Burnout and Internal Medicine Resident Work-Hour Restrictions

Background  Burnout is very common in internal medicine residents. Effective July 2003, all residents were restricted to work less than an average of 80 hours per week and no more than 30 hours of continuous duty for patient care and educational obligations. We evaluated rates of burnout in internal medicine residents before and after the implementation of the new work-hour restriction.

Methods  University of Colorado Health Science Center internal medicine residents were surveyed in May 2003 and May 2004. The survey contained the Maslach Burnout Inventory, organized into 3 subscales (ie, emotional exhaustion, depersonalization, and personal accomplishment); the Primary Care Evaluation of Mental Disorders depression screen; and self-reported quality of care and education.

Results  The response rate was 87% (121 of 139 residents) and 74% (106 of 143 residents) in 2003 and 2004, respectively. Self-reported hours worked decreased from a mean of 74.6 to 67.1 (P = .003). In 2004, 13% fewer residents experienced high emotional exhaustion (42% vs 29%; P = .03). There was a trend toward fewer residents with high depersonalization (61% vs 55%; P = .13) and fewer residents with a positive depression screen (51% vs 41%; P = .11). Personal accomplishment did not change. The assessment of self-reported quality of care did not significantly change from 2003 to 2004. Residents reported attending fewer educational conferences per month (18.99 vs 15.56; P = .01). Overall residency satisfaction decreased 6 mm on a 100-mm visual analogue score (P = .02).

Conclusions  Burnout continues to be a major problem. Reducing hours may be the first step to reduce burnout but may also affect education and quality of care.

Conceptually, burnout describes a process beginning with high and sustained levels of stress resulting in feelings of irritability, fatigue, detachment, and cynicism.¹ In the helping professions, stress originates from frequent intense interactions with clients with complex problems.² These high demands, combined with lack of autonomy or support, result in burned-out employees.^3-6 The hallmark features of burnout are a combination of emotional exhaustion (EE), depersonalization (DP), and a reduced sense of personal accomplishment (PA).⁷ Emotional exhaustion is the principal dimension associated with job absenteeism and performance deficits.^8-10

Internal medicine residents are at particular risk for burnout.¹¹ They work long hours in a stressful environment with minimal control over their work environment or schedule. They spend most of their time in hospitals taking care of acutely ill patients. Health care has become increasingly complex with many diagnostic tests and interventions to master, and residents may not have yet acquired all the knowledge and abilities to care for their patients, causing internal stress and strife about the quality of their care.

The death of Libby Zion in March 1984 spurred a national reaction when the grand jury found fault with the system of residency training and physician staffing. In July 2003, the Accreditation Council for Graduate Medical Education (ACGME) mandated that residents work no more than 24 consecutive hours with an additional 6 hours for educational and administrative activities.¹² The ACGME began to enforce the 80-hour work week. The effect on resident burnout of these new work-hour restrictions and the resulting change in work schedule is unknown. Our objective was to compare rates of burnout in internal medicine residents before and after the implementation of the new work-hour restriction and evaluate the effect of these changes on the quality of education and care.

All University of Colorado Health Science Center (UCHSC) internal medicine residents were eligible to participate in the study. The UCHSC internal medicine training program includes residents distributed across 3 postgraduate years and 3 tracks; preliminary (6 residents total), primary care (10 residents per postgraduate year class), and categorical (36 residents per postgraduate year class). The residents rotate through 5 hospitals including the University of Colorado Hospital, Denver Health (county hospital); the Denver Veterans Affairs Medical Center, Rose Medical Center; and Presbyterian/St Luke’s Medical Center. In 2003, residents took 9 months of call during the first year, 8 months of call during the second year, and 4 months of call during the third year. Admissions were capped at 5 new patients per call per intern and 10 new patients per call per supervising resident. Categorical and preliminary residents had 1 half day of clinic per week, while primary care residents had 2 half days of clinic per week. Elective rotations did not include night call. During the 2003 academic year, residents took call no more than every fourth night. Prior to the 2004 academic year, changes were made to ensure that no resident worked more than 30 continuous hours and 80 hours per week on average as mandated by new ACGME regulations. These changes included 1 less call month for interns and a day float system that employed a resident to participate in rounds with the postcall teams to assure continuity of care. In addition, work hours were monitored through an online self-reporting system, and residents were consistently reminded of the work-hour regulations.

