Errors in the practice of medicine confront physicians with a dilemma: we want to shed light on our mistakes so that we can learn from and share the lessons they would teach us, but we hesitate to expose ourselves to collegial scrutiny for fear of embarrassment, lost reputation or discipline. We hear compelling calls to see errors as vicarious commodities for shared professional learning1 yet struggle to surmount the impediments to discussion that remain diverse and daunting.2 3

These impediments can be lessened if forums to discuss errors provide constructive criticism and collegial reassurance,4 but in the absence of such supportive venues, it is hard to avoid the charge that medicine has no place for its errors5 or that formal error discussions, such as morbidity and mortality conferences, remain incomplete.6 7 Residency training programmes are particularly in need of opportunities to discuss and learn from mistakes through dialogue with peers and supervisors.8 9 Training programmes that provide such opportunities may reap the additional benefit of lessening the risk of burnout, depression and lost empathy among their trainees.10 Moreover, institutions that facilitate discussions about errors can use these exchanges to encourage two other forms of error disclosure: reporting errors to institutions to improve patient safety11 and communicating errors to affected patients as part of respectful clinical care.12

To improve our understanding of error discussions, we surveyed faculty and resident physicians in teaching hospitals. We selected this setting because of the formative role teaching hospitals play in the development of physicians’ attitudes toward error discussions and their subsequent practice patterns.

METHODS

Participants

Faculty and resident physicians were drawn from three medical centres located in Midwest, Mid-Atlantic and Northeast regions of the US. Surveys were completed between June and September, 2004, with the exception of 32 resident physicians and 51 faculty physicians at one study site who completed surveys between January and March, 2005.

After approval by the Institutional Review Boards at each of the participating institutions, potential participants were invited to complete a self-administered, paper-based survey at the end of an organised educational activity, as an independent activity or through mailed correspondence depending on local institutional preferences. Participation was voluntary and included token incentives, and the questionnaire required approximately 20 minutes to complete. In order to send a second or third invitation to those who did not respond to mailed correspondence, participation was tracked by using a carefully worded response card separate from the survey. The card allowed potential participants to communicate that they had either completed the survey or did not wish to do so, without revealing whether or not they had actually completed the survey. No personal identifying information was collected, and participants were assured that they and their institutions would remain anonymous.

Resident physicians were from the disciplines of paediatrics, internal medicine and family medicine. Faculty physicians were from the disciplines of general paediatrics, general internal medicine and family medicine, along with 36 paediatric specialists from one study site. The paediatric specialists were surveyed as a comparison group to contrast with the generalist faculty physicians. When all variables were analysed for differences, responses from paediatric specialists were not significantly different from generalist faculty responses, so responses from the paediatric specialists were included for analysis.

Survey questionnaire

To assess respondents’ willingness to discuss errors, we used a hypothetical error vignette similar to that employed by Blendon et al,13 followed by three different outcomes of varying severity—no harm, minor harm and major harm (see box 1). The question at the end of each outcome was followed by response options of very likely, likely, not sure, unlikely and very unlikely. A modified version of the hypothetical vignette and responses was given to paediatric faculty and residents by changing the patient to a 7-year-old boy and adjusting the third clinical outcome by making no mention of a myocardial infarction and concluding that “the patient’s condition stabilises and he is transferred out of the intensive care unit after 24 hours”.

Based on an empirically derived taxonomy of factors that facilitate or impede error disclosure,3 questions concerning attitudes, experiences and practices related to error discussions were designed to probe the domains of responsibility to profession, attitudinal barriers and fears and anxieties. Attitudinal questions used 5-point Likert scale responses ranging from strongly agree to strongly disagree.

Demographic variables included training level, specialty, gender, belief in forgiveness, experience giving medical-legal testimony and being named as a defendant in a malpractice case. To assure anonymity of responses, we did not query age, year of graduation or race/ethnicity.

