Our search strategy identified 4732 articles, of which 684 met the criteria for full text review (fig 2). A total of 170 studies involving 239 246 physicians (150 cross sectional studies including 231 964 physicians and 20 prospective or longitudinal studies including 7282 physicians) met the eligibility criteria. The characteristics of the included studies are summarised in appendix 6, and citations are provided in appendix 7. Agreement between reviewers for study inclusion was high (κ 0.89, 95% confidence interval 0.81 to 0.96).
77 (45%) of the 170 studies were conducted in the US, 48 (28%) in European countries, four (2%) of which were in the UK two (1%) in the African Region, eight (5%) in the Region of the Americas, two (1%) in the South-East Asian Region, three (2%) in the Eastern Mediterranean Region, 29 (17%) in the Western Pacific Region, and one (1%) multinational study.38394041424344 107 (63%) of the studies were based in a hospital setting, 33 (19%) involved mixed settings, 29 (17%) were based in primary care setting, and one study45 was unclear but involved medically qualified academics. We aimed to exclude studies with fewer than 70% of responses from physicians, however, in reality, only three (2%) of 170 studies with a mixed sample of physicians (70% and over) and other health professionals were included in our analyses.
The median number of physicians across studies was 312 (interquartile range 162-1015 ) with a median age of 42 years (32-48) and where data for sex were reported, 112 (66%) studies involved mostly male physicians. The physician specialty varied across studies: 42 (25%) mixed specialties, 32 (19%) internal medicine, 21 (12%) surgery (ie, trauma, plastic, and neurosurgical), 19 (11%) emergency medicine and intensive care, 11 (6%) general practitioners, eight (5%) interns or residents, eight (5%) paediatrics, seven (4%) oncology (ie, gynaecologist, radiation, or palliative care), six (4%) neurology, three (2%) psychiatry, and 13 (8%) involving other specialties. Physicians had more than seven years of experience in 52 (31%) studies, a mixture of experience was reported in 47 (28%) studies, and 38 (22%) studies involved residents, junior doctors, or interns with fewer than seven years of experience.
The most common measure of burnout was the full 22-item Maslach Burnout Inventory (81 (48%) of 170 studies). An abbreviated version of the Maslach Burnout Inventory was used in 50 (29%) studies, other types were used in 34 (20%) studies and only five (3%) studies used the Copenhagen burnout inventory (see appendix 8 in supplement for breakdown of the measures used). Thirty one (18%) studies reported secondary measures of depression and 24 (14%) studies reported emotional distress, which were analysed separately. In terms of career engagement for physicians, 81 (48%) studies reported on decreased job satisfaction compared with increased job satisfaction, 19 (11%) on career choice regret compared with being satisfied with career choice, three (2%)42464748 poor career development compared with good career development, nine (5%)474950515253545556575859 on reduced productivity compared with sustained productivity, and 36 (21%) on turnover intention compared with retention. Concerning quality of patient care outcomes, 39 (23%) studies reported patient safety incidents compared with no patient safety incidents, 43 (25%) reported indicators of low professionalism compared with maintained professionalism, and eight (5%) studies reported measures of patient dissatisfaction compared with satisfied patients. Nineteen (11%) studies reported more than one of these outcomes.
Of the 119 (70%) studies reporting career engagement, all were self-reported by the physician. Physicians self-reported across most of the studies for patient safety incidents (31 (79%) of 39) and professionalism (37 (80%) of 46 studies), whereas the remaining studies used patient record reviews and surveillance systems. Patient satisfaction was based on self-reports by patients.
