Using Student-Reported Digital Daily Logs and Human-Verified AI-Assisted Reflection Coding to Describe Clinical...

In Japan, the 2021 revision to the Medical Practitioners Act formally supported students' participation in supervised clinical practice. However, monitoring day-to-day clinical experiences across community-based clerkship sites remains challenging. Student-reported digital daily logs and reflection sheets provide complementary descriptive information across such sites.

This study aimed to investigate whether a digital educational data workflow, combining student-reported daily clinical activity logs with human-verified artificial intelligence (AI)-assisted reflection coding, can be used to describe clinical learning experiences across distributed community-based clerkship sites.

In this exploratory observational study, we explored variations across placement contexts, including whether sites included community-based attending physicians (CAPs) trained in the university faculty development (FD) program. Data from 112 fifth-year medical students who had completed 3-week community-based clinical clerkships at sites with trained CAPs (n=34) or at sites without trained CAPs (n=78) were analyzed. Students recorded daily experiences of 50 Model Core Curriculum (MCC)-based clinical activities using a Google Spreadsheet-based digital daily log. The primary outcome was the number of student-reported examination-based, lower-risk clinical activity types recorded as performed at least once. Log completion, defined as the presence of any substantive daily log entry, was assessed across eligible student-days, and a classification-independent sensitivity analysis examined all 50 activity items. Reflection sheets were deductively coded according to MCC learning objectives using a human-verified AI-assisted preliminary sorting and candidate-coding workflow. Furthermore, exploratory sensitivity analyses examined the role of the placement context.

Daily log entries were available for 1491 (99.6%) of 1497 eligible student-days. Students at sites with trained CAPs had more student-reported examination-based, lower-risk clinical activity types recorded as performed at least once than those at sites without trained CAPs. No statistically significant difference was observed for basic clinical and emergency procedures. The classification-independent analysis across all 50 items showed a similar pattern. Reflection sheet analysis yielded 4097 textual units. Placement category was closely associated with medically underserved placement, home visit exposure, and baseline rural self-efficacy.

The digital workflow yielded nearly complete student-reported daily data and described variation in clinical experiences across clerkship sites. These differences should be interpreted as variation across placement contexts rather than evidence of an independent effect of supervision by trained CAPs. Reflection sheets provided complementary descriptive information, and the AI component required human verification.