Managing Safety in Teaching Clinical Skills: Investigating Framing Errors in Cardiopulmonary Resuscitation Training through a Multi-Arm Randomized Controlled Equivalence Study

Introduction

Medical errors are a central concern for patient safety [1]. Over the past decades, extensive efforts have focused on safety management and the role of errors in improving patient outcomes [1–5]. For instance, Donchin et al. [4] investigated human errors in the ICU using a human factors engineering approach. Their findings suggest that strategies grounded in human factors principles can reduce error rates in intensive care settings, thereby enhancing patient safety.

Safety management encompasses the prevention of undesirable events, such as accidents or incidents, and involves regulatory or control mechanisms [6]. It is closely linked to an organization’s safety culture and incorporates both organizational and behavioral components of systems and processes [7, 8]. Despite the importance of medical error culture, medical students and trainees are often overlooked, even though they represent a key group for instilling a culture of error awareness across an organization [9, 10]. In Germany alone, approximately 10,000 medical students graduate annually and enter the healthcare system, highlighting the importance of targeting this population [11]. Moreover, recent research indicates that German hospitals still have room for improvement in fostering a robust error culture [12]. Therefore, integrating error-focused education early in medical training could significantly strengthen organizational safety culture.

This study examines two common approaches to handling errors in professional practice. First, error avoidance (EA) emphasizes minimizing errors whenever possible [13]. Within the EA framework, errors are often viewed as unnecessary for learning [14]. Given the potentially severe and even life-threatening consequences of medical errors, the healthcare system naturally prioritizes error reduction to protect patient safety [1, 15, 16]. Second, error management (EM) encourages a constructive view of errors, promoting them as learning opportunities while actively supporting learners to engage with mistakes [14, 17]. From the EM perspective, errors contribute to personal learning and can enhance skill acquisition [14, 17]. While errors may cause temporary frustration, they frequently provide valuable insights that help individuals handle similar challenges in the future [18]. EM, therefore, focuses on mitigating adverse outcomes of errors rather than eliminating errors themselves [13, 19]. EM has been shown to foster organizational learning, innovation, and psychological safety by enabling team members to communicate errors openly and effectively [3, 20, 21].

The observed advantages of EM compared with EA can be framed within a theoretical context [22], which serves as the foundation for this study. Our framework draws on motivation theory, specifically John W. Atkinson’s choice under risk model [23], a key theory addressing the need for achievement. According to this model, the need for achievement has two components: the drive to succeed and the drive to avoid failure. In addition to individual traits, these components are shaped by task characteristics such as difficulty and the perceived value of success or failure. The interplay between these factors influences behavior in achievement situations, determining whether an individual approaches a task with optimism or retreats in fear.

Regarding task framing—the focus of our intervention—additional insight comes from goal framing theory [24], which emphasizes the role of social context in shaping motivation. In other words, a person’s behavior in a given situation is affected by the ‘mindset’ activated at the moment. In this study, the EM or EA instruction was used to shape this mindset. Based on this framework, we hypothesized that EM framing would enhance motivation to achieve, whereas EA framing would increase motivation to avoid failure.

Performance outcomes of error avoidance (EA) and error management (EM) strategies have been explored in various contexts, such as software training [5, 17, 18, 25–27]. However, research on these approaches within medical practice is limited [14, 28], and existing results are often inconsistent [29]. In particular, it remains unclear whether EM-guided training might increase the frequency of performance errors; in other words, it still needs to be demonstrated that EM does not negatively affect learning outcomes. Establishing such equivalence could also reassure practitioners who are cautious about adopting EM due to concerns that learners might internalize incorrect knowledge or develop unsafe patient-handling habits [10, 30]. Medical simulation provides an ideal setting for this investigation, as error-handling strategies can be safely practiced without endangering patient safety [31].

To address this gap in the medical domain, we conducted a study examining the training of medical students in emergency skills. Sudden cardiac arrest is among the leading causes of death worldwide [32, 33], and basic life support (BLS) represents the critical first-line response. Mastery of cardiopulmonary resuscitation (CPR) is a fundamental clinical competency, vital for ensuring patient safety. Because the quality of CPR directly affects survival outcomes, it is arguably one of the most essential practical skills in clinical training [34]. However, frequent errors during CPR can significantly reduce patient survival rates [35–37]. Common performance mistakes include compressions that are too shallow or performed at an incorrect rate [38]. While prior research has examined approaches to teaching high-quality CPR, focusing on feedback methods and training designs [39–42], the influence of different error-framing strategies on CPR skill acquisition remains underexplored.

As previously discussed, EM offers several advantages over EA in terms of long-term benefits for psychological safety and organizational error culture. Nevertheless, its impact on short-term performance has not been fully established. In this study, we evaluated the CPR performance of undergraduate medical students following training with either EM or EA instructions. Our primary goal was to determine whether EM-guided training yields performance outcomes comparable to those of EA-guided training.

Materials and Methods

Trial registration

The study was officially registered under the identifier DRKS00029981 at https://www.drks.de.

