Abstract
Objective structured clinical examination (OSCE) is widely used to assess medical students' clinical skills. Virtual OSCEs were used in place of in-person OSCEs during the COVID-19 pandemic; however, their reliability is yet to be robustly analyzed. By applying generalizability (G) theory, this study aimed to evaluate the reliability of a hybrid OSCE, which admixed in-person and online methods, and gain insights into improving OSCEs' reliability. During the 2020-2021 hybrid OSCEs, one examinee, one rater, and a vinyl mannequin for physical examination participated onsite, and a standardized simulated patient (SP) for medical interviewing and another rater joined online in one virtual breakout room on an audiovisual conferencing system. G-coefficients and 95% confidence intervals of the borderline score, namely border zone (BZ), under the standard 6-station, 2-rater, and 6-item setting were calculated. G-coefficients of in-person (2017-2019) and hybrid OSCEs (2020-2021) under the standard setting were estimated to be 0.624, 0.770, 0.782, 0.759, and 0.823, respectively. The BZ scores were estimated to be 2.43-3.57, 2.55-3.45, 2.59-3.41, 2.59-3.41, and 2.51-3.49, respectively, in the score range from 1 to 6. Although hybrid OSCEs showed reliability comparable to in-person OSCEs, they need further improvement as a very high-stakes examination. In addition to increasing clinical vignettes, having more proficient online/on-demand raters and/or online SPs for medical interviews could improve the reliability of OSCEs. Reliability can also be ensured through supplementary examination and by increasing the number of online raters for a small number of students within the BZs.