Correlates of physician burnout across regions and specialties: a meta-analysis

Background Health care organizations globally realize the need to address physician burnout due to its close linkages with quality of care, retention and migration. The many functions of health human resources include identifying and managing burnout risk factors for health professionals, while also promoting effective coping. Our study of physician burnout aims to show: (1) which correlates are most strongly associated with emotional exhaustion (EE) and depersonalization (DP), and (2) whether the associations vary across regions and specialties. Methods Meta-analysis allowed us to examine a diverse range of correlates. Our search yielded 65 samples of physicians from various regions and specialties. Results EE was negatively associated with autonomy, positive work attitudes, and quality and safety culture. It was positively associated with workload, constraining organizational structure, incivility/conflicts/violence, low quality and safety standards, negative work attitudes, work-life conflict, and contributors to poor mental health. We found a similar but weaker pattern of associations for DP. Physicians in the Americas experienced lower EE levels than physicians in Europe when quality and safety culture and career development opportunities were both strong, and when they used problem-focused coping. The former experienced higher EE levels when work-life conflict was strong and they used ineffective coping. Physicians in Europe experienced lower EE levels than physicians in the Americas with positive work attitudes. We found a similar but weaker pattern of associations for DP. Outpatient specialties experienced higher EE levels than inpatient specialties when organization structures were constraining and contributors to poor mental health were present. The former experienced lower EE levels when autonomy was present. Inpatient specialties experienced lower EE levels than outpatient specialties with positive work attitudes. As above, we found a similar but weaker pattern of associations for DP. Conclusions Although we could not infer causality, our findings suggest: (1) that EE represents the core burnout dimension; (2) that certain individual and organizational-level correlates are associated with reduced physician burnout; (3) the benefits of directing resources where they are most needed to physicians of different regions and specialties; and (4) a call for research to link physician burnout with performance.


Background
Health care organizations globally realize the need to address physician burnout due to its close linkages with quality of care, retention and migration. A 2008 World Health Organization (WHO) report found that the major factors for turnover and migration were poor or dangerous working conditions, insufficient resources, limited career opportunities, and economic instability [1]. The field of health human resources (HHR) deals with human resource issues for workers in the health sector, and has been suggested as a way to strengthen health system performance and to improve well-being for health professionals [2]. The many functions of HHR include identifying and managing the individual and environmental burnout risk factors, while simultaneously promoting effective coping [3][4][5].
Burnout is a specific pattern of response to chronic work-related stress that is a serious issue for many physicians [6]. Physician burnout is characterized primarily by a depletion of mental energy, known as emotional exhaustion (EE). With such depletion, providers feel unable to give of themselves, which leads to cynical attitudes and detached feelings toward patients, known as depersonalization (DP). The third burnout dimension is negative self-appraisal, especially in the competencies required to work with others, known as diminished personal accomplishment [6]. Our study will focus on the EE and DP dimensions only.
The frameworks to explain the development of burnout in health professionals have ranged from personal characteristics to work organization variables or a combination of the two. For example, Wiskow et al.'s model emphasizes the impact of the work environment, which is influenced by: (a) organizational functionality; (b) organizational culture; (c) management and patient support; (d) staff development; and (e) work-family balance [7]. These elements have been linked to burnout, medical errors and quality of care [7,8]. In turn, burnout is posited to be a risk factor for increased turnover and migration in physicians [2,8,9]. Existing evidence supports models with personal and work characteristics. The three levels of change to reduce burnout risk are: (1) modifying the organizational structure and work processes; (2) improving the fit between the organization and the individual physician, including professional development programs to facilitate better adaption to the work environment; and (3) individual-level actions to reduce stress and poor health symptoms through effective coping and promoting healthy behaviors [2,3,10].
The aims of our study of physicians are to determine which correlates would be most strongly associated with EE and DP, and whether the associations would vary across geographical regions and specialties. The three levels of burnout risk served as the framework for the categorization of variables that we created in this study. Our findings will help the field of HHR to identify personal and work characteristics that are the most significant risk factors for EE and DP, and direct resources most needed to physicians of different regions and specialties.

