GPT-4-Based Virtual Patient Training for the PECARN Pediatric Head CT Rules

Background: Generative AI chatbots such as GPT-4 can role-play patients from an educator-written prompt, allowing learners to practice history-taking as if with a real patient. The PECARN head CT rules guide CT decisions in children with minor head trauma and require structured history-taking.

Objective: To compare GPT-4-based virtual patient training with conventional self-study for learning the PECARN head CT rules, and to examine whether the platform improves history-taking and communication skills.

Methods: Prospective, randomized, parallel-group pilot trial at Samsung Medical Center, Seoul, Korea. Licensed physicians with no experience as an emergency department attending physician or with less than 5 years of clinical experience were recruited through the hospital intranet and online physician communities. After written informed consent, participants drew a slip from a box to be allocated 1:1 to the control or virtual patient group. All participants completed a baseline knowledge checklist (10 short-answer and 10 case-based items) and a mini-clinical evaluation exercise (mini-CEX) on pediatric head trauma cases, including a clinical decision assessment of PECARN risk category and need for head CT. All then received a 5-minute lecture on the PECARN rules. The control group studied for up to 1 hour with the printed PECARN publication, a summary, and six training cases in print. The virtual patient group practiced for up to 1 hour on the Med2Lab platform, interviewing six GPT-4 virtual patients built on PECARN scenarios, with case review by a virtual instructor after each case. The knowledge checklist and a mini-CEX on a new case were repeated after training, followed by questionnaires on confidence and satisfaction; the virtual patient group also completed the System Usability Scale. The planned sample was 40 participants (45 allowing for dropout).

After a 5-minute lecture on the PECARN head CT rules, participants practiced history-taking for up to 1 hour on the Med2Lab platform with six GPT-4-based virtual patients representing pediatric head trauma cases, with case review by a virtual instructor after each case.

Behavioral: GPT-4-based virtual patient training

After the same 5-minute lecture on the PECARN head CT rules, participants studied for up to 1 hour with printed materials: the original PECARN publication, a summary, and the same six training cases in printed form.

Behavioral: Self-study with printed materials

Free-text history-taking practice with six GPT-4-based virtual patients on the Med2Lab platform (up to 1 hour). Each virtual patient is defined by an educator-written prompt describing a PECARN-based pediatric head trauma scenario and behavioral rules; after each case a virtual instructor reviews the case and the learner's clinical reasoning.

Self-directed study for up to 1 hour with printed materials: the original PECARN head injury prediction rule publication, a summary of the rules, and the same six pediatric head trauma training cases in printed form.

Plazo: Baseline and immediately after the training session (same day)

History-taking and clinical performance on a standardized pediatric head trauma case, rated by two assessors with the Royal Brisbane and Women's Hospital Emergency Department Mini-CEX form; the same assessors rate before and after training. Difference in change between groups.

Plazo: Baseline and immediately after the training session (same day)

Communication skills during the Mini-CEX case, rated with the communication skills assessment items (LCSAS) by the same assessors before and after training. Difference in change between groups.

Plazo: Baseline and immediately after the training session (same day)

Knowledge checklist on the PECARN head CT rules consisting of 10 short-answer items and 10 case-based decision items, administered before and after training. Difference in change between groups.

Plazo: Baseline and immediately after the training session (same day)

Correct determination of the PECARN risk category and of the need for head CT on the standardized pediatric head trauma case (clinical decision assessment), before and after training. Difference in change between groups.

Plazo: Immediately after the training session (same day)

Post-training questionnaire on confidence and satisfaction with the education, completed by both groups.

Plazo: Immediately after the training session (same day)

Usability of the Med2Lab virtual patient platform measured with the 10-item System Usability Scale (score 0-100; higher scores indicate better usability), virtual patient group only.

Evaluating the Effectiveness and Usability of History Taking and Clinical Decision Training With a ChatGPT-Based Virtual Patient Platform: A Randomized Pilot Trial