Burnout in Pediatric Residents: Three Years of National Survey Data

The PRB-RSC membership as well as the overall design, participant eligibility and recruitment and use of standard instruments in the de-identified annual survey through the Association of Pediatric Program Directors’ Longitudinal Educational Assessment Research Network have been described previously.^39,40  In 2016, the PRB-RSC consisted of 34 US residency programs; in 2017, 43 programs participated; and in 2018, 49 programs participated.

Briefly, participants were eligible if they were residents in categorical or combined (eg, medicine-pediatrics, pediatrics-psychiatry, etc) programs at PRB-RSC sites. There were no exclusion criteria.

The final survey consisted of 141 items, took 12 to 15 minutes to complete, and included demographic characteristics, residency experiences, and widely used scales to measure personal attributes associated with burnout and well-being (described below) and questions about satisfaction with social support, career, and learning environment.

The 22-item Maslach Burnout Inventory Human Services Survey was used to assess burnout.^41,42  Burnout was defined as having high subscale scores for personal emotional exhaustion (≥27) and/or depersonalization (≥13).^43,44 

Current physical and mental health was assessed by using the Patient-Reported Outcomes Measurement Information System questions; raw scores were converted to T-scores with a population mean score of 50.^45,46  Sleepiness was measured by using the Epworth Sleepiness Scale.^47–49  Stress was measured with the Perceived Stress Scale; scores among health professionals typically range from 14 to 18.^50–56  Mindfulness was assessed with the 10-item Cognitive and Affective Mindfulness Scale, Revised; the average item score in normative populations is 2.8 ± 0.5.^57–59  Resilience was assessed by using with the Brief Resilience Scale.⁶⁰  Self-compassion was measured by using Neff’s 12-item measure of self-compassion; average item scores in normative populations range from 2.7 to 3.2.⁶¹  CCC was assessed by using the 10-item Confidence in Providing Compassionate Care Scale; average scores in other studies of health professionals range from 60 to 80.⁶²  Empathy was measured by using the Davis empathy scales, with perspective-taking and empathic concern subscales used in the analyses.⁶³ 

Additional items that were aimed at assessing recent training in mind-body skills and satisfaction with career, support, personal life, and the educational environment were included as exploratory variables.

Descriptive statistics were used to describe participants’ demographic characteristics, burnout, and residency experiences. Fisher’s exact tests, χ² tests, t tests, and analysis of variance tests were used to determine if characteristics were associated with residency response rate and burnout, as appropriate. To account for the clustering of learners in programs, mixed-effects linear and logistic regression models were fitted to predict factors associated with burnout in a cross-sectional analysis for each year and all 3 years, with a random intercept for program included in every model and all predictors entered simultaneously to obtain adjusted odds ratios for each. In the data sets for each year and for all years combined, we used multiple imputation by chained equations to impute missing continuous variable predictors from other continuous variable predictors present. We conducted a longitudinal analysis using 2017 and 2018 data, including variables in the model based on their cross-sectional relationships with burnout and potential for intervention: self-compassion, mindfulness, empathy, satisfaction with work-life balance, overall satisfaction with their learning environment, current sleepiness, recent time off, and rotation. Because burnout is a dichotomous outcome, the analysis was a logistic mixed model, reporting odds ratios and 95% confidence intervals. Because stress and CCC are continuous variables, the analysis was a linear mixed model estimate of standardized, z-scored outcomes. In all analyses, our a priori level of significance was P < .05, and 2-sided tests were performed. Statistical analysis was performed with R 3.4, using the lme4 package for mixed modeling and the mice package for multiple imputation.

Each PRB-RSC member obtained approval from their local institutional review board.

In 2016, 61% of 2723 eligible residents from 34 programs participated; in 2017, 66% of 3273 residents from 43 programs participated; and in 2018, 61% of 3657 residents from 49 programs participated.

