The ABCD of AI-Enabled Clinical Reasoning: A Guided Framework to Transform Learner Interactions and Decision Quality

To evaluate the efficacy of a structured guidance framework articulate, brainstorm and benchmark, critique and customize, and decide and discuss (ABCD) in improving AI-assisted clinical decision-making among surgical trainees.

We conducted a randomized controlled trial with 72 surgical residents, comparing the ABCD framework (intervention) against unguided AI use (control). All participants analyzed 8 standardized gastric cancer cases using an LLM (DeepSeek). Primary outcome was clinical decision quality assessed via expert-blind rating (Clinical Decision Quality Assessment Scale, CDQAS). Secondary outcomes included AI interaction process quality (AI-Clinical Interaction Process Quality Scale, ACPQS, evaluating 5 domains: Structured information articulation, deep inquiry and reasoning, critical appraisal, contextual integration, and interaction efficiency and focus), self-efficacy (AI-Assisted Learning Self-Efficacy Scale, AI-SE), objective behavioral metrics (evidence solicitation frequency [ESF] and "What-If" scenario prompting [WISP]), and knowledge integration (match to expert-derived key points, MEKP). We performed mediation analysis to elucidate the mechanism of effect.

The guided group demonstrated superior clinical decision quality (CDQAS: 16.2 vs 13.5, p < 0.05, d = 1.15), with significant gains in personalization and decision process completeness. The intervention fundamentally improved interaction quality (ACPQS: 20.8 vs 15.1, p < 0.05, d = 1.75), markedly increasing evidence solicitation (3.0 vs 0.7 instances/case, p < 0.05) and critical "what-if" questioning (2.0 vs. 0.8, p < 0.05). Self-efficacy in critical AI appraisal improved more in the intervention group (+8.5 vs +4.1, p < 0.05). Mediation analysis revealed that 88 % of the improvement in decision quality was mediated by the enhanced interaction process.

The ABCD framework effectively transforms AI from a passive information tool into a catalyst for disciplined clinical reasoning. By structuring interactions to mandate articulation, evidence-benchmarking, critique, and synthesis, it cultivates essential metacognitive skills, offering a practical, theory-driven model for the responsible integration of generative AI into surgical education.