JMIR Publications

ABSTRACT

Background:

Virtual Patients (VPs) are digital simulations that can be used for practicing clinical reasoning (CR) in controlled educational environments. Traditional computer-based VP platforms often lack the authenticity and interactivity that is needed for effective CR skill training. Social robotic VPs combined with artificial intelligence (AI) have shown promise in enhancing realism and engagement. However, quantitative evidence comparing this emerging technology with conventional VP platforms within the context of CR training remains limited.

Objective:

The aim of this investigation was to quantitatively compare VP design elements that support CR training for medical students between an AI-enhanced social robotic VP platform and a conventional computer-based VP software, as self-reported by the students.

Methods:

We conducted a randomised cross-over trial involving 178 sixth-semester medical students at Karolinska Institutet, Stockholm, Sweden. Students experienced a social robotic VP platform with an integrated large language model to support dialogue (SARI) and a conventional computer-based VP platform (VIC) during CR training sessions. VP design was evaluated using a validated questionnaire. Student preferences for CR training were assessed using categorical responses and a visual analogue scale.

Results:

SARI demonstrated significant superiority over VIC across multiple domains of VP design. Students rated SARI higher for authenticity of patient encounters (median [IQR]: 4.0 [3.5–4.5] versus 3.0 [2.5–3.5]; p<0.001), professional approach during consultation (median [IQR]: 4.5 [4.0–4.8] versus 4.0 [3.5–4.5]; p<0.001), coaching during consultation (median [IQR]: 4.3 [4.0–4.7] versus 4.0 [3.7–4.7]; p<0.001), and learning effect of consultation (median [IQR]: 4.4 [4.0–5.0] versus 4.0 [3.5–4.5]; p<0.001), all on scales from 1 to 5. Students strongly favoured SARI over VIC for CR training (72% versus 14%; OR: 27.1; 95% CI: 14.3–53.7; p<0.001). These preferences were consistent in subgroup analyses of student groups of interest (female versus male students, students with or without prior VP experience, and students first introduced to SARI or VIC). Comparisons were made between SARI and VIC, SARI and equal preference, and SARI and not SARI (i.e., VIC and equal preference combined). The only comparisons not reaching statistical significance were among students with prior VP experience and those first introduced to VIC in the SARI versus not SARI comparison, although a numerical preference for SARI was still observed in these groups.

Conclusions:

AI-enhanced social robotic VPs appear to offer superior design characteristics and, in our investigation, were strongly favoured by medical students for CR training compared with conventional computer-based software. These results corroborate prior research supporting the integration of AI with social robotics as a promising approach for VP simulations aimed at preparing medical students for real clinical encounters.