“Physician heal thyself.” This directive presents a significant challenge for persons entering the field of medicine. Acculturation into the profession includes tacit inculcation of many beliefs (1), including the idea that physicians should not become ill, and if they do, that they should continue working anyway. Physicians take these lessons to heart and are less likely than others to take sick leave (2), to have their own personal physician, and to undergo routine preventive screening tests they frequently recommend for others (3). Although they tend to die at later ages than the general population (4), physicians have higher rates of depression (3, 4), suicide attempts and completed suicide (5), and substance abuse (6).

Negative mood has been shown to impact doctors’ ability to be empathetic with patients (7), and job dissatisfaction has been linked to difficulties in providing quality, accessible health care to patients (8). Physician sleep deprivation, substance abuse, anxiety and depression have been implicated as factors contributing to medical errors and inadequate patient safety (9, 10), and extended trainee work hours result in higher rates of work-related injuries and motor vehicle accidents among residents in training (11, 12).

Given the significant potential for negative outcomes to physicians’ own health and to the health and safety of their patients, examination of the natural history of this acculturation process about physician self-care and wellness is critical to quality health care. Health and health risk behaviors as well as beliefs about self-care and treatment represent a significant part of the formation of professional identity for physicians. Assessing the development of related behavior patterns and beliefs over the continuum of medical training is important for the development of adequate education and support strategies for those entering and already in the profession.

The aim of this cross-sectional study is to describe the prevalence of health-promoting and health-risking behaviors among physicians and physicians-in-training, with particular attention to prevalence of healthy and unhealthy behaviors by gender, age, and level of training.

Subjects

Between 2001 and 2004, alumni, faculty, matriculating students, and residents affiliated with a midwestern medical school were surveyed by mail. Alumni, faculty, and residents were surveyed during the first year of data collection. New matriculants were surveyed in the summer of each year, so entering students across 4 years were included in this study. All alumni for whom addresses could be found were sent surveys. Faculty included only clinical faculty who were practicing physicians, and therefore they were combined with alumni as a single group of “attending physicians.” Previous analyses have shown that there was no difference between alumni and faculty on the survey factors of Empathy, Tolerance, Wellness, or Spirituality (13). Because wellness behaviors were the focus of the present project, these two groups were combined for our analyses.

Follow-up reminders were sent to encourage survey completion. Informed consent was implied by participants’ completion and return of the survey instrument. Details of subject recruitment and survey methodology have been reported previously (13). This study was reviewed and approved by the institution’s Committee on Research in Human Subjects.

Survey recipients included 2341 subjects, comprised of 1151 alumni, 274 faculty, 360 matriculating students, and 556 residents. Matriculating students completed the survey during the summer prior to matriculation or immediately after arrival at medical school. A total of 1037 respondents (44.3%) completed the survey, including 550 alumni (47.8%), 123 faculty (44.9%), 266 matriculants (73.9%), and 98 residents (17.6%). Respondents with missing data for race/ethnicity (N=167) and gender (N=2) were dropped from analyses because these were major variables in our final adjusted analysis. The final analyses for this article after dropping these respondents included 962 individuals, comprised of matriculating medical students (N=248), residents (N=96), or attending physicians (N=618).

Survey Design

The focus of this article is the Empathy, Spirituality, and Wellness in Medicine wellness scale (13), which includes self-reported Likert-type items describing satisfaction with career, personal health, and work or training, as well as questions about frequency of tobacco and alcohol use. Responses were rated on a Likert-type scale from 1–5, with 1 representing “always,” 2 representing “most of the time,” 3 representing “sometimes,” 4 representing “rarely,” and 5 representing “never.” Although the Wellness scale has demonstrated sufficient reliability (Cronbach alpha = 0.59), the focus of the present article was on individual items reflecting health and wellness behaviors. Group differences have previously been demonstrated for the overall Wellness factor (13), with men scoring higher than women and medical students scoring higher than residents, alumni, and clinical faculty. This article further explored these findings by examining individual items related to health behaviors.

Health-risking behaviors included alcohol and tobacco use and measures of frequency for feeling “depressed or sad” and “anxious or stressed.” The item regarding alcohol use asked respondents to describe the frequency with which they consume more than three alcoholic beverages in a 4-hour period during an average week. This definition approximates standardized definitions of binge drinking, which has been described as four or more drinks in a row for women and five or more drinks in a row for men (14).