In Table 1, we show that the prevalence of burnout exceeded 50% in all 3 years. There were no consistent significant differences between residents meeting criteria for burnout and those not meeting criteria for burnout in terms of any demographic characteristic (Table 1). There were also no significant differences in having received training in mind-body skills (Table 2); participating in a special residency pathway, such as global health or primary care track; or recently experiencing a patient’s death (Table 3). Certain factors, such as residency year and program size, were associated with burnout in some years but not in others; for example, there were significant differences in burnout rates among postgraduate year 1 (PGY1), postgraduate year 2 (PGY2), and postgraduate year 3 (PGY3) in 2016 and 2017 but not in 2018. On the other hand, 2018 was the only year in which there were significant differences in burnout by program size (Table 1).

Compared with residents who did not meet criteria for burnout, those who did reported significantly worse mental health, more sleepiness, and greater stress (Table 2). Residents meeting burnout criteria consistently reported lower mindfulness and self-compassion scores, less CCC, and lower levels of empathy and resilience (P < .001 for all).

Residents who met criteria for burnout were more likely to be on high-acuity rotations (Table 3). They were approximately twice as likely to report recently having made a medical error, more likely to report a work-life conflict, less likely to have had a vacation within the past month, and less likely to have had a recent weekend off than those who did not meet criteria for burnout (P < .001 for all).

Residents who met burnout criteria consistently reported significantly less satisfaction with support from family, spouse, friends, faculty, and colleagues (Table 4). They also reported significantly lower quality of life, less satisfaction with their choice to go into pediatrics, and less satisfaction with the balance between personal and professional life (P < .001). Residents who were not burned out were much more likely than burned-out residents to strongly agree that they worked in a “collaborative rather than competitive environment,” that resident education and mentoring were high priorities in their programs, and that they were strongly satisfied with their learning environment (P < .001 for all factors for all 3 years).

In the cross-sectional mixed model logistic regression analysis, the variable most strongly associated with burnout was perceived stress (Table 5); other variables significantly associated with an increased risk of burnout were sleepiness, having made a recent medical error, and being dissatisfied with work-life balance. Four factors were consistently associated with a lower risk of burnout: empathy, self-compassion, overall quality of life, and CCC (P < .05 for each).

In the longitudinal logistic regression model predicting 2018 burnout, the strongest predictors were 2017 burnout and 2018 recent error, sleepiness, no recent weekend off, and current high-acuity rotation; after controlling for these factors, the only 2017 factor associated with 2018 burnout was quality of life (Table 6).

In the longitudinal regression for 2018 stress, the strongest predictors were 2017 stress and 2018 recent error, sleepiness, and no recent weekend off; after controlling for these predictors, the only 2017 factor longitudinally associated with 2018 stress was self-compassion (Table 6).

In the longitudinal regression for 2018 CCC, the strongest predictors were 2017 CCC and 2018 recent error; after controlling for these predictors, the 2017 factors longitudinally associated with 2018 CCC were mindfulness, satisfaction with the learning environment and career choice of pediatrics, and empathy (Table 6).

In this 3-year national survey of US pediatric residents, overall rates of burnout consistently exceeded 50%. We confirmed several previously identified risk factors (stress, sleepiness, recent medical error, and high-acuity rotation) and protective factors (mindfulness, self-compassion, CCC, and empathy) for burnout. We also built on findings from our earlier study, that is, that mindfulness and self-compassion were longitudinally associated with lower levels of stress and greater CCC, even after controlling for powerful factors like current sleepiness and recent error.³⁹  These results offer insights into potential targets for future interventions and research.

The stable 54% to 56% rate of burnout found from 2016 to 2018 in this study is similar to burnout rates previously reported in US studies of pediatric residents.^13,20  In surveys of Argentinian and Saudi pediatric residents, researchers have reported burnout rates >65%.^64,65  These consistent findings in our large national data set strengthen the conclusion that burnout is a prevalent, persistent problem.