Methods
We chose meta-analysis in this study. The use of multi-sample data of physicians from different regions and specialties allows for the examination of a more diverse range of risk factors than would be possible with any single-sample data. Our study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines and reporting standards [11].

Literature search
We searched for published studies from 1991 to 2011, using the terms, 'physician/doctor emotional exhaustion, ' 'physician/doctor burnout, ' and 'physician/doctor coping, ' with the search engines: Cochrane, Embase, TheFreeLibrary.com, Google, Google Scholar, LILACS, PsycINFO, PubMed, SciELO and Scopus. Our search yielded 92 studies of medical doctors, but 27 were excluded either because they each included physicians with other health professionals (k = 13), or did not report or respond to email requests for the necessary statistics (k = 14). The remaining (K = 65) sampled exclusively physicians and provided either the sample correlations (r) or statistics that could be converted to r. The studies used in our meta-analysis are listed in the Appendix. Four of these were published in Spanish, with the variables and text recorded and translated into English by the fifth author.

Procedure
For the coding of sample characteristics, we coded each study sample on the method of survey administration, response rate, sample size, gender distribution, mean years of age, mean years in practice, country of the sample and medical specialty distribution.
For the coding of statistics, the internal consistency reliability estimates (Cronbach's α) and the associations between each correlate with EE and DP were recorded.
For conversion to r, 30 studies provided either a 2 × 2 χ 2 , t-ratio, one-way F-ratio or odds ratio (OR). To convert the χ 2 to r, we used the formula [12]: and to convert the t-ratio or F-ratio to r, we used the formula [12]: To convert the OR to r, we used the formula [13]: For the correlates, the classification of variables as either environmental drivers or constraints were informed by Lewin's field theory [14], which posits that behavior is the function of the person and the environment, and Lowe and Chan's classification of healthy work environment indicators [15]. The remaining variables were categorized either as work-life conflict, contributors to good health, contributors to poor mental health or coping strategies (see the correlates under each of the categories in Table 1). The fourth author classified all the variables, and the first author checked the categorizations. The inter-rater agreement for the classification was 100%.
The correlates classified under work engagement drivers are: recognition/feedback, autonomy, organization/peer support and adequate resources. The correlates classified under work engagement constraints are: professional values (for example, compromise of beliefs), organization structures (for example, supervision, inflexible work arrangements), inadequate resources, role ambiguity/conflict, insufficient input, workload, inadequate skills/preparation, and position-specific demands (for example, patient suffering and emotions). Work attitudes drivers include job and professional satisfaction, and organizational commitment. Work attitudes constraints include lack of motivation, career regret and intent to leave profession. The correlates classified under health and safety drivers are: quality and safety culture (for example, time for patients, management of patient-load), clinical skills, and professional development. The correlates classified under health and safety constraints are: incivility/conflicts/violence, and lack of quality and safety (for example, ergonomics and work-related hazards). Work-life/home conflict is incompatibility between professional and personal obligations and commitments. Contributors to poor mental health include fatigue, anxiety and depression. Contributors to good health include relaxation, hobbies, time for self and others. The correlates classified under adaptive coping are: social support (family, relatives, friends, outside acquaintances) and problemfocused (for example, prioritization of goals, finding meaning, spirituality). Ineffective coping includes overeating, inactivity and emotion-focused.

