A modified murtagh-style safe diagnostic rubric for general practice training: preliminary validity evidence from a real-world implementation study

BACKGROUND: Safe diagnostic reasoning is a central but challenging competency in general practice training, particularly in frontline settings where information is limited, uncertainty is common, and diagnostic error has direct implications for patient safety. Although Murtagh's diagnostic framework provides a clinically intuitive structure, its educational application and structured assessment in lower-resource, non-English-speaking training environments remain underexplored. The objective of this study was to evaluate a modified Murtagh-style safe diagnostic rubric (MM-SAFE-Dx) and its application within a general practice training context. METHODS: This study was conducted as a single-centre retrospective real-world implementation study at The First Affiliated Hospital of Harbin Medical University in Northeast China. In 2025, a modified Murtagh-style safe diagnostic training programme was delivered across three voluntary rounds. Eighty-three general practice trainees and residents were included, contributing 145 scored encounters. Baseline and follow-up case-based assessments were analyzed using nonparametric methods. Preliminary real-world validity evidence for the MM-SAFE-Dx rubric was examined in terms of internal consistency, baseline-stratified discrimination, responsiveness, context sensitivity, and practical scoring stability. RESULTS: The overall educational signal was modest but directionally favourable. When separating first-exposure from repeated-use effects, baseline to first exposure showed minimal change, whereas subsequent rounds demonstrated more coherent directional improvements consistent with cumulative learning processes, although not all comparisons reached statistical significance. Sensitivity analyses further indicated that uncorrected baseline stratification yielded counterintuitive patterns dominated by short-term effects among lower-performing learners, whereas longitudinally informed stratification produced more interpretable results. These findings suggest that baseline measurements derived from first-time exposure to a complex diagnostic rubric may not reliably reflect underlying diagnostic competence. CONCLUSIONS: In this low-resource, heterogeneous general practice training environment, the MM-SAFE-Dx rubric demonstrated preliminary real-world validity and practical educational utility. Beyond evaluating a specific programme, this study highlights a broader methodological issue: structured diagnostic tools introduced into unfamiliar and resource-variable settings may require both guided implementation and context-sensitive analytical calibration before their educational effect can be meaningfully interpreted.