This pre-post efficacy study of the VCA app for error disclosure was conducted to determine whether laypersons’ ratings of resident error disclosure improve with practice and feedback. With input from experienced attending physicians, we made 12 cases, which included 4 cases specific to the following three fields: internal medicine, pathology, and obstetrics and gynecology (OBGYN). Each case consisted of 3 or 4 vignettes depicting sequential stages in a conversation, such as initially sharing information about a mistake or responding to a patient’s emotional response (Multimedia Appendix 1). We recruited resident physicians at an academic center to complete each of the four cases (ie, 2 cases at 2 time points). Participating residents were randomly assigned to 1 of 2 groups within their specialty that differed only by the order of case assignment (Figure 1). Residents completed 2 cases at time 1; feedback was provided via the app after approximately 4 weeks, and residents were asked to complete the remaining two cases at time 2. The time points were spaced by 4 weeks, so that residents could be recruited while they were available during teaching conferences. We asked residents to not discuss the cases with colleagues until after study completion. We collected crowdsourced ratings of residents’ performances and surveys from residents.

The VCA app was described previously [16,23], and this project used the same software for presenting vignettes, recording residents’ responses, and delivering feedback. For each specialty, 2 cases portrayed an angry patient response, and another 2 cases portrayed a sad emotional response. Professional actors portrayed the patients. We designed cases to be of equivalent error severity. Further, 2 attending physicians and 3 senior residents pilot-tested the cases in their specialty and provided feedback via structured interviews to improve the cases’ relevance, quality, and believability.

Feedback reports, which were viewed in the VCA app, presented users with their overall communication scores for each vignette, the average scores of their peers, and written advice derived from the comments of raters who responded to the following question: “what would you want the provider to say if you were the patient in this situation?” Users could replay their own responses and listen to a highly rated response from a peer (Figure 2).

We recruited residents from March to June 2020 at all training levels (postgraduate years 1-4) from 3 departments at the University of Washington. The departments represented a procedural specialty (OBGYN), a diagnostic specialty (pathology), and a nonprocedural specialty (internal medicine). In the United States, pathology and OBGYN residency training typically lasts 4 years, and internal medicine residency training lasts 3 years. We invited all 237 OBGYN (n=28), internal medicine (n=183), and pathology (n=26) residents. Recruitment for the pathology and OBGYN specialties occurred at program-wide teaching conferences, with protected time for VCA use. The COVID-19 pandemic disrupted the in-person internal medicine conferences that were planned for recruitment; most internal medicine residents were recruited via email invitation. Each conference included a 10-minute orientation to VCA and 30 minutes for completing 2 cases. Residents received a US $50 gift card after completing all 4 VCA cases. At the time, the residencies did not have required program-wide training for error disclosure, and this trial did not provide didactic training.

The University of Washington Institutional Review Board determined that this study was exempt from a review of resident, layperson, and patient advocate participants based on its policies, procedures, and guidance (case identifier: STUDY00008246) [26]. Risks and benefits were explained verbally; consent was implied by voluntary participation in the VCA exercise.

Residents provided audio responses to each vignette through the VCA software. Audio responses were bundled into rating tasks for MTurk layperson raters who met the following criteria: US residents, those aged 18 years or older, and those able to speak and read English. Raters for OBGYN vignettes were further restricted to women. To participate, raters completed an audio check, answered demographic questions, read a description of the vignette that was written in lay language, viewed the patient video, and listened to the first resident response. The raters then responded to 6 items that covered domains related to accountability, honesty, apology, empathy, caring, and overall response (Multimedia Appendix 2) before advancing to the next response. The items used a 5-point scale that was labeled with “Poor,” “Fair,” “Good,” “Very good,” and “Excellent.” The instrument was developed by the investigators because the core competency assessment tool developed by the US ACGME only measures residents’ disclosure of patient safety events as “participates,” “discloses,” or “models disclosure,” rather than assessing disclosure quality or the patient experience of disclosure [27]. Although a very limited number of tools exist for rating residents’ error disclosure to standardized patients [7], they include questions about body language or are intended for faculty raters; therefore, they were not appropriate for incorporation in the VCA app. After rating the last response in the set, the raters were prompted to enter free text in response to the following question: “What would you want the provider to say if you were the patient in this situation?” Crowdsourced raters received variable compensation amounts based on a rate of US $0.20 per rating.

Residents were surveyed before and after each video-based communication assessment with the questionnaires that were built into the VCA app. The initial survey asked about the residents’ sex, their level of training, and whether they had personally disclosed medical harm to a patient. Residents answered the following two items, using a 5-point scale ranging from “strongly agree” to “strongly disagree”: “I am confident in my ability to share information with a patient after medical harm” and “I am confident in my ability to respond to patient emotions after medical harm.” After assessment completion, residents were asked to rate the VCA app’s ease of use and the relevance of the cases. They were also asked to rate their performance in the domains of accountability, clear explanation, apology, empathy, caring, and overall response (Multimedia Appendix 2). The items were rated on a 5-point Likert scale that was labeled with “Poor,” “Fair,” “Good,” “Very good,” and “Excellent.” Before time 2, residents were asked if they had “incorporated the feedback into how [they] communicate with patients generally (not just communication about medical harm events).” Due to a technical error, this second survey was not shown to OBGYN residents.