We developed an 82-question self-administered postal survey.¹³ Surveys were mailed to each resident’s home in May 2003 and again in May 2004, nearly 1 year after the new work schedule was implemented. Residents were asked to answer questions in reference to their previous month’s rotation. Colorado Multiple Institutional Review Board approved the research protocol.

The Maslach Burnout Inventory (MBI) consists of a 22-item self-administered 7-point Likert scale questionnaire organized into the following 3 subscales: 9 questions on EE, 5 questions on DP, and 8 questions on PA.² Based on a normative sample of 1104 medical workers, responders in the upper third were considered high scores for each subscale (ie, EE ≥27, DP ≥10, and PA ≤33).^2,11 To convert a 3-dimensional definition of burnout into a unidimensional construct, we followed the “Exhaustion +1” criteria, in that residents with high burnout were defined as having high scores in EE and DP, while burnout was defined as high EE or DP.^2,14,15 Personal accomplishment is not included because it correlates less with measures of psychiatric and physiologic strain.² Individuals with burnout report somatic complaints such as headache and sleep disturbance, feelings of hopelessness, irritability, and impatience, as well as substance abuse and a desire to “get away from people.”^16-18

The Cognitive Hardiness Scale (CHS) consists of a 30-item self-administered 7-point Likert scale composed of attitudes and beliefs about work and life, including commitment, challenge, and control. Scores are reported as a T score with the mean set at 50 with a standard deviation of 10. A score greater than 59 is considered high and identifies individuals with relative invulnerability to stress-related illness such as burnout.¹⁹ They tend to view life’s changes as challenges and opportunity for growth.²⁰ In a study of college students, CHS correlates with MBI subscales: EE (r = −0.61), DP (r = −0.66), and PA (r = 0.45).⁶

The Primary Care Evaluation of Mental Disorders (PRIME-MD) Patient Health Questionnaire (PHQ) screens for 5 mental disorders.²¹ The depression subset consists of a self-administered 2-question Likert scale.

The residents’ opinions on the quality of education and care were also assessed. Self-reported patient care was measured using 8 previously studied Likert questions developed by Shanafelt et al.²² Other questions included self-reported work and sleep hours, the type of clinical rotation just completed, teaching experiences, and overall satisfaction with residency measured on a 100-mm visual analog score anchored on each end by “very satisfied” and “very dissatisfied.”

We compared the responses from the 2 surveys using paired 2-tailed t tests and analysis of variance for continuous data and χ² tests for categorical data (SPSS Inc, Chicago, Ill). When expected cell numbers were low (<5), we used the Fisher exact test. P<.05 was considered statistically significant.

Our study sample size was sufficient to detect moderate effect sizes for our primary continuous variables. For example, the mean ± SD EE score in 2003 was 24.7 ± 9.0. We had a power of 0.91 to detect a difference of 4 points in the 2004 scores, assuming an α level of .05 (Pass 2002; NCSS Inc, Kaysville, Utah). For categorical variables, the study power is less, and we are able to detect only large effect sizes.²³

The response rate was 87% (121/139) in 2003 and 74% (106/143) in 2004 (P = .006). Respondent demographics were similar in both years (Table 1). Two thirds of the respondents were in the 26- to 30-year age group for both survey years. Comparing all responding residents in 2003 and 2004, the number completing an in-hospital call rotation was the same (76% vs 74%) for both years. The mean hardiness score was slightly higher in 2004 (55.2 vs 57.2; P = .09).

The number of residents with high EE decreased by 13% in 2004 (P = .03) (Table 2). There was a trend toward diminished DP (61% vs 55%; P = .13), burnout (74% vs 58%; P = .06), and high burnout (36% vs 25%; P = .05). Residents, however, maintained their sense of PA through the change in work schedule (Figure). Depression also trended downward in 2004.