Analyses
For point estimates, our meta-analysis did not include any correlate examined in only one sample. Where a study had two or more separate item measures for a given correlate, we first calculated their mean r with the burnout dimension. For each correlate with a k ≥2, we calculated the weighted mean meta-correlation (ρ), and ρ corrected for within-sample measurement unreliability (ρ c ), using the formula [16]: where r xy = r of correlate with burnout dimension, α x = reliability estimate of correlate, α y = reliability estimate of burnout dimension. We substituted the value of the weighted mean of Cronbach's α or 1 when no reliability estimate was provided. We considered ρ c ≥0.30 to have practical significance for evaluation purposes. For example, a ρ c = 0.30 between a work constraint and burnout could mean that 66% of physicians in restrictive environments have high EE levels, and 66% of those in supportive environments have low EE levels [17].
For dispersion around ρ c , we calculated the variance of ρ c (σ 2 ρ c ), and the Q-test for homogeneity of r [16], where significance indicates that the associations vary across k. For homogeneous k, the standard error of ρ c (SE ρ c ) formula is [18]: and for heterogeneous k, the SE ρ c formula is [18]: where σ 2 res = σ 2 ρ c -σ 2 ρ. The SE ρ c was used to construct the 95% confidence interval (CI) of ρ c . For group differences, for heterogeneous k, we compared the difference in ρ c 's between regions and between specialty groups using the formula [12]: To check for publication bias, the file drawer problem exists when studies with significant results are published, while those with non-significant results are not reported. This and other types of publication bias are evident when the funnel plot (r by n) is asymmetrical or skewed [19]. Publication bias was checked by: (1) estimating the k with non-significant r that would be needed to increase the ρ c 's significance level to ≥0.05 (that is, fail-safe k or k fs ) for each correlate [20], and (2) examining the funnel plots of correlates with k ≥15.

Analytical software
We used the META 5.3 meta-analysis program (National Collegiate Software Clearinghouse, Raleigh, NC, USA) [21] to estimate the weighted mean of Cronbach's α, ρ, ρ c , σ 2 ρ c , SE ρ c , 95% CI of ρ c , Q-test for homogeneity of r, and k fs . We used the Microsoft Excel 2010 (Microsoft, Redmond, WA, USA) spreadsheet to convert the χ 2 , tratio, one-way F-ratio and OR statistics to r's; compute the K and N, and each group k and n descriptive statistics; and one-way F-and Z-difference tests. We used the Excel scatter chart program to create the funnel plot. Table 2 shows that the overall K = 65; N = 28,882; weighted mean for years of age = 45, weighted mean for years in practice = 15, and weighted mean for proportion of males = 73%. All research participants were administered questionnaires either through postal mail (61%), email (4%), in person (24%) or unspecified (12%), and the weighted mean response rate = 62%.

Sample characteristics
For the Americas, k = 26, n = 12,457; for Europe, k = 28, n = 13,085, and for Australia/Asia, k = 11, n = 3,340. On average, the European samples were older (47 years) than either the American (42 years) or Australian/Asian (41 years) samples. On average, the American samples provided a lower response rate (53%) than either the European (69%) or Australian/Asian (68%) samples.
We divided the samples into three specialty groups. The first was where, within a study sample, all the physicians saw their patients in hospital settings (inpatient specialties); the second was where, within a study sample, all of the physicians saw their patients in non-hospital settings, such as in walk-in clinics (outpatient specialties); and the third was where, within a study sample, some physicians saw patients in hospital settings and other physicians saw patients in non-hospital settings (mix of inpatient and outpatient specialties). For the inpatient specialty group (anesthesiology, internal, gynecology, oncology, otolaryngology, pediatric, surgical), k = 25, n = 10,935; for the outpatient specialty group (emergency medicine, infectious diseases, general/family, ophthalmology, psychiatry), k = 17, n = 4,775; and for the mixed group, k = 23, n = 13,172. On average, the outpatient specialty group had fewer years of practice experience (13 years) than either the inpatient specialty (16 years) or mixed (15 years) groups. On average, the outpatient specialty group provided a higher response rate (76%) than either the inpatient specialty (63%) or mixed groups (57%). Table 1 shows the k, n and the weighted mean of Cronbach's α of each variable. The weighted mean of Cronbach's α ranged from 0.61 to 0.89 for the correlates, with 15/17 (88%) above 0.70. The weighted mean of  (14) 62 (19) Values presented as weighted mean (SD). k, number of samples; n k , cumulative n across k. a P <0.05; b P <0.01.