We sought at least 8 raters per response after removing raters with indications of inattention or low contributions to reliability [28]. Ratings were aggregated across raters, items, and vignettes to create an overall score for each resident at time 1 and at time 2. Scores were created by averaging multiple responses to 6 Likert scale questions that were designed to assess general communication skills; thus, the continuous scores presented in this paper were derived from ordinal approximations of continuous variables [29,30]. The reliability and generalizability of the representative cases were analyzed and reported separately [23]. We used a 2×3 split-plot ANOVA to test the effects of time (pre-post) and specialty on communication ratings.

We used paired samples 2-tailed t tests to examine whether residents’ self-confidence in their ability to share information and respond to emotions increased from time 1 to time 2. To determine if self-confidence was related to actual ratings, data were subjected to multiple linear regression analyses wherein residents’ self-confidence in their abilities predicted pre- and postfeedback ratings. To determine if changes in such confidence from time 1 to time 2 were associated with changes in actual ratings, difference scores were created; one score represented the difference in residents’ ratings between rounds, and the other represented the difference in residents’ reported self-confidence in their abilities between rounds. Difference scores were subjected to a Pearson correlation analysis. To determine if self-reports of experiences with personally disclosing a harmful error to a patient (“yes” vs “no”) before time 1 were associated with time 1 performance, we used an independent samples 2-tailed t test. To determine if years in training were associated with ratings, we performed a Pearson correlation analysis. To determine if physicians’ specialty was associated with prior disclosure experience, we used a chi-square test.

To determine if communication ratings varied with the emotional affects of patients, we created 2 affect scores for each resident. One score represented a resident’s average rating across all vignettes for the two angry affect cases, and the other represented a resident’s average rating for the two sad affect cases. Scores were subjected to a paired samples 2-tailed t test.

Of the 238 residents from all specialty departments who were contacted to volunteer for this study, 62 (26%) completed the first two VCA cases (time 1) and received feedback (Table 1). Of these 62 residents, 48 (77%) completed the postfeedback cases (time 2). Less than half of all residents (27/62, 44%) reported prior experience with disclosing medical harm to patients; experience differed significantly among specialties (P<.001). Further, 1 of 17 (6%) pathology residents reported previously participating in such conversations, whereas 15 of 23 (65%) OBGYN residents and 11 of 22 (50%) internal medicine residents reported prior experience with such conversations.

Prior to time 2, of the 48 returning residents, 23 completed a survey about incorporating the VCA feedback in general communications with patients. About half (11/23, 48%) agreed or strongly agreed that they had incorporated the feedback, 39.1% (9/23) neither agreed nor disagreed, and 13% (3/23) disagreed that they had incorporated the feedback.

Of the 62 residents, 38 (61%) completed surveys about the VCA app. A majority (32/38, 84%) found the cases to be “very much” or “completely” relevant to their practice. Additionally, 71% (27/38) found the VCA app to be “very much” or “completely” easy to navigate. We achieved a mean of 8.63 crowdsourced raters per case after removing raters with low contributions to reliability, resulting in an average cost of US $8.90 to rate the responses from 1 resident.

Based on the ANOVA for examining changes in mean resident communication ratings from time 1 to time 2, we found that residents were rated significantly higher at time 2 (mean 3.75, SD 0.16) than at time 1 (mean 3.53, SD 0.25; P<.001).

Among the 30 residents who completed surveys at both time points, confidence in the ability to share information about medical harm increased from time 1 (mean 2.87, SD 0.73) to time 2 (mean 3.47, SD 0.63; P<.001). Residents’ confidence in their ability to respond to patients’ and families’ emotions after medical harm events also increased from time 1 (mean 3.20, SD 0.71) to time 2 (mean 3.60, SD 0.72; P=.005). The multiple linear regression analysis showed no relationship between confidence in such abilities and performance on either the prefeedback ratings or postfeedback ratings. Based on the difference scores evaluated with the Pearson correlation analysis, we found no relationship between increases in confidence over time and increases in ratings over time.

No differences in communication ratings based on specialty were detected. We found no significant relationship between residents’ years in training and overall communication ratings (P=.44). However, residents who reported prior experience with disclosing medical harm to patients performed better at time 1 (mean 3.63, SD 0.23) compared to those without prior disclosure experience (mean 3.46, SD 0.25; P=.02).

Residents’ communication was rated significantly higher for angry cases (mean 3.69, SD 0.22) versus sad cases (mean 3.58, SD 0.21; P=.01).