Ethics

Approval for the study was obtained from the Institutional Ethics Review Board of the University Hospital, RWTH Aachen University (document EK-22-290) on Sept 8, 2022, in compliance with the ethical standards outlined in the Declaration of Helsinki [43] by the World Medical Association.

Study design

This investigation was structured as a multi-arm, parallel-group randomized controlled trial, comprising three distinct experimental conditions: (1) basic life support (BLS) training with explicit error management (EM) instructions, (2) BLS training guided by error avoidance (EA) instructions, and (3) standard BLS training without additional error-focused instructions (control group) (Figure 1). The study followed both the CONSORT guidelines for multi-arm parallel-group randomized trials and the CONSORT standards for equivalence trials [44–46].

We opted for an equivalence trial rather than a superiority trial for several reasons. Error management (EM) has been shown to enhance learning behaviors, facilitate organizational learning, and mitigate the negative consequences of mistakes. It also fosters psychological safety and encourages open communication within teams [3, 20, 21]. Such behaviors are linked to higher rates of error reporting, which in turn can improve patient safety among nurses [47]. However, while EM’s advantages are well-established in clinical practice, it remains unclear whether EM and error avoidance (EA) yield equivalent outcomes in medical education. Potential risks, including learners adopting incorrect knowledge and ethical concerns, warrant careful evaluation. Equivalence trials are designed to assess whether two interventions produce comparable effects on a targeted outcome [48]. The primary goal of this study was to determine if EM and EA produce similar results in CPR performance. These trial designs are beneficial in pragmatic or applied clinical research [49], especially when one intervention (here, EM) has recognized benefits. Still, the interest lies in confirming that it performs no worse than the alternative regarding a key outcome (in this case, performance during training). By evaluating these effects, we aim to inform discussions on the effective integration of EM into medical education and its broader influence on cultivating a positive error culture and enhancing patient safety.

Participants

The study enrolled first-year undergraduate students from medicine, dentistry, physiotherapy, and midwifery programs. Data collection occurred during a mandatory introductory emergency medicine course between October 12th and 27th, 2022. All participants provided written informed consent. No exclusion criteria were applied to maintain high ecological validity.

Sample size determination

Sample size calculations for equivalence testing followed the method described by Blackwelder [50], using the Sealed Envelope power calculator [51]. Assuming an α level of 0.05, 80% power, and an expected success rate of 46% based on prior institutional data, the required sample size was 230 participants (115 per group) for each two-group comparison. These targets were achieved across all study comparisons.

Randomization

Participants were assigned to study conditions (EM, EA, or control) in clusters of 12 using the Research Randomizer tool (https://www.randomizer.org/). Each session was conducted in a designated training room, and participants were placed in the rooms according to the randomization schedule. No cross-allocation or switching between groups occurred during the study.

Intervention

All participants underwent interprofessional BLS training following the Peyton 4-step teaching model [52]. In the first phase (demonstration), the tutor performed BLS at a normal pace without commentary. During the second phase (deconstruction), the tutor repeated the procedure, explaining each critical component of the BLS steps. In the third phase (comprehension), participants took an active role by instructing the tutor on the correct performance of BLS, with the tutor intervening only when prompted. In the final phase (performance), participants independently executed BLS on a ResusciAnne™ manikin while receiving guidance and feedback from a trained tutor. The manikin, a standard resuscitation training model, features a torso, head, and limbs, providing realistic resistance for chest compressions, with an audible click signaling correct compression depth.

Additionally, participants used a feedback device displaying real-time compression rate, depth, and full chest release. Steps 1–3 were conducted in a plenary session of 36 participants in a lecture hall. At the same time, the performance phase involved three smaller groups (EM, EA, control) of 12 participants, each assigned to an individual training room according to randomization.

Before the performance phase, the intervention groups (EM and EA) received standardized error-handling instructions via a purpose-designed video. Key points were reinforced with posters displayed in the training rooms (Supplementary Material). The content, informed by literature and adapted for the BLS context [5, 18, 25–27, 53], differed between groups: the EM group was taught to view errors as positive and integral to learning (e.g., “Errors are expected while learning the resuscitation algorithm and are an essential part of mastering the skills”), whereas the EA group was instructed to minimize errors proactively (e.g., “Try to avoid mistakes during training; think ahead about how to prevent them”). The control group proceeded directly to the Performance phase without additional error-focused instructions.

Primary outcome: CPR skill assessment

The primary outcome was participants’ CPR performance, evaluated immediately before (t0) and one week after (t1) training using the ResusciAnne™ manikin (Laerdal, Stavanger, Norway). During the assessment, no feedback device or audible click was provided. Participants received standardized instructions to perform resuscitation based on prior knowledge (t0) or newly acquired skills (t1). Tutors followed strict guidelines to avoid providing feedback beyond the initial instruction: “Please imagine that you see an unconscious person lying on the ground and come to help. Pretend that I am not there.” The scenario ended two minutes after the first chest compression to ensure uniform compression times across participants. CPR performance metrics included compression depth (CD) and compression rate (CR), which were recorded using the Laerdal PC Skill Reporting System Software (Version 2.4.1). According to AHA guidelines, CD was considered correct if it averaged 50–59 compressions per minute, and CR was considered correct if it averaged 100–120 compressions per minute [54].

