Traditionally, peer assessment in medical training has been used to obtain quantitative information on a trainee’s performance, with examples including peer ranking and peer rating [1]. In recent years, peer assessment has been considered to be an effective learning strategy [2]. In a study of objective structured clinical examinations (OSCEs), peer assessment was found to be beneficial for learners, who received high-quality feedback and experienced stress reduction, and for peer assessors (PAs) in terms of improved teaching and clinical skills [3]. Basic life support (BLS) skills are a key element of the undergraduate curriculum that medical students should acquire. Peer-led BLS training for medical students provided high-quality education that was as effective as professional training [4]. In studies of peer assessment in BLS training, senior medical students were able to make reliable assessments of their peers’ performance [5]. The interrater reliability between professional faculty assessors (FAs) and PAs has likewise been shown to be good [6]. However, quite a few studies of peer assessment in skills training were limited to BLS. Few studies have been conducted on peer assessment of advanced resuscitation skills, such as tracheal intubation and manual defibrillation, among medical students. To introduce peer assessment for educational purposes, its reliability and validity should be established, but studies have not provided sufficient psychometric data [7].

The objective of this study was to examine the agreement level between medical students’ peer assessments and a professional faculty member’s assessments in an advanced resuscitation skills examination. Thus, the research question of this study was as follows: How reliable is the peer assessment of medical students in an advanced resuscitation skills examination compared to that of professional faculty?

This study was approved by the institutional review board of Dong-A University (IRB approval no., 2-1040709-AB-N-01-202101-HR-003-02). The requirement to obtain informed consent was waived because participants took part in the activity as part of their educational curriculum.

This was a retrospective exploratory study aiming to determine the interrater reliability of peer assessment.

In total, 54 fourth-year medical students who attended the emergency medicine (EM) clinical clerkship course were included in this study as PAs. They were divided into 8 groups (6–7 students per group) and participated in the course every 2 weeks from April to July 2020. A professional faculty member who was a qualified advanced life support instructor and emergency physician participated as an FA.

The measurement variables were the BLS skills, tracheal intubation skills, and manual defibrillation skills assessed by peers and faculty.

We used the paired t-test to compare scores between the FA and PAs and used the intraclass correlation coefficient (ICC) to analyze interrater reliability. ICC estimates and their 95% confidence intervals (CIs) were calculated using IBM SPSS ver. 25.0 (IBM Corp., Armonk, NY, USA) based on a 1-way random effects model because each subject was rated by a different set of raters (44 different FA-PA pairs). Values less than 0.5 are indicative of poor reliability, values between 0.5 and 0.75 indicate moderate reliability, values between 0.75 and 0.9 indicate good reliability, and values greater than 0.90 indicate excellent reliability [10]. Bland-Altman plots were used to graphically evaluate the agreement of FA and PA scores by plotting the differences between the FA and PA scores against the FA scores using MedCalc ver. 19.6 (MedCalc Software Ltd., Ostend, Belgium). We used Gwet’s first-order agreement coefficient (AC1) to assess the agreement of individual checklist items between FA and PAs using R ver. 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria; 2020) with the ‘rel’ package version 1.4.2 [11]. In the data processing, if either the FA or PA scores was missing, the other was also considered to be missing. A P-value of <0.05 was considered to indicate statistical significance. The R code is available in Supplement 1.

The agreement for the overall checklist items was poor to moderate (ICC, 0.55; 95% CI, 0.31 to 0.73; P<0.01). Likewise, poor to moderate agreement was also found for BLS, tracheal intubation, and manual defibrillation (Table 2, Dataset 2). The Bland-Altman plot revealed that the 95% limits of agreement were -1.7 to 6.0 points (Fig. 2B).

The agreement for the global rating was poor (ICC, 0.10; 95% CI, -0.20 to 0.38; P<0.25). The Bland-Altman plot revealed that the 95% limits of agreement were -1.1 to 3.2 points, and the PA score was lower than the FA score in only 1 case (Fig. 2C).