Health-promoting behaviors included degree of social connectedness (15) as well as satisfaction with career choice and work (16, 17). Responses to two items, “having at least one person with whom to discuss major life events” and “belonging to a nonmedical organization,” addressed social connectedness. Two other items, “satisfied with choice of careers” and “find fulfillment in school or work” addressed career satisfaction. Respondents were also asked to report the frequency with which they exercised 30 minutes or more, with which they “got enough sleep” and with which they used seat belts.

Analysis

All Likert-type items were examined as both continuous and dichotomous variables, with no significant differences in the analysis. Dichotomous results (agrees or strongly agrees versus all others) are presented here for the sake of clarity. Analyses were conducted using training level as the independent variable (matriculating medical students, residents, and attending physicians). Dependent variables were responses to items relevant to health and wellness behaviors, connectedness, and work satisfaction. Among demographic characteristics, age was stratified based on size of groups; since only 10 physicians were age 60 or older, the age of 55 and older (with 27 respondents) was felt to be the appropriate cutoff for analysis. Descriptive statistics were examined for sociodemographic characteristics, and Likert-type items were analyzed using unadjusted (bivariate) logistic regression. Multivariate logistic regression was then conducted, controlling for the potential confounders of gender and race. All data were analyzed using Stata statistical software, version 9.2. Statistical significance was determined using a two-tailed p value of 0.05.

Respondents were grouped as either matriculants (25.8%), residents (10.0%), or attending physicians (64.2%). For matriculants, 47.6% were male with mean age 24.1 (range=20–50); for residents, 54.8% were male with mean age 30.7 (range=25–44); and for attending physicians, 68.8% were males with mean age 43.2 (range=27–82). 84.5% of the sample self-reported their ethnicity as “white, non-Hispanic,” with “Asian or Pacific Islander” the next most frequent at 7.0%, and this was comparable across groups (Table 1 ).

Health-Promoting Behaviors and Conditions

The mean overall wellness factor score among all respondents was 3.7 (range 2.4, 4.7; SD = 0.5). Multivariate analysis showed medical students to be nearly three times more likely than attending physicians (OR 2.8, 95% CI 2.0–3.8) and over four times more likely than residents (OR 0.6, 95% CI 0.4–1.0) to score above the overall mean (Table 2 ).

Most participants were satisfied with their personal health (77.4%), with no statistically significant difference for this variable among the three groups in either bivariate or multivariate analyses. Exercise at least twice a week was reported by 69.7% of respondents. Adjusted logistic regression revealed that medical students were over twice as likely as attending physicians (OR 2.6, 95% CI 1.8–3.8) and nearly three times as likely as residents (OR 0.9, 95% CI 0.6–1.4) to exercise at this modest level. Adjusted logistic regression revealed that the difference in getting “enough sleep” between residents and attending physicians was statistically significant (OR 0.3, 95% CI 0.2–0.5) whereas that between medical students and residents was not. Adjusted multivariate analysis showed a statistically significant difference in appropriate seatbelt use between residents and attending physicians (OR 0.5, 95% CI 0.3–0.8), but not between medical students and residents.

Stratified analysis of all individual health-promoting behavior items revealed no statistically significant differences among subgroups based on gender or age groups. Further analysis revealed a nearly linear relationship between age and predicted mean overall Wellness factor score (p<0.01), with increasing age associated with higher wellness scores. Residents in general (p<0.0001) and female residents in particular (p<0.0001) had the lowest probability of scoring above the study population mean for overall wellness.

Health-Risking Behaviors and Conditions

Smokers comprised 7.4% of the population, with no statistically significant differences between the three groups found in bivariate analysis (Table 1 and Table 2 ). Multivariate analysis suggested that medical students were nearly twice as likely as attending physicians (OR 1.9, 95% CI 1.1–3.4) to smoke at least once weekly, but no statistically significant difference between medical students and residents was noted. Stratified analysis of tobacco use revealed that the predictive probability of using tobacco decreased with increasing training level (p<0.05), and that women in general were far less likely than men to report tobacco use (p<0.0001) (Figure 1 ). Men in the 25- to 34-year-old age group were most likely to report tobacco use of any kind, with a predicted probability of 0.18 that they would use tobacco (p<0.0001).