The mean number of hours worked per week decreased from 74.6 hours in 2003 to 67.1 hours in 2004 (P = .003) (Table 2). The interquartile range for 2003 was 55 to 88 hours per week, whereas in 2004 it decreased to 50 to 80 hours. The attendance at educational conferences declined from 18.99 conferences per month in 2003 to 15.56 conferences per month in 2004 (P = .01). The morning report conference attendance decreased 20%, resulting in 2 fewer morning reports per month in 2004 (P = .02). The amount of sleep per night while on call (2.83 vs 2.56 hours; P = .12), and the percentage of residents spending 2 or more hours with their attending physician each day did not significantly change (54% vs 55%; P = .61). The number of residents reporting reading medical topics more than 2 hours per week increased from 42% to 54% (P = .06). However, overall satisfaction with the residency program decreased from a mean of 76.3 mm on a 100-mm visual analogue score to 70.3 mm (P = .02).

The assessment of self-reported quality of care did not significantly change from 2003 to 2004. Looking at each question individually, the percentage of residents that responded “at least monthly” to the 5 patient care questions did not significantly change (Table 3). A similar percentage of residents reported engaging in at least 6 of the 8 poor quality-of-care measures “at least monthly” in 2003 and 2004 (25% vs 24%; P = .51).

Comparing the residents at the same level in 2003 and 2004 (Table 4), interns in 2004 trended toward less EE compared with interns in 2003 (raw score, 22.5 vs 26.4; P = .08). A similar trend was noted for second-year residents, but no difference was noted for third-year residents. Depersonalization and PA did not differ for any resident year. There was a nonsignificant trend toward higher hardiness in 2004 for all postgraduate years. Work hours per week decreased for interns (−7.6 hours; P = .01) and second-year residents (−12.8 hours; P = .006) but increased for third-year residents (+3.4 hours; P = .45). Overall satisfaction with the training program increased as the residents moved to a higher level in 2003, but in 2004, satisfaction decreased by 14% for third-year residents (79.4 vs 68.6; P = .02). Sleep also decreased by 1 hour for third-year residents while on call in 2004 (3.35 vs 2.30; P = .006). Residents in all 3 years were attending fewer morning reports in 2004, especially third-year residents (15.1 vs 11.4; P = .048).

Few studies have looked at the direct impact of the work-hour restrictions on internal medicine residents. While the average number of hours worked per week for all residents was already less than 80 in 2003, a significant number of residents were exceeding the 80-hour limit. The work schedule also changed drastically with the elimination of traditional 36-hour continuous on-call, moving toward a coverage system involving more patient “hand-offs.” Covering resident were responsible for making patient care decisions once the admitting residents had completed their shift.

We originally hypothesized that the burnout rates would increase after the change in work schedule. This a priori hypothesis was driven by the supposition that the stress induced from passing and assuming care of patients from fellow residents would outweigh the benefit of fewer work hours. However, comparing residents in 2003 to 2004, we observed a decrease in all burnout subscales, most notably in EE. Comparing individual postgraduate years showed a trend toward less burnout. Since resident hardiness was not different between the years, personality factors that influence burnout rates did not influence the burnout rate.

To our knowledge, this is the first prospective, longitudinal evaluation of an entire internal medicine residency program’s self-reported quality of care. A retrospective evaluation of care in New York Hospital before and after implementation of work-hour restrictions suggested that work-hour restrictions led to delays in test ordering by house staff and increased complications without change in serious complication or length of stay.²⁴ This retrospective study assessed changes within a residency program rather than changes with individual residents. Evaluating individual residents, we observed a nonstatistically significant trend toward improved care. This lack of significance may represent a lack of effect on quality outcomes or the fact that the questions were not sensitive enough to detect a difference. In addition, it may take longer than 1 year for the new work schedule to affect quality-of-care measures. Even though it has been shown that residents with burnout report higher levels of suboptimal care, it is unknown if the relationship between work hours and quality of care is mediated via burnout.²² Landrigan et al²⁵ evaluated 20 interns in the intensive care unit, noting that those working a traditional call schedule made significantly more medical errors than when the same interns worked a restricted work schedule (relative reduction, 35.9%). This study used nonblinded direct observation of interns as the principal means of detecting medical error, which may have introduced bias. It is difficult to objectively assess an individual resident’s quality of care through surveys or medical chart review owing to the team approach to care and the insensitivity of using end points such as mortality or transfer to higher level of care.