The agreement was highest for checklist item 1 (checking responsiveness), item 4 (starting chest compressions immediately after checking the pulse), and item 8 (selecting the appropriate size of a laryngoscope blade for an adult) with a Gwet’s AC1 of 1.0. However, the agreement was poor for checklist items 13 (ventilating the patient at an appropriate rate using bag-mask ventilation) and 14 (ventilating with an appropriate tidal volume) (Table 1, Fig. 3).

Students gave significantly higher scores than the FA on both the checklist scale and global rating. The overall agreement of the assessments by medical students was poor to moderate, and the ventilation-related items had the poorest agreement.

Students seemed to acknowledge their peers’ partially performed skills as performed skills on the binary checklist consisting only of “performed (yes)” or “not performed (no).” In contrast, the faculty member seemed to have strict standards for partially performed skills. The items related to quantitative parameters such as rates and amounts had poor agreement, such as checking the pulse for 5–10 seconds, a compression rate of 100–120 per minute, compression depth between 5 and 6 cm, a ventilation rate of 10 per minute, and 1/3–1/2 the total volume of the ventilation bag. The global ratings of proficiency given by the PAs were not reliable at all. Proficiency implies a thorough competence derived from training and practice. The criteria of proficiency assessment require a qualitative judgments and could be more subjective than quantitative judgments. Students seem to have been generous with their qualitative judgments of the global rating. The poor agreement on some BLS- and ventilation-related items is considered to have been a technical problem caused by the absence of quantitative measuring equipment.

Most peer assessment studies in OSCEs involving resuscitation skills were limited to BLS without tracheal intubation and manual defibrillation [5]. In a study of 9 peer/faculty pairs assessing the BLS of 162 medical students, interobserver agreement was high (>95%). Unlike our study, students who participated as PAs in that study attended a 1-day course given by a European Resuscitation Council BLS instructor and worked with experienced faculty as BLS assessors for 4 hours to consolidate their assessment skills. This implies that PAs should be trained, and that at least 1 day of training is required to train a reliable PA for resuscitation skills assessments. Additionally, the use of devices that objectively measure compression depth and rate, ventilation volume, and similar parameters may improve agreement between FAs and PAs.

According to a review by Khan et al. [12], PAs provided higher global rating scores with moderate to high agreement, variable checklist scores depending on the station, and valuable feedback compared with FAs. The finding that PAs assigned higher global rating scores is consistent with this study, while the finding regarding agreement was not. The 3 main contributions to the reliability of peer assessment are the observed number of relevant performances, the number of peers involved, and the number of competencies being evaluated. Increasing the number of evaluators has a smaller impact on reliability than increasing the number of observations [13]. The poor agreement of global ratings in this study seems to be an issue relating to the aspects of competence being evaluated, rather than a problem of assessor training.

Finn and Garner [14] outlined tips for implementing successful peer assessments, which include considering the relevance of the curriculum, reflective learning, support, and resource allocation in planning. The implementation of peer assessment appears to require a more comprehensive approach, as well as assessor training.

There are several limitations to this study. First, the encounters observed by assessors were not sufficient. Due to coronavirus disease 2019-related social distancing measures, students were not allowed to concentrate in a small OSCE station as PAs. Thus, each student had only the opportunity to observe and assess 1 other student. This is thought to be a major cause of the poor observed agreement. Second, we did not investigate the impact of peer assessment on learning in OSCEs. One of the advantages of peer assessment is learning from peers and the assessment process. This requires students to carefully examine their peers’ performance according to specific criteria. This activity also helps them diversify their approaches and strategies in learning and understanding high- or lower-quality performance. Therefore, students can better understand their learning through insights into performance quality [15]. However, this study did not explore the effects of peer assessment on learning outcomes in depth. The impact of peer assessment on students’ learning will require long-term, qualitative follow-up. In principle, students who show a high level of agreement with an FA may evaluate on behalf of the FA. However, the present study did not analyze this possibility, and further research is needed.

Senior medical students showed unreliable agreement in peer assessments of advanced resuscitation skills compared to the faculty member’s assessments. Students were more generous in global ratings and showed variable scores on the checklist scale compared with faculty assessments. If a peer assessment is planned in skills education, comprehensive preparation and sufficient assessor training should be carried out in advance.