Development and Validation of a Large Language Model–Based System for Medical History-Taking Training: Prospective Multicase Study on Evaluation Stability, Human-AI Consistency, and Transparency
Notice bibliographique
Résumé
Background: History-taking is crucial in medical training. However, current methods often lack consistent feedback and standardized evaluation and have limited access to standardized patient (SP) resources. Artificial intelligence (AI)-powered simulated patients offer a promising solution; however, challenges such as human-AI consistency, evaluation stability, and transparency remain underexplored in multicase clinical scenarios. Objective: This study aimed to develop and validate the AI-Powered Medical History-Taking Training and Evaluation System (AMTES), based on DeepSeek-V2.5 (DeepSeek), to assess its stability, human-AI consistency, and transparency in clinical scenarios with varying symptoms and difficulty levels. Methods: We developed AMTES, a system using multiple strategies to ensure dialog quality and automated assessment. A prospective study with 31 medical students evaluated AMTES's performance across 3 cases of varying complexity: a simple case (cough), a moderate case (frequent urination), and a complex case (abdominal pain). To validate our design, we conducted systematic baseline comparisons to measure the incremental improvements from each level of our design approach and tested the framework's generalizability by implementing it with an alternative large language model (LLM) Qwen-Max (Qwen AI; version 20250409), under a zero-modification condition. Results: A total of 31 students practiced with our AMTES. During the training, students generated 8606 questions across 93 history-taking sessions. AMTES achieved high dialog accuracy: 98.6% (SD 1.5%) for cough, 99.0% (SD 1.1%) for frequent urination, and 97.9% (SD 2.2%) for abdominal pain, with contextual appropriateness exceeding 99%. The system's automated assessments demonstrated exceptional stability and high human-AI consistency, supported by transparent, evidence-based rationales. Specifically, the coefficients of variation (CV) were low across total scores (0.87%-1.12%) and item-level scoring (0.55%-0.73%). Total score consistency was robust, with the intraclass correlation coefficients (ICCs) exceeding 0.923 across all scenarios, showing strong agreement. The item-level consistency was remarkably high, consistently above 95%, even for complex cases like abdominal pain (95.75% consistency). In systematic baseline comparisons, the fully-processed system improved ICCs from 0.414/0.500 to 0.923/0.972 (moderate and complex cases), with all CVs ≤1.2% across the 3 cases. A zero-modification implementation of our evaluation framework with an alternative LLM (Qwen-Max) achieved near-identical performance, with the item-level consistency rates over 94.5% and ICCs exceeding 0.89. Overall, 87% of students found AMTES helpful, and 83% expressed a desire to use it again in the future. Conclusions: Our data showed that AMTES demonstrates significant educational value through its LLM-based virtual SPs, which successfully provided authentic clinical dialogs with high response accuracy and delivered consistent, transparent educational feedback. Combined with strong user approval, these findings highlight AMTES's potential as a valuable, adaptable, and generalizable tool for medical history-taking training across various educational contexts.
Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.
Comment cette classification a été obtenuedéplier
Prédiction distillée sur la base complète
Imitation des enseignantsNi prévalence calibrée, ni vérité terrain. Validation humaine à venir. Apprise à partir de 10 348 étiquettes directes de Codex et de 10 348 étiquettes directes de Gemma. Le mode candidate est l'union des têtes enseignantes seuillées; le consensus est leur intersection. Ces sorties portent le statut machine_predicted_unvalidated et ne sont ni des étiquettes humaines ni des étiquettes directes de modèles de pointe.
Scores Codex et Gemma par catégorie
| Catégorie | Codex | Gemma |
|---|---|---|
| Métarecherche | 0,003 | 0,003 |
| Méta-épidémiologie (sens strict) | 0,000 | 0,000 |
| Méta-épidémiologie (sens large) | 0,000 | 0,000 |
| Bibliométrie | 0,000 | 0,000 |
| Études des sciences et des technologies | 0,000 | 0,000 |
| Communication savante | 0,000 | 0,000 |
| Science ouverte | 0,000 | 0,000 |
| Intégrité de la recherche | 0,000 | 0,000 |
| Charge utile insuffisante (le modèle a refusé de juger) | 0,000 | 0,000 |
Scores machine (provisoires)
Les deux têtes enseignantes du modèle étudiant, lues sur ce travail. Un score ordonne la base pour la relecture; il n'affirme jamais une catégorie, et le statut de validation accompagne chaque rangée tel quel.
Scores de référence d'un modèle non mature (critères de maturité non atteints, 7 itérations). Un score ordonne; il n'affirme jamais une catégorie.
score_only:v0-immature-baseline · tel quel depuis la passe de notation : score_only signifie que le nombre peut ordonner les travaux, et qu'aucune étiquette de catégorie n'en découleClassification
machine, non validéePrédiction automatique; un appel candidat d’une seule tête enseignante, pas un consensus.
Le détail, modèle par modèle et score par score, se trouve en fin de page sous « Comment cette classification a été obtenue ».