The goal of this clinical trial is to learn whether access to an artificial intelligence (AI) clinical decision support assistant can improve diagnostic accuracy during real-world telemedicine consultations among primary care physicians in El Salvador. The main questions it aims to answer are: * Does access to the AI assistant increase the proportion of correct diagnoses compared to telemedicine without AI assistance? * Does the effect of the AI assistant differ according to the physician's prior experience using AI in telemedicine? Researchers will compare physicians with the AI assistant enabled to physicians with the AI assistant temporarily disabled to see if access to AI improves diagnostic accuracy. Participants (physicians) will: * Provide telemedicine consultations as part of their routine clinical duties. * Be randomly assigned to either have the AI assistant enabled or disabled during the study period. * Continue documenting clinical encounters in the electronic platform as usual. * Have their anonymized consultation notes reviewed by an independent expert panel to determine diagnostic accuracy.
Study Type
INTERVENTIONAL
Allocation
RANDOMIZED
Purpose
DIAGNOSTIC
Masking
DOUBLE
Enrollment
180
An AI tool integrated into the telemedicine platform, built on Google's Gemini Large Language Models (LLMs). The system operates via two modules: (1) a clinical history assistant that supports structured documentation of patient information in real-time and (2) a pre-diagnosis tool that analyzes documented clinical data to generate differential diagnosis suggestions for the physician's consideration. The model uses contextual prompting to ensure suggestions are culturally and clinically appropriate for El Salvador.
Standard primary care consultation via videocall without the assistance of artificial intelligence tools. Physicians rely solely on their own clinical judgment and manual documentation without automated summaries or diagnostic prompts.
Hospital Nacional El Salvador
San Salvador, El Salvador
Diagnostic Accuracy
The proportion of consultations where the primary diagnosis recorded by the participating physician matches the "gold standard" reference diagnosis. The reference diagnosis is established by a panel of three independent, blinded expert evaluators reviewing the anonymized clinical notes. A diagnosis is considered "correct" (value = 1) if it matches the reference diagnosis within the same clinically equivalent diagnostic group; otherwise, it is considered "incorrect" (value = 0). The analysis will compare the proportion of correct diagnoses between the AI-enabled and AI-disabled arms.
Time frame: Through study completion, ~ 12-16 weeks
Diagnostic Concordance
The level of agreement between the physician's diagnosis and the expert reference diagnosis, measured using Cohen's Kappa coefficient. This measure evaluates the reliability of the diagnoses beyond simple percentage agreement, accounting for agreement occurring by chance.
Time frame: Through study completion, ~12-16 weeks
Diagnostic Accuracy Stratified by Physician Experience Level
Evaluation of diagnostic accuracy (proportion of correct diagnoses) compared between subgroups of physicians with "High Experience" (≥1 year in the program or ≥20 consultations) versus "Low Experience" (\<1 year in the program or \<20 consultations).
Time frame: Through study completion, ~12-16 weeks
Diagnostic Accuracy Stratified by Clinical System
The proportion of correct diagnoses stratified by the physiological system of the pathology: Respiratory, Digestive, Urinary, or Ophthalmic. This outcome assesses if the AI's performance or utility varies depending on the specific type of clinical condition.
Time frame: Through study completion, ~12-16 weeks.
This platform is for informational purposes only and does not constitute medical advice. Always consult a qualified healthcare professional.