Secondary outcome: Subjective self-assessment

Participants completed an online questionnaire both before and after the BLS training. The survey collected demographic information and assessed self-reported confidence in performing CPR and managing emergencies. Responses were recorded on a 6-point Likert scale ranging from 1 (“strongly disagree”) to 6 (“strongly agree”).

Statistical analysis

All data analyses were performed using IBM SPSS Statistics version 28 (IBM Corp., Armonk, NY, USA). Equivalence testing has recently been highlighted in the literature as a suitable method for comparing different educational strategies [49]. Accordingly, equivalence was evaluated by comparing the proportion of participants achieving correct compression depth (CD) and compression rate (CR) across the three study groups. Two-sided 95% confidence intervals (CIs) were calculated following CONSORT guidelines for equivalence trials [45]. Results were considered significant when the empirical percentage differences’ 95% CIs did not overlap with the predefined equivalence margins. CIs for differences in proportions were estimated using the Wilson score interval method for independent samples [55]. The same approach was applied to 95% CIs of differences in Likert-scale confidence ratings between groups.

Equivalence margins

Equivalence thresholds were determined based on historical data from prior training sessions at our center [56–58]. Observed success rates for CPR following Peyton’s 4-step BLS instruction among novices ranged from 45% to 64% for CD and 33% to 52% for CR, covering a 19-percentage-point span. Any performance outcomes within these ranges for a different training method were considered equivalent. Therefore, the equivalence margins for comparing EM, EA, and control groups were set at Δ = 19% and −Δ = −19% for both CD and CR. Due to the limited prior data on confidence ratings, a difference of ±0.5 points on the 6-point Likert scale (approximately 8%) was designated as the equivalence margin.

Results

Sample characteristics

Data were collected from a total of 430 participants. Three participants were excluded due to missing group allocation, incorrect entries, or non-participation, resulting in a final sample of 427 individuals (70.7% female, 28.8% male, 0.5% diverse; mean age 20.7 ± 3.5 years). Partially completed datasets were retained for analysis. A chi-squared (χ²) test of independence confirmed that randomization was successful, as no significant demographic differences were observed among the study groups (Table 1).

Notes: M = mean; SD = standard deviation; ^a = participation in an emergency course within the previous year.

Descriptive data

Table 2 presents the participants’ performance outcomes and self-reported confidence levels before and after training for each of the three study groups.

Notes: t0 = pre-training assessment; t1 = post-training assessment; CD = compression depth; CR = compression rate; CPR = cardiopulmonary resuscitation; M = mean; SD = standard deviation; IQR = quartile range; NOR = non-outlier range.

Equivalence analysis

Figure 2 illustrates the outcomes of the equivalence analyses, showing the proportional differences between the EM, EA, and control groups with 95% confidence intervals. The blue lines represent the predefined equivalence thresholds (−Δ and Δ).

Compression depth (CD)

One week after BLS training (t1), the proportion of participants achieving correct CD was 63.8% in the EM group, 40.5% in the EA group, and 40.4% in the control group. The comparison between EA and EM showed a proportional difference of 23.3 percentage points (95% CI = 11.4% to 34.2%), indicating equivalence, though EM showed a tendency toward higher performance. Comparing EM with the control group revealed a 23.4% difference (95% CI = 11.5% to 34.2%), also demonstrating equivalence with a trend favoring EM. The EA versus control comparison yielded a 0.1pp difference (95% CI = −11.6% to 11.7%), indicating clear equivalence between these groups.

Compression rate (CR)

For CR at t1, correct performance was achieved by 42.8% of participants in the EM group, 45.8% in the EA group, and 47.1% in the control group. The proportional difference between EA and EM was 3.0pp (95% CI = −14.7% to 8.7%), demonstrating significant equivalence. EM versus control resulted in a 4.3pp difference (95% CI = −15.8% to 7.4%), and EA versus control showed a 1.3pp difference (95% CI = −13.0% to 10.6%), both indicating significant equivalence among the groups.

Subjective self-assessment

At t1, mean differences in self-reported confidence between EM and EA were minimal: 0.02 points (95% CI = −0.27 to 0.23) for confidence in performing CPR and 0.01 points (95% CI = −0.30 to 0.26) for confidence in handling an emergency. These results indicate significant equivalence between EM and EA for both measures. Comparisons between EM and the control group showed mean differences of 0.16 points (95% CI = −0.08 to 0.40) for CPR confidence and 0.19 points (95% CI = −0.09 to 0.46) for emergency confidence, again indicating equivalence. Similarly, EA versus control differences were 0.18 points (95% CI = −0.07 to 0.43) for CPR confidence and 0.20 points (95% CI = −0.07 to 0.48) for emergency confidence, confirming significant equivalence. All results are summarized in Table 2 and visualized in Figure 3.