Based on prior focus group data, we did not make a distinction between medical error and medical mistake and printed the following statement in the questionnaire: “Definition: we use ‘medical error’ and ‘medical mistake’ interchangeably to describe a preventable adverse event that affects a patient by prolonging treatment or causing discomfort, disability or death”.

The questionnaire was pilot tested with 16 participants for face validity, clarity and stability over time, as previously described.12 Items that had a test-retest reliability correlation coefficient less than 0.50 were not included in the final questionnaire. All of the final questions had good to excellent reliability, with test-retest Spearman’s Rho > = 0.6.14 The questionnaire is available upon request from the first author.

Statistical analysis

Answers from the questionnaires were entered manually into an Access data file and then uploaded into PC SAS V.8.1 (SAS Institute Inc, Cary, NC, USA). For analysis, Likert scale responses were dichotomised as follows: (1) likely/very likely versus not sure/unlikely/very unlikely and (2) agree/strongly agree versus neutral/disagree/strongly disagree. By grouping undecided and negative responses together, this dichotomisation intentionally placed primary analytic focus on positive responses. To simplify reporting in the results, likely signifies the combination of “likely” and “very likely” responses, and agree signifies the combination of “agree” and “strongly agree” responses.

We calculated frequency distributions of responses and used the two-tailed Fisher’s exact test or the χ² statistic to test response differences between groups such as the differences between faculty and resident responses. To determine significant predictors of our dependent variables—hypothetical error discussions and past role-modelling—multivariate analyses were conducted using backwards stepwise regression on the following independent variables: gender, training level, specialty, belief in forgiveness, experience giving malpractice-related testimony, being named as a defendant in a malpractice case, and each of the items listed in tables 1 and 2 (except for the question regarding role modelling). Only variables significant at the α = 0.05 level were retained in the final multivariate model for each of the dependent variables.

RESULTS

Response rates and demographic characteristics

Surveys were completed by 138 faculty physicians and 200 resident physicians. The overall response rate was 74%, with subgroup response rates of 82% (faculty) and 69% (residents). Table 3 describes respondents’ demographic characteristics.

Discussing hypothetical errors

The likelihood of discussing a hypothetical error depended on the outcome of the error: 77% of respondents indicated they would likely discuss a hypothetical error if it resulted in no harm, 87% if it resulted in minor harm and 94% if it resulted in major harm (table 4). Cross frequency analyses between these responses revealed a hierarchical relationship: 98% of those who would likely discuss a “no harm” error would also discuss a “minor harm” error, and 99% of those who would discuss a “minor harm” error would also discuss a “major harm” error.

Attitudes regarding errors and discussions with colleagues

As shown in table 1, a variety of motivations to discuss errors were affirmed by a majority of respondents: to learn whether colleagues would have made the same judgment or decision (91%), to allow colleagues to learn from one’s mistake (80%), to receive support and understanding (79%) and to unburden oneself (60%). Moreover, 70% of respondents agreed that discussing their errors with colleagues strengthens their relationships with those colleagues. However, 27% agreed that such discussions are hard because of potential damage to one’s reputation, and a majority of respondents agreed that “in my experience, physicians tend to expect perfect performance from each other” (64%) and “competition in medical education and training encourages students and trainees to keep silent about their mistakes” (69%).

Experiences and practices regarding error discussions

Table 2 describes the frequency of experiences and practices among faculty and residents relevant to error discussions. Approximately half of respondents had observed, during their training, a more experienced clinician discuss a medical mistake, and a similar proportion felt that their residency programme provided a supportive environment in which medical mistakes could be discussed. Most respondents knew at least one colleague who would be a supportive listener if a mistake needed to be discussed, and most also indicated that they usually discuss their mistakes with colleagues. Fifty-seven percent of respondents had tried, at least once, to serve as a role model by discussing one of their own mistakes with students or residents. Nearly two-thirds of respondents reported they usually tell their significant others about their mistakes

Most resident physicians reported knowing a fellow resident (87%) or faculty physician (81%) who would be a supportive listener if a mistake needed to be discussed. However, 4% of residents acknowledged not telling their supervising physician about a mistake that prolonged treatment or caused discomfort, disability or death and 13% of residents believed that they do not need to tell their supervising physician about mistakes that cause temporary and minor harm to a patient.