Reliability estimates
Cronbach's α ranged from 0.84 to 0.90 for EE and from 0.68 to 0.80 for DP. Tables 3 and 4 show the k, n, ρ, ρ c , σ 2 ρc , 95% CI of ρ c , Q-test, and k fs . Table 3 reveals that EE had 25 correlates with k ≥2, and 17/25 (68%) had ρ c 's ≥0. 30. Autonomy (ρ c = −0.36) was the strongest correlate of the work engagement drivers; workload (ρ c = 0.66) and organizational structure (ρ c = 0.45) were the strongest correlates of the work engagement constraints. EE was associated with the work attitude drivers (ρ c = −0.47) and work attitude constraints (ρ c = 0.46). Quality and safety culture (ρ c = −0.34) was the strongest correlate of the health and safety drivers; incivility/conflicts/violence (ρ c = 0.41), and lack of quality and safety (ρ c = 0.42) were equally strong correlates of the health and safety constraints. EE was strongly associated with work-life conflict (ρ c = 0.49), and contributors to poor mental health (ρ c = 0.62), moderately associated with contributors to good health (ρ c = −0.32) and ineffective coping strategies (ρ c = 0.33).

Overall associations
EE had 22/25 (88%) correlates with k fs ≥10, and 17/25 (68%) correlates with a k fs /k ratio ≥4/1, indicating minimal risks of the file drawer problem. Figures 1, 2, 3 and 4 show the plots of r by n for workload, work attitude drivers, lack of quality and safety, and contributors to poor mental 95% CI ρ c , confidence interval; σ 2 ρc , variance of ρ c ; k, number of samples; k fs , fail-safe k for critical ρ c ≥0.05; n, sample size across k; ρ, weighted mean metacorrelation; ρ c , ρ after correcting for measurement unreliability with value ≥0.30 in bold; Q, homogeneity of r test, where significance indicates that the ρ c 's vary across k. a P <0.05; b P <0.001.
health. All four were funnel-shaped with three symmetrical, indicating minimal risks of publication bias [17]. Table 4 reveals that DP had 22 correlates with k ≥2, and 11/22 (50%) had ρ c 's ≥0. 30. Three correlates, adequate resources, inadequate resources and insufficient inputs, each had k = 1 and were not included in the meta-analysis for DP. Organizational structure (ρ c = 0.47) was the strongest correlate of the work engagement constraints. DP was moderately associated with the work attitude drivers (ρ c = −0.36) and work attitude constraints (ρ c = 0.32). Quality and safety culture (ρ c = −0.35) was the strongest correlate of the health and safety drivers; incivility/conflicts/violence (ρ c = 0.51) was a stronger correlate than lack of quality and safety (ρ c = 0.33) of the health and safety constraints. DP was moderately associated with work-life conflict (ρ c = 0.34), and contributors to poor mental health (ρ c = 0.34). DP had 18/22 (82%) correlates with k fs ≥10, and 13/22 (59%) correlates with a k fs /k ratio ≥4/1, indicating minimal risks of the file drawer problem. Figure 5 shows the plot of r by n for contributors to poor mental health. It was funnel-shaped and symmetrical, indicating a minimal risk of publication bias.