Recently, a study compared 20 interns on a traditional intensive care unit work schedule to one that was modified to limit continuous clinical responsibility to 16 hours.²⁶ Interns on the traditional work schedule worked more hours and slept less. While working during on-call nights, 80% of these interns exhibited significantly more attention failures identified with continuous electrooculography showing slow-rolling eye movement. This difference was not noted during the day. This study did not correlate residents with attention failures to medical errors. However, lack of sleep has also been shown to impact cognitive performance.²⁷ Such decrease in sleep noted in our study may be due to the fact that the residents knew that they were going home at 1:00 PM the following day and did not have to pace themselves through their call night.

Although we did not directly measure each resident’s clinical knowledge, the overall educational experience appeared to be negatively affected by the change in work schedule. Despite having more time away from clinical responsibilities, the residents reported attending fewer conferences, with the majority of the decrease seen in the later postgraduate years. This likely represents a conscious decision to sacrifice lecture time in favor of completing clinical tasks prior to the mandatory end of shift. Furthermore, direct contact with attending physicians, and presumably teaching, did not increase during the transition. The only educational variable to improve was the amount of time spent reading medical topics. The long-term implications of these educational effects are unknown, but it appears that work-hour restrictions may have a net negative effect on resident education.

Unfortunately, an unintended consequence of the work-hour changes was that overall satisfaction with the training program decreased. The decrease in satisfaction was limited to those resident levels that experienced the transition to the new work schedule. It appears that satisfaction is less likely to be directly related to work hours because second-year residents were working less but were less satisfied. Satisfaction is more likely linked to the fact that senior residents were attending fewer conferences, spending more time working as interns, and taking care of patients without continuity of care.

As in any survey study, there are notable limitations. We do not have data on the nonresponders. They may be more burned out than those who responded. Nevertheless, the high response rate overall suggests that the data are an accurate representation of the residency program. Some may criticize that our burnout rates may underestimate the actual rate owing to surveying late in the academic year. This could have created a “light at the end of the tunnel” effect, but our pre–work restriction burnout rates were similar to other pre–work restriction programs.^22,28-31 Importantly, as with all surveys, recall bias and inaccuracies of self-report are a concern. However, we blinded the residents to the hypothesis and the nature of the validated instruments, so they were unable to consciously manipulate the results. Finally, the effect of burnout on reporting is unknown. Residents who are more burned out may tend to exaggerate or embellish their responses more so than residents who are not burned out. The difference in burnout rates noted in our study could be due to differences in response rates, if burned out residents were less likely to return the survey. Self-reported work hours in our study may also be inaccurate, but in a recent study, self-reported and observed work hours correlated well (mean r = 0.98; P<.001).²⁴ Because our study compares one year to the next, the same factors would affect both years. Furthermore, our sample size limited our study power to detect small to moderate effects, especially for categorical variables, such as high burnout.

Our study does have several strengths compared with previous studies of burnout in residents. These include the longitudinal design, the large size of our training program, and the high response rate. Using the cognitive hardiness scores, we were also able to compare the residents as a whole from 2003 to 2004. The cognitive hardiness scores, a personality test that determines susceptibility to burnout, allowed us to control for personality differences that have been shown to influence burnout sometimes more than situational variables.³¹ We believe that this study, conducted at a large academic multihospital training center, is similar to many other residency training programs and can be generalized to other academic centers.

Since the Libby Zion case, the medical profession has been more self-critical of its training system. In the June 2002 report from the ACGME Work Group on Duty Hours, “the goal of simultaneously fostering high-quality education and patient care and resident well-being provides the rationale of the ACGME standards.”¹² Our prospective study shows that the change in work hours had an inconsistent effect on our training program. The longer-term effects on residents are unknown. Will these residents provide better care as attending physicians because they are less burned out, or will the care suffer because of lack of education and commitment to their patients? Also, we do not know if residents training in other hospitals experience the same stressors as our residents. Decreasing work hours may be one intervention to decrease burnout. Further interventions need to be developed and implemented to help residents cope with a stressful period so that they may acquire the skills to provide high-quality care and sustain a long, productive career.

Correspondence: Ravi Gopal, MD, Denver Veterans Affairs Medical Center, Box 11B, 1055 Clermont St, Denver, CO 80220 (Ravi.Gopal@med.va.gov).

Accepted for Publication: June 21, 2005.

Author Contributions: Dr Gopal had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Financial Disclosure: None.

Funding/Support: This research was funded through an internal small grant from the General Internal Medicine Division of the Department of Medicine at the University of Colorado Health Science Center.