JMIR Publications

ABSTRACT

Background:

Videolaryngoscopy (VL) is increasingly used for emergency intubation, and simulation-based training (SBT) is an effective approach to training the skill. SBT typically requires an instructor for real-time feedback, which limits scalability in settings where faculty time is scarce. Self-directed learning supported by artificial intelligence (AI) guidance modules built into VL devices could extend SBT to such settings, but has not been directly compared with instructor coaching within an SBT curriculum.

Objective:

We compared AI-assisted self-directed and instructor-guided SBT for VL intubation on time to the Minimum Passing Standard (MPS), immediate performance, one-week retention, and learner experience.

Methods:

This single-center mixed-methods randomized study enrolled 122 medical interns with no prior VL experience, who were randomized 1:1 to instructor-guided training or AI-assisted self-directed training using a VL device's on-screen guidance module. In both arms, participants practiced until they declared readiness, at which point feedback was discontinued and they performed two consecutive testing attempts; the MPS required first-pass success in under 60 seconds without critical errors. Time to MPS was the cumulative practice time from the first practice attempt to the readiness declaration. The 122 post-training interviews underwent reflexive thematic analysis.

Results:

All 122 participants met the MPS without reaching the practice cap. Time to MPS did not differ between arms (medians 194 vs 203 seconds; 95% CI for between-arm difference −37 to +45 seconds; P>.99). At short-term (one-week) retention (76% follow-up), first-pass success was numerically higher with instructor guidance (80.4% vs 68.1%; absolute difference +12.3%, 95% CI −5.2% to +29.9%; P=.24). Six themes emerged: in both arms the videolaryngoscope screen itself was the main teacher; feedback mostly confirmed rather than corrected what learners were doing; only the instructor arm named errors with their clinical consequences; AI-arm participants met recognition lag, false negatives, and conflicting cues and disengaged from the module; the instructor arm added encouragement and reassurance the AI did not; and in both arms confidence stayed bounded by the gap between manikin and patient.

Conclusions:

AI-assisted and instructor-guided SBT produced similar time to MPS on a high-fidelity simulator. A nonsignificant short-term retention trend favored instructor guidance, which the interviews linked to the instructor's corrective feedback and reassurance — features the AI did not provide.