Our findings on potential risk factors for burnout confirm earlier studies in other groups and identify attractive targets for interventions to improve resident well-being. For example, we confirmed that being on an intensive care rotation is associated with a higher risk of burnout.⁶⁶  Studies in which researchers evaluate interventions that ease the stress of high-acuity rotations are needed. We also found 40% increased odds of reporting a recent medical error in residents who met criteria for burnout, echoing earlier findings in practicing physicians and other specialties.^10,22,67,68  It is difficult to determine if making errors leads to burnout or whether being burned out leads to errors; prospective research is needed to identify strategies that reduce human error to protect against burnout while improving patient safety. Like others, we found that sleepiness, fatigue, stress, work-life conflicts, and dissatisfaction were associated with an increased risk of burnout.^{4,26,34,69–73}  Interventions addressing these factors may be worthwhile to test to reduce burnout.⁷⁴ 

On the other hand, we did not confirm some reported risk factors for burnout, for example, sex and debt.^{15,21,26,71,75–78}  The reasons for this difference are unclear and must be tested in future studies. Also, unlike Pantaleoni et al,²⁰  we did not find that burnout rates consistently increased from the beginning to the middle of residency training. Future studies are needed to better understand the natural history of burnout.

With our data, we confirm cross-sectional studies reporting several protective factors for burnout, such as that more mindfulness is associated with a lower risk of burnout.^{30,31,33,72,79–82}  Also confirmed in our data are earlier findings that self-compassion, CCC, and empathy are associated with a lower burnout risk.^{30,38,76,82,83}  Controlled trials are needed to evaluate the impact on burnout of programs that improve mindfulness, self-compassion, and CCC.

With this study, we extend previous findings by providing extensive exploratory data on resident satisfaction and attitudes. We confirmed the finding that among Dutch residents, a better perception of the quality of the learning environment was associated with a lower risk of burnout.⁸⁴  Longitudinal controlled studies are needed to evaluate the impact of changes in the learning environment on burnout and strategies to enhance satisfaction with support.

We used a similar longitudinal model to extend our earlier research to predict 2018 burnout, stress, and CCC, controlling for the effect of additional variables (2018 recent error, sleepiness, no recent weekend off, and current rotation).³⁹  After controlling for these factors, 2017 quality of life was protective for 2018 burnout, 2017 self-compassion was protective for 2018 stress, and 2017 mindfulness was protective for 2018 CCC. That is, both the data from 2016 to 2017 and from 2017 to 2018 suggest that mindfulness and self-compassion may be longitudinally related to better scores on factors closely related to burnout, stress, and CCC.

Despite its large size and longitudinal nature, results of this voluntary survey of pediatric residents may not be generalizable to other specialties or stages of training. All of the variables evaluated depend solely on self-report; burned-out residents may view experiences, support, or programmatic activities more negatively than residents who are not burned out. Future researchers should use more objective measures and longitudinal analyses. The annual time frame for this study may not capture shorter- or longer-term effects. The survey was not used to examine unit, system, or cultural factors that may have affected burnout. Future researchers would benefit from having more data at the programmatic level to better understand where and how to best leverage interventions to improve resident well-being and reduce burnout. Future researchers should also include emerging potential cultural risk factors for burnout, including sexism, harassment, bullying, discrimination, or violence, and institutional factors, such as use of the electronic health record, scribes, and access to social workers and psychologists.^85,86 

In this national longitudinal study, >50% of pediatric residents met criteria for burnout. Promising targets for intervention include program-level interventions during stressful high-acuity rotations and/or in the time period after a major medical error as well as the timing of weekends off and individual training in protective factors, such as adequate sleep, mindfulness, empathy, and compassion. Future researchers should also address broader factors affecting burnout at institutional and cultural levels.

We thank all the residents who completed the surveys and all of the local program directors involved in the study. We also thank Beth King at the Association of Pediatric Program Directors’ Longitudinal Education Assessment Research Network for her invaluable work in coordinating communication with all sites, assisting with local institutional review board applications, administering the surveys, and coordinating the data collection. The PRB-RSC institutions and site investigators can be found at www.pedsresilience.com/organization/member-institutions.

CCC

confidence in providing compassionate care

PGY1

postgraduate year 1

PGY2

postgraduate year 2

PGY3

postgraduate year 3

PRB-RSC

Pediatric Resident Burnout and Resilience Study Consortium