JMIR Publications

ABSTRACT

Background:

AI-supported serious games are increasingly proposed as scalable environments for clinical reasoning practice in nursing education, where opportunities for repeated, safe rehearsal are constrained. Unlike conventional serious games, which deliver pre-authored feedback tied to fixed decision points, large language models allow learners to interrogate their own decisions through open-ended dialogue. However, evidence regarding learning outcomes, student experiences, and implementation of serious games incorporating conversational AI remains limited, particularly across multiple educational settings.

Objective:

This study evaluated an AI-supported serious game for cardiovascular nursing care across 2 universities, examining pre-post changes in critical thinking dispositions, clinical decision-making, and case assessment performance, as well as student game experience, acceptability of the AI-supported debriefing, and implementation-related outcomes. Between-site differences in change were explored.

Methods:

A multicenter, single-arm pre-post study was conducted with second-year undergraduate nursing students at 2 public universities in Türkiye during the summer recess, outside any teaching or assessment relationship. Students completed a case-based game covering the nursing process for a patient with heart failure and coronary artery disease, concluding with a domain-constrained conversational agent that facilitated reflective debriefing. Selected game components covering patient communication, data collection, and nursing diagnosis underwent expert content validation before implementation (S-CVI/Ave=1.00 for relevance and clarity). Implementation-related outcomes included recruitment, completion, technical access failures, and voluntary re-engagement; acceptability was assessed using the Game Experience Scale (GAMEX) and a feedback form. Learning-related measures were analyzed using Wilcoxon signed-rank and Mann-Whitney U tests, with rank analysis of covariance as a post hoc sensitivity check. Reporting followed CHERRIES.

Results:

Of 260 students invited, 118 (45.4%) consented, and 108 (91.5%) completed the protocol (54 per site). Voluntary replay occurred in 25 of 108 participants (23.1%), and 107 of 108 participants (99.1%) reported willingness to replay. The mean total GAMEX score was 96.63 (SD 10.71), with no significant difference between institutions (Institution A: mean 95.44, SD 11.25; Institution B: mean 97.81, SD 10.11; P=.33). The mean rated usefulness of the AI-supported debriefing was 4.53 (SD 0.57) out of 5. Within- institution increases occurred at both institutions for critical thinking (Institution A: r=0.69; Institution B: r=0.29), clinical decision-making (r=0.74 and r=0.36, respectively), and case assessment performance (r=0.88 and r=0.96, respectively). Exploratory between-site comparisons of change differed for critical thinking (P=.005) and clinical decision-making (P<.001) but not case assessment performance (P=.53).

Conclusions:

The AI-supported serious game was successfully implemented across both universities and was positively received by students. Pre-post increases were observed in critical thinking, clinical decision-making, and case assessment performance; however, the single-arm design and baseline differences between institutions preclude causal attribution to the intervention. Future controlled studies should determine the educational contribution of the serious game and its AI-supported reflective component and examine how learner and institutional characteristics influence outcomes.