One hundred and 30 studies provided a representative sample of the target population (76% met criterion 1); 103 studies provided an ample sample size of physicians (61% met criterion 2); 58 studies reported a response rate of 60% or greater (34% met criterion 3); 25 studies satisfied low risk of bias for the ascertainment of exposure mostly due to many of the surveys being self-reported (15% met criterion 4); 100 of the studies adequately adjusted for confounding factors (59% met criterion 5); 165 reported a low risk of bias due to assessment of outcome (97% met criterion 6); and 118 studies had used adequate statistical tests and measures to report their findings (69% met criterion 7). Overall, 32 (19%) of the studies reported low risk of bias (total score 6-7), 23 (14%) reported high risk of bias (total score: 0-2), and 115 (67%) studies reported medium risk of bias (total score 3-5). The full results of the Newcastle Ottawa critical appraisals are presented in appendix 9.
Meta-analysis of association of burnout with career engagement and quality of patient care
The results of all the meta-analyses are provided in table 1. All forest plots for each outcome are available in appendix 10. Only significant results are reported here.
Physician burnout was associated with almost fourfold decreases in job satisfaction compared with increased job satisfaction based on measures of overall burnout (3.79, 95% confidence interval 3.24 to 4.43, I2=97%, k=73 studies, n=146 980 physicians), emotional exhaustion (4.81, 3.67 to 6.30, I2=98%, k=33, n=22 699), depersonalisation (2.89, 2.37 to 3.53, I2=92%, k=30, n=22 002) and personal accomplishment (2.88, 2.28 to 3.63, I2=93%, k=32, n=27 374). Burnout was associated with threefold increases in career choice regrets compared with being satisfied with their career choice based on measures of overall burnout (3.49, 2.43 to 5.00, I2=97%, k=16, n=33 871) and emotional exhaustion (4.16, 3.34 to 5.19, I2=90%, k=4, n=2014). Burnout was associated with up to threefold increases in turnover intention compared with retention based on measures of overall burnout (3.10, 2.30 to 4.17, I2=97%, k=25, n=32 271; fig 3), emotional exhaustion (2.81, 1.80 to 4.40, I2=99%, k=16, n=23 625), and depersonalisation (1.82, 1.26 to 2.62, I2=99%, k=11, n=23 257; fig 4) but no effect was seen for personal accomplishment. Burnout was associated with small but significant decreases in productivity compared with sustained productivity based on measures of overall burnout (1.82, 1.08 to 3.07, I2=83%, k=7, n=9581), depersonalisation (1.23, 1.18 to 1.28, I2=96%, k=3, n=2969) and personal accomplishment (1.53, 1.43 to 1.63, I2=97%, k=3, n=2969). Finally, only two studies464748 reported a significant pooled association between overall burnout and career development concerns compared with good career development (3.77, 2.77 to 5.14, I2=0%, k=2, n=3411).
Physician burnout was associated with double the risk of patient safety incidents compared with no patient safety incidents based on measures of overall burnout (odds ratio 2.04, 95% confidence interval 1.69 to 2.45, I2=87%, k=35, n=41 059; fig 5), emotional exhaustion (2.15, 1.82 to 2.53, I2=73%, k=17, n=20 213), depersonalisation (2.44, 1.84 to 3.23, I2=90%, k=14, n=19 616), and personal accomplishment (1.47, 1.20 to 1.80, I2=87%, k=14, n=19 616; fig 6). Burnout was associated with more than twofold decreases in professionalism compared with maintained professionalism based on measures of overall burnout (2.33, 1.96 to 2.70, I2=96%, k=40, n=32 321), emotional exhaustion (2.45, 1.71 to 3.53, I2=94%, k=16, n=11 861), depersonalisation (2.93, 1.93 to 4.46, I2=93%, k=12, n=10 488), and personal accomplishment (2.17, 1.36 to 3.46, I2=92%, k=9, n=2992). Burnout was also associated with up to threefold decreases in patient satisfaction compared with patients being satisfied based on measures of overall burnout (2.22, 1.38 to 3.57, I2=75%, k=8, n=1002), depersonalisation (3.82, 1.57 to 9.29, I2=81%, k=6, n=571), and personal accomplishment (1.79, 1.44 to 2.81, I2=5%, k=5, n=527). Publication bias was found after visual inspection of funnel plots and the test statistics for most comparisons were significant (appendix 13).