Differences based on training level, gender and malpractice experience

As noted in tables 1 and 2, resident physicians were more likely than faculty physicians to be motivated to discuss errors by a desire to allow colleagues to learn from their mistakes, but they were more concerned than faculty about receiving blame or loss of reputation from colleagues when disclosing mistakes in general. Residents were more likely than faculty to have observed a more experienced clinician discuss a mistake and were also more likely to feel that their training programmes provided a supportive environment in which to talk about mistakes.

Women were more likely than men to be their own worst critics after making an error (96% vs 87%, p = 0.002) and more concerned about receiving blame (66% vs 46%, p = 0.003) or loss of reputation from colleagues (70% vs 53%, p = 0.001). Women were also more likely than men to be motivated to discuss errors to learn whether a colleague would have made the same decision (95% vs 87%, p = 0.01), to receive support and understanding (84% vs 73%, p = 0.01), and to unburden themselves (66% vs 55%, p = 0.04), and they were more likely to feel that their training programmes provided a supportive environment for error discussions (56% vs 42%, p = 0.01).

Respondents who had been defendants in a malpractice case were less likely to report that they usually discuss their errors with colleagues (58% vs 75%, p<0.05).

Multivariate analyses: variables associated with discussing hypothetical errors and trying to role-model error discussions

Eight variables remained significant (α = 0.05) in one or more of the models for the three hypothetical vignettes (with variable outcomes of no harm, minor harm and major harm) or for having tried to serve as a role model in discussing a mistake with students or residents.

Variables associated with discussing an error resulting in no harm included two positive and two negative independent predictors: usually discussing mistakes with colleagues (OR 10.17, 95% CI 5.16 to 20.06), knowing at least one colleague who would be a supportive listener (OR 2.67, 95% CI 1.02 to 7.00), faculty (versus resident) status (OR 0.29, 95% CI 0.14 to 0.59) and feeling it is hard discuss mistakes because of potential damage to one’s reputation (OR 0.45, 95% CI 0.22 to 0.90). Variables associated with discussing an error resulting in minor harm included one positive and one negative independent predictor: usually discussing mistakes with colleagues (OR 18.46, 95% CI 7.76 to 43.91) and faculty (versus resident) status (OR 0.21, 95% CI 0.09 to 0.48). Variables associated with discussing an error resulting in major harm included three positive independent predictors: female gender (OR 3.09, 95% CI 1.00 to 9.50), usually discussing mistakes with colleagues (OR 7.97, 95% CI 2.38 to 26.66) and believing that discussing mistakes strengthens professional relationships (OR 3.71, 95% CI 1.17 to 11.74).

Variables associated with having tried to serve as a role model by discussing an error included four positive independent predictors: faculty (versus resident) status (OR 3.80, 95% CI 2.05 to 7.05), having observed a more experienced clinician discuss a mistake (OR 4.17, 95% CI 2.34 to 7.42), wanting colleagues to learn from one’s mistakes (OR 3.00, 95% CI 1.54 to 5.82) and believing that discussing mistakes strengthens professional relationships (OR 2.15, 95% CI 1.22 to 3.79).

DISCUSSION

The results of this study provide new and confirming information about physicians’ attitudes and practices regarding the discussion of medical errors among themselves within the formative environment of a teaching hospital. Our data suggest that most physicians report that they usually discuss their medical errors with colleagues and do so for professional and personal reasons related to a commitment to shared learning and the need for support. We also found that most respondents believed that discussing errors with colleagues strengthens professional relationships and that most knew at least one colleague who would be a supportive listener if they needed to discuss a mistake. Nevertheless, it is notable that 27% of respondents indicated they do not usually discuss their mistakes with colleagues, 30% did not agree that discussing errors strengthens professional relationships and 11% did not know a colleague who would be a supportive listener.