After their first VCA use, residents’ mean self-rating of their overall response was 3.82 (SD 0.80). Residents reported their performance in sincerely expressing regret to patients (mean 4.05, SD 0.77), acknowledging and validating patients’ emotions (mean 4.00, SD 0.66), showing that they care about the patient (mean 3.79, SD 0.70), expressing accountability for the harmful event (mean 3.68, SD 0.70), and explaining things in a way that the patient could understand (mean 3.45, SD 0.80).

We measured laypeople’s assessment of residents’ error disclosure skills before and after they received numerical and written feedback. Residents’ mean ratings on a 5-point scale improved from 3.53 at baseline to 3.75 after feedback (P<.001). This finding provides novel evidence that simulated practice and feedback from laypeople can improve resident physicians’ error disclosure communication. The VCA app represents a novel, scalable, and statistically reliable tool for educators seeking to satisfy the ACGME mandate that residents “participate in the disclosure of patient safety events, real or simulated” [5]. Our findings indicate that the VCA app can be used across multiple specialties. This tool could particularly support educators and residents in diagnostic specialties, such as pathology, who may have fewer opportunities to participate in real-life error disclosure.

Consistent with prior literature about the accuracy of physician self-assessment [24,31], we found that residents’ confidence in their error disclosure skills did not correlate with laypeople’s ratings for these skills. Although physicians should still reflect on their performance, educators can emphasize the use of crowdsourced ratings as a more patient-centered and reliable way to assess error disclosure preparedness. This finding aligns with recommendations by disclosure experts that physicians should seek advice before discussing harmful events with patients [32,33]. Coaches often use brief practice and feedback to help clinicians recognize ineffective communication habits and phrasing before actual disclosures.

The higher ratings of residents when addressing patients with an angry affect versus patients with a sad affect warrant further study. This finding runs counter to predictions that residents might react defensively to confrontational, angry patients, which would be expected to result in lower communication ratings. One possible explanation for this finding is that angry patients’ challenging comments precipitated the explicit acknowledgement of their anger, whereas sad, withdrawn patients did not prompt a direct expression of acknowledgment or support. This would exacerbate existing physician tendencies to avoid discussing patient emotions, as indicated by observations that only 55% of attending surgeons who perform simulated error disclosure attempt to acknowledge or validate patients’ emotions [34]. A second hypothesis is that residents feel less shame when causing anger rather than sadness and respond readily with supportive expressions. Alternatively, the residents in this study may have been exposed to different instructions (ie, outside of this study) for handling patient emotions. Future research could pair an analysis of VCA response content with novel surveys about raters’ expectations for emotional support to characterize effective approaches for specific patient emotions. Error disclosure curricula should prepare trainees to tailor their communication approach to different patient and family emotional responses [35].

Our work has limitations. First, the VCA app assesses the skills needed for effective communication after harm but excludes other important areas, such as nonverbal communication. This limitation is offset by the strength that participants received actionable and focused feedback about their phrasing, which addresses a top concern of physicians and helps with not overwhelming them with advice across multiple domains [1,3]. Second, residents did not receive just-in-time coaching or a lecture on error disclosure—practices that might improve performance. Third, this study did not assess long-term skill maintenance. Fourth, the educational significance of the effect size is unknown, as we did not establish a threshold for either competence or mastery. Fifth, the limitations of the study population include recruitment at a single academic center and heterogeneity in the cohort’s training levels. A minority of eligible internal medicine residents participated (22/183, 12%), and not all participants completed all cases, which may have introduced selection bias. To address this, subsequent studies should be embedded in mandatory curricula rather than be based on volunteer participation. Sixth, we did not collect data that would allow us to explain why some residents reported that they did not incorporate the feedback into their communication practices or why some did not complete the cases at the second time point; these findings warrant further study. The efficacy and real-world implementation of this work remain unknown; we tested the VCA app as a stand-alone intervention instead of incorporating it into a longitudinal curriculum with a lecture. Lastly, we did not simultaneously measure faculty ratings of residents. Although this limited our ability to make comparisons between faculty and crowdsourced laypeople, we do not believe that this is a weakness. Rather, it highlights 2 key strengths of the VCA app. First, laypeople directly represent the ultimate arbiters of physician communication effectiveness—patients themselves. Second, crowdsourced laypeople can be recruited rapidly for statistically reliable sample sizes and at lower costs when compared to faculty.

The VCA app for error disclosure allows users to practice in a safe environment, provides formative feedback, and appears to facilitate skill acquisition. If these findings are replicated, the VCA app will likely offer a scalable way for residency leaders to meet ACGME mandates and assess residents’ skills. Yet, important questions remain about how best to incorporate the VCA app in graduate medical education. Multisite confirmatory studies should examine the effectiveness of using the VCA app in conjunction with didactic sessions on error disclosure and coaching from teaching faculty, as well as longitudinal skill development. Additional work could determine the efficacy of the VCA app for other challenging communication scenarios and for other learner groups, including practicing physicians. Finally, the viewpoints of residency leaders and residents about VCA acceptability, feasibility, and appropriateness will be needed to ensure adoption and sustainable use.