Some studies had used different scales to measure burnout, therefore, we also did analyses using standardised mean difference to account for measures of different length (see forest plots in appendix 11). However, we found no significant differences in this analysis and the results were consistent with those reported when analysed with the odds ratio.
The subgroup meta-analyses for the different measures of burnout used for the outcomes job satisfaction, patient safety incident, professionalism, and turnover intention is provided in appendix 8 in supplement.
For job satisfaction, the abbreviated version of Maslach Burnout Inventory provided the largest association with burnout (odds ratio 4.62, 95% confidence interval 3.21 to 6.65, I2=99%) and smallest with the Copenhagen burnout inventory (2.59, 2.22 to 3.01, I2=95%). The Copenhagen inventory had the highest association of burnout with patient safety incidents (3.59, 2.92 to 4.42, I2=95%) and the abbreviated versions of the Maslach Burnout Inventory had the lowest association (1.68, 1.16 to 2.43, I2=79%). The association between burnout and low professionalism was greatest when using an abbreviated version of the Maslach Burnout Inventory (2.91, 1.65 to 5.13, I2=87%) and lowest when using the Copenhagen inventory (1.89, 1.69 to 2.12, I2=43%). The association between burnout and turnover intention was greatest when other non-specific measures of burnout were used (7.23, 5.93 to 3.18, I2=77%) and lowest when an abbreviated version of Maslach Burnout Inventory was used (2.53, 1.39 to 4.59, I2=98%). No significant differences were noted between the different burnout measures when tested using the ratio of odds ratios (appendix 8).
The results of the univariable and multivariable meta-regression analyses are provided in appendix 12. In the univariable regression results, a stronger association of overall burnout with low job satisfaction was found in physicians working in hospitals compared with primary care settings (1.88, 0.91 to 3.86, P=0.09), and more specifically in emergency medicine and intensive care (2.16, 0.98 to 4.76, P=0.06) compared with a general internal medicine specialty, and in physicians older than the age of 50 years compared with individuals aged 31-50 years (2.41, 1.02 to 5.64, P=0.04). The association was weakest in GPs (0.16, 0.03 to 0.88, P=0.04). However, these associations did not remain significant in the multivariable regressions. The association between burnout and patient safety incidents in the univariable regression results was found to be larger in younger physicians (20-30 years; 1.88, 1.07 to 3.29, P=0.03), working in emergency medicine and intensive care settings (2.10, 1.09 to 3.56, P=0.02), or in training based in the Commonwealth region (3.03, 0.83 to 11.25, P=0.09). The only association to remain significant in the multivariable regression results was that found in younger physicians (1.55, 0.94 to 2.56, P=0.08). The univariable regression results of the association of burnout with low professionalism was found to be smaller in physicians aged older than 50 years (0.36, 0.19 to 0.69, P=0.003) and larger in physicians still in training or residency (2.27, 1.45 to 3.60, P=0.001), who worked in a hospital (2.16, 1.46 to 3.19, P<0.001), specifically in the emergency medicine specialty and intensive care (1.48, 1.01 to 2.34, P=0.04), or when situated in a low to middle income country (1.68, 0.94 to 2.97, P=0.08). Multivariable regression results show that the association remained significant in middle aged physicians aged 31-50 years (0.45, 0.26 to 0.76, P=0.003), working in a hospital (3.82, 1.84 to 8.00, P<0.001), or specialising in cancer (0.25, 0.09 to 0.74, P=0.01) or neurology (0.22, 0.07 to 0.73, P=0.01). The univariable regression results of the association of burnout with career choice regret was found to be largest in physicians with a specialisation in emergency medicine and intensive care (2.89, 0.97 to 14.89, P=0.10) and neurology (2.52, 0.82 to 7.80, P=0.10). The association between burnout and turnover intentions did not vary according to any other factors included in the univariable regression analyses (appendix 12). No significant associations were found between burnout and job satisfaction.