Formal error discussions in the medical profession have traditionally taken place in morbidity and mortality conferences. However, the potential for candid and supportive discussion of errors in such settings is limited by concerns about blame,15 public humiliation,6 lack of senior clinicians who role-model by acknowledging their own errors,7 and inattention to the emotional impact of errors on physicians.4 Ideally, such forums would simultaneously promote shared professional learning and provide emotional support for the physicians involved, but this dual agenda may be unrealistic in “public” settings of shared professional learning. Our data suggest that the majority of physicians in teaching hospitals discuss their errors with colleagues, but many of these discussions are likely occurring informally within the confines of trusted professional relationships rather than in formal educational settings.

We found that the likelihood of discussing a hypothetical error depends on the seriousness of the error’s outcome: 77% of respondents indicated they would likely discuss a hypothetical error if it resulted in no harm, 87% if it resulted in minor harm and 94% if it resulted in major harm. This graduated approach to error discussions suggests that physicians are driven more by the need for personal support (which presumably increases with the severity of an error’s outcome) than by the desire for shared learning (which would derive from the error independent of its outcome). But an alternative conclusion is suggested by a recent qualitative study of residents and students, many of whom indicated that they learned most from errors when harm was caused.16

About one in 10 respondents did not know even one colleague who would be a supportive listener if they needed to discuss a medical error and 27% of our respondents found it hard to talk with colleagues about errors for fear of damage to their professional reputation. Such findings emphasise the point that physicians can be the “second victims” of an error4 17 and therefore need access to supportive colleagues, especially when they are in training.10 Unconditional support for physicians is a vital service18 and it need not come at the expense of an error’s objective assessment.4 Fortunately, some hospitals are beginning to establish peer support teams designed to provide such assistance in the aftermath of error.19 20

Resident physicians in our study were more likely than faculty physicians to be motivated to discuss errors by a desire to allow colleagues to learn from their mistakes, and they were more likely to discuss a hypothetical error resulting in no harm or minor harm. They were also more likely to have observed a more experienced clinician discuss a mistake and to feel that their training programmes provided a supportive environment in which to talk about mistakes. These data may signal positive changes in the culture of training. Yet not all residents were inclined to be candid with their supervisors, and about one in five did not know a faculty physician who would be a supportive listener if a mistake needed to be discussed. Nevertheless, the residents in our study appeared more inclined to discuss errors with their supervising physicians than residents in three other studies, in which 29%–46% of trainees had not disclosed their most serious error to their attending physician.8 10 21

Graduate medical education requires supervising physicians committed to meeting the needs of residents after errors by providing a balance of emotional support and medical instruction.22 23 But data from a study of primary care preceptors confirms how difficult it is to strike this balance: approximately half of the faculty preceptors were inclined to offer reassurance after an error, but the same proportion indicated that a resident’s error would negatively impact the resident’s written evaluation.24 Our study confirms that physicians perceive themselves as highly self-critical, giving credence to Mizrahi’s25 observation that residents may come to believe that “because they perceive themselves as their own worst judges, they should be their only judges.” Faculty physicians should take this mindset into consideration and identify approaches to error discussions that disarm defensiveness and clear a path to constructive self-reflection.

We found that 57% of respondents had tried to serve as a role model by discussing one of their own mistakes, and there was a strong correlation between having observed a more experienced clinician discuss one of their own errors and having tried to role model such a discussion. Though such role-modelling appears to be infrequent in formal educational settings,7 it is not known how frequently supervising physicians refer to their own experiences of error to reduce learners’ feelings of isolation and encourage a shared sense of humility based on a common fallibility. Such role-modelling may help counter the competitiveness in training environments that—according to two thirds of our respondents—encourages trainees to keep silent about their mistakes.

