years old, and involved in diagnosing rare diseases. Data will be anonymized, and demographic information such as age, gender, specialty, years of practice, and specific training in genetics or metabolic disorders will be collected.

DIAGEN-IA is a platform co-designed with input from neuropediatricians and geneticists, integrating the Orphanet ontologies (ORDO, HPO, and HOOM) to ensure comprehensive diagnostic support. The application employs a Bayesian network model to recommend differential diagnoses and appropriate tests based on phenotypic characteristics. This AI-driven approach enables interpretable decision-making and models uncertainty inherent in rare disease diagnosis. The platform operates on a client-server architecture and supports seamless integration into clinical workflows.

Primary outcomes include the proportion of referrals deemed inappropriate, assessed using a 5-point Likert scale by clinical geneticists. Secondary outcomes focus on referral quality, evaluated against standardized criteria, and user satisfaction, measured with the validated CSQ-8 questionnaire. User satisfaction will be assessed at one and six months during the intervention phase. Results will inform strategies to optimize referral processes and diagnostic accuracy in pediatric care.