Group differences
The interpretation of the overall ρ c 's must be qualified due to heterogeneity of r's across k on both burnout dimensions. The r's were heterogeneous on 18/25 (72%) correlates for EE, and on 12/22 (55%) correlates for DP. For these correlates, we compared the significance of ρ c differences between the two largest regions, the Americas and Europe, and between inpatient and outpatient specialty groups. We did not compare with the mixed group because both 95% CI ρ c , confidence interval; σ 2 ρc , variance of ρ c ; k, number of samples; k fs , fail-safe k for critical ρ c ≥0.05; n, sample size across k; ρ, weighted mean metacorrelation; ρ c , ρ after correcting for measurement unreliability with value ≥0.30 in bold; Q, homogeneity of r test, where significance indicates that the ρ c 's vary across k. a P <0.05; b P <0.001. inpatient and outpatient specialties were combined in a given sample. Tables 5 and 6 show the k, n, ρ c and Z-difference between regions. Table 5 reveals that EE had 13/18 (72%) correlates with significant ρ c differences. The ρ c 's of the Americas were stronger than Europe on 11 correlates. The most notable differences were in quality and safety culture (−0.56 versus −0.25), professional development (−0.41 versus −0.22), work-life conflict (0.57 versus 0.40), problem-focused coping (−0.44 versus −0.17), and ineffective coping (0.53 versus 0.32). Physicians in the Americas were at lower risk than physicians in Europe for EE when quality and safety culture and career development opportunities were present, and problem-focused coping was used. The former were at higher risk than the latter when work-life conflict was present, and ineffective coping was used. The ρ c of the work attitude drivers was stronger for Europe (−0.64) than for the Americas (−0.28), indicating that the former was at lower risk than the latter for EE when their attitudes were positive. Table 6 reveals that DP had 8/12 (67%) correlates with significant ρ c differences. The ρ c 's of the Americas were stronger than Europe on seven correlates. The most notable differences were in lack of quality and safety (0.47 versus 0.29), and work-life conflict (0.42 versus 0.27). Physicians in the Americas were at higher risk than physicians in Europe for DP when quality and safety was compromised and work-life conflict was present. Tables 7 and 8 show the k, n, ρ c and Z-difference between specialty groups. Table 7 reveals that EE had 8/16 (50%) correlates with significant ρ c differences. The ρ c 's of the outpatient specialties were stronger than the inpatient specialties on seven correlates. The most notable differences were in autonomy (−0.79 versus −0.57), organizational structures (0.72 versus 0.32) and contributors to poor mental health (0.77 versus 0.56). The former  were at higher risk than the latter for EE when the organization of work was constraining and poor mental health were present, but were at lower risk for EE with autonomy and use of problem-focused coping. The ρ c of the work attitude drivers was stronger for the inpatient specialties (−0.44) than for the outpatient specialties (−0.29), indicating that the former was at lower risk for EE when their attitudes were positive. Table 8 reveals that DP had 3/10 (30%) correlates with significant ρ c differences. The ρ c 's of the outpatient specialties were stronger than the inpatient specialties for work-life/home conflict (0.46 versus 0.32), and contributors to poor mental health (0.63 versus 0.35), indicating that the former were at higher risk than the latter for DP when work-life conflict and poor mental health was present. The ρ c of the work attitude drivers was stronger ρ c for the inpatient specialties (−0.45) than the outpatient specialties (−0.29), indicating that the former was at lower risk for DP when their attitudes were positive. Figure 6 shows the overall associations and reveals some significant trends. The ρ c 's with burnout were stronger for constraints than for drivers. Similarly, the ρ c 's with burnout were stronger for work-life conflict and contributors to poor mental health than contributors to good health. EE was more strongly associated with a greater number of correlates than DP. EE's stronger ties with the environmental drivers and constraints support Maslach's contention that it represents the core aspect of burnout [22]. The results also support Maslach's position that EE is more closely tied to health states. The implication is that while drivers are important, the management of constraints may be even more critical for physicians who experience      high EE. In summary, our findings suggest that attempts to reduce burnout risk could operate at three levels: individual (healthy lifestyle/behaviors, adequate coping), the individual and the environment (social support structures, relationships, improving person-organization fit), and at the organizational level (adequate working conditions, organization of work, design) [7,9,10].

Drivers and constraints of EE
Excessive and unevenly distributed workloads are fairly pervasive constraints, and were strongly associated with EE. The improvement of work processes, flow and interpersonal relationships (quality and safety) were drivers associated with reduced EE. A positive work attitude was another driver associated with reduced EE, and suggests the benefit of fostering greater organizational commitment and career satisfaction. Contributors to poor mental health and work/life conflict were strongly associated with EE, indicating the importance of self-care practices, and improved personal and family management.