Because medical errors can be so overwhelming, it is not surprising that 65% of our respondents reported that they usually tell trusted loved ones about their errors, consistent with other studies in which rates of disclosure to non-colleagues have ranged from 53%–67%.10 18 21 Physicians depend on friends and family to serve as confidants, especially when discussions with colleagues are perceived as focusing too much on facts and not enough on feelings.4 While such confidants can be profound sources of comfort, insights from communication studies point out that such discussions blur the boundaries between work and private life and may impose considerable burdens on relationships already taxed by the demands the medical profession places on families.26

Our study makes important contributions to the literature on error communication among physicians by reporting attitudes derived from an empirically-based taxonomy of factors that facilitate or impede error disclosure3; simultaneously exploring attitudes, experiences and practices (including the influence of role-modelling on error discussions); employing a hypothetical vignette with variable outcomes to assess how the outcome of an error may impact a physician’s willingness to discuss it; and comparing faculty and resident physicians to assess differences based on training level. Some of the taxonomy-based attitudinal questions were drawn from the domain of “responsibility to profession” which contains four facilitating factors: (1) desire to share lessons learned from errors; (2) desire to serve as a role model in disclosing errors or breaking bad news; (3) desire to strengthen inter-professional relationships and build inter-professional trust; and (4) desire to change professional culture by accepting medicine’s imperfections and lessen the focus on managing malpractice risks.3 These four factors represent a wide range of deontological and consequentialist ethical values, and they also reflect the significance of virtue ethics for physicians and the need for communities of practice within which virtue and wisdom are developed and sustained in the company of fellow professionals.27 28

Our study also had limitations. Even though the survey was anonymous, social desirability bias may have led some respondents to give answers that were perceived to be more socially acceptable. Second, we employed a hypothetical error vignette with outcomes of variable severity that provided dependent variables for multivariate analysis. However, answers to hypothetical scenarios may not predict actual behaviour. Third, due to the skewed distribution of responses to the hypothetical error vignettes (skewed towards “agree” responses), the confidence intervals in many of the multivariate analyses were wide. Fourth, although responses were generally similar across the specialties we studied (paediatrics, internal medicine and family medicine) our results represent data from a population of physicians comprised mainly of generalists based in teaching hospitals, so our results may not be generalisable to physicians in other specialties or in other practice settings. Lastly, our data were collected in 2004–2005 and may not reflect more current attitudes or practices.

Discussions of medical errors by physicians will always be a challenging yet vital responsibility, one that cannot be avoided if we want to learn from our mistakes and receive support as we work through their implications. This responsibility derives from a commitment to respect our patients and our colleagues, and it is directed towards quality patient care and the integrity of the profession. Such commitment is increasingly articulated within a framework of medical professionalism,29^–31 the wide scope of which may encourage physicians to appreciate the interconnected reasons not only for discussing errors with colleagues, but also for reporting errors to institutions and disclosing errors to patients.11 12

To facilitate discussions of errors in teaching hospitals, physicians need carefully arranged educational venues and access to trustworthy colleagues in order to promote and coordinate the contrasting goals of professional learning and emotional support after errors occur.23 32 Role models need to communicate humility, appropriate self-criticism, and a shared sense of professional responsibility as they discuss their own errors.1 33 34 Special attention should be devoted to trainees, some of whom may have no colleagues to whom they feel they can turn when private discussions about errors are desired. Training programmes should encourage mentoring relationships that serve as havens of support after errors occur and manifest the paradox of accepting fallibility without lessening the commitment to excellence. Many years ago Osler observed that it is necessary to accept that “slips in observation are inevitable” and “that errors in judgment must occur in the practice of an art which consists largely in balancing probabilities,” and he was convinced that if mistakes are “acknowledged and regretted” we will draw from our errors the very lessons that will enable us to avoid their repetition.35 Perhaps the answer to the paradox of fallibility and excellence is to be found in Osler’s insight that fallibility can, if honestly engaged, promote excellence.