Drivers and constraints of DP
A culture of quality and safety and positive work attitudes were critical drivers associated with reduced DP. Contributors to poor mental health and work/life conflict were associated with DP, although the link between poor mental  Figure 6 Summary of overall associations. The direction of ρ c is indicated by either '+' or '−', with the large font indicating high magnitude of association.
health and EE was much stronger. Again, our findings underscore the necessity of self-care and finding the right career balance.

Americas versus Europe
The moderating role of region may be partly due to age differences, with physicians from Europe, on average, being five years older than physicians from the Americas. The age-related maturity may have enabled many of the European physicians to better manage the risk factors. Work-life conflict had stronger associations with burnout in physicians from the Americas than physicians from Europe. In addition to the greater maturity levels, physicians from Europe may have received more extended family support than their colleagues from the Americas. A caveat worth noting is that the dissimilar mean response rates between the physicians from Europe and the Americas may have distorted the group differences in ρ c 's.
For physicians from the Americas, possible ways to reduce EE include rebalancing the constraints of heavy workload and position-specific demands, while improving quality and safety culture, and professional development. For physicians from Europe, possible ways to reduce EE include managing the factors that contribute to poor health. For physicians from the Americas, possible ways to reduce DP include cultivating a climate that generates positive work attitudes, quality and safety culture, work-life balance, and managing the contributors to poor health.
Our findings suggest that factors other than culture and economics should be considered when comparing these two regions. The challenges and complexities inherent in the physicians' work may limit the scope of any proposed changes. Redistributing patient-loads may be difficult within health systems faced with chronic resource constraints. The changes in work routines, resource distribution and decision-making processes may be resisted by physicians and other health professionals. Applying Wiskow et al.'s three levels of change require an integrated, systems approach based on careful planning and coordinated implementation [7].

Inpatient versus outpatient specialties
The moderating role of specialties may be partly due to differences in practice experience, with inpatient specialties having, on average, three more years of practice than outpatient specialties. The increased knowledge commensurate with experience may have enabled many of the inpatient specialties to better manage the risk factors. The stronger associations between EE and its correlates for the outpatient specialties also suggest increased difficulties with work organization and processes due to geographical isolation, and the transient nature of patient relations. The most significant finding was the link between contributors to poor mental health and burnout, with the ρ c stronger for outpatient than inpatient specialties. This may indicate that managing the financial, logistical and other businessrelated needs of outreach clinics exacts a severe toll on their health. Their health deterioration is associated with increased workload and challenges over and above the demands of clinical practice. Outpatient specialties in managed care systems may experience negative health states due to the highly regulated environment, which limits their autonomy, decision input and ability to develop long-term professional relationships with patients [23]. Possible ways to reduce EE for them include providing informational and material resources, training/development programs, and collegial and administrative support. A caveat worth noting is that the dissimilar mean response rates between the outpatient and inpatient specialties may have distorted the group differences in ρ c 's.

Study limitations
One study limitation is the lack of uniform standards in the reporting of sample characteristics and r's. The conversion of ORs may have yielded imprecise r estimates [13]. A second is the inability to infer causality. Did poor health lead to physician burnout, or vice-versa? Similarly, did work-life conflict precede or follow from burnout? A third is that health contexts may have influenced some of our results, but characteristics of health systems (for example, public versus private) were not always reported. A fourth is the paucity of research on physicians in Africa or the Middle East. Finally, we were unable to collect or interpret studies published in languages other than English or Spanish.

Conclusions
Our study found that reducing the individual and organizational-level risk factors is associated with decreased burnout. Documenting the regional and specialty differences lays the foundation for directing resources where they are most needed. Our findings also reveal the lack of research linking physician burnout with performance. A US study found that physician DP was associated with diminished patient satisfaction and longer post-discharge recovery time [24]. Additional studies could link physician burnout with quality of care and medical errors, which have been found to be negatively associated with patient safety and recovery [25]. Research could examine how physician burnout relates to health behaviors, professional development, communication skills, and overall quality of life.