Drinking behavior differed by training and gender. Among all respondents, 22.1% reported drinking more than three alcoholic beverages within a 4-hour period at least once weekly, with medical students nearly four times as likely as attending physicians (OR 3.9, 95% CI 2.7–5.6) and over six times as likely as residents (OR 0.6, 95% CI 0.3–1.3) to report heavy drinking. Stratified analysis of alcohol use revealed that nearly one quarter of those who reported drinking heavily did so 2–6 times weekly (Figure 1 ). Residents were least likely, and medical students most likely, to report heavy levels of drinking (p<0.0001). This prevalence pattern held up across genders, but women in general were less likely to report heavy drinking patterns than were men (p<0.0001). Both men and women in the under-25-year-old age group were the most likely to report such levels, with men in this age group having an adjusted probability of heavy drinking of 0.56, and women of 0.30 (p<0.0001).

Reported feelings of depression were more common among residents and attendings than entering medical students. In multivariate analysis, medical students were half as likely (OR 0.5, 95% CI 0.3–0.7) as attending physicians to report this finding. Stratified analysis of depressed mood revealed that residents were most likely to report frequent feelings of sadness or depression regardless of gender (adjusted probability 0.38), but that female residents had an adjusted probability of 0.55 for having frequent depressed feelings, compared to male residents at 0.24, and to all other groups (p<0.0001). When the population was stratified by both age and gender, the oldest physicians, 55 years old and older, had among the highest probabilities of reporting depression, at 0.49 for both males and females. Only females between the ages of 25 and 34 were higher, at an adjusted probability of 0.52 (p<0.001).

Feeling “anxious” or “stressed” was common in this population, with 74.4% reporting these feelings “more than rarely,” or a 1, 2, or 3 on the Likert scale. The final adjusted regression showed medical students were half as likely as attendings (OR 0.5, 95% CI 0.3–0.6) to report these feelings “more than rarely,” with no statistically significant difference between the likelihood of residents (OR 0.7, 95% CI 0.4–1.2) to do so compared with either of the other two groups. Stratified analysis of anxious mood states revealed no statistically significant differences in this variable based on gender or age groups.

Social Connectedness and Work Satisfaction

Respondents to this survey reported high levels of social connectedness (Table 1 ). Adjusted multivariate regression analysis showed that medical students were far more likely than attending physicians to have someone with whom they could discuss major events (OR 11.6, 95% CI 1.6–86.5), but no significant difference was found between medical students and residents or residents and attendings for this item. Medical students were more likely than attendings (OR 1.7, 95% CI 1.0–2.7) to belong to a nonmedical organization, but no other group differences were found. For the aggregate social connectedness item, medical students were more likely than attendings (OR 6.8, 95% CI 1.6–28.7) and residents less likely than medical students (OR 0.5, 95% CI 0.2–1.0) to exhibit a high degree of overall social connectedness. No statistically significant differences based on gender or age groups were found for any of these items.

The level of career satisfaction in our study population was also relatively high. Medical students were significantly more likely than attendings (OR 11.6, 95% CI 3.6–37.2) and residents (OR 1.0, 95% CI 0.5–2.1) to report high career satisfaction. Medical students were three times as likely as attendings (OR 3.0, 95% CI 1.2–7.9) and six times as likely as residents (OR 0.5, 95% CI 0.2–1.0) to report finding fulfillment in school or work; they were nearly seven times as likely as attendings (OR 6.8, 95% CI 1.6–28.7) and over 13 times as likely as residents (OR 0.5, 95% CI 0.2–1.0) to exhibit a high degree of overall career satisfaction (Table 2 ).

The goal of this project was to describe the prevalence of health related behaviors at various levels of medical training and to explore the relationship between sociodemographic variables and these findings. Because of findings showing increased rates of depression, suicide, and substance abuse among physicians (3–6), it is important to begin to examine the prevalence of these and other related wellness behaviors in the medical community. Overall, we found that matriculating medical students scored highest on the Wellness factor from the Empathy, Spirituality, and Wellness in Medicine (ESWIM) survey. Medical students were more satisfied with their career choice, were more likely to belong to nonmedical organizations, exercised more regularly, and reported getting more sleep on a regular basis. Residents and attending physicians reported lower scores on these variables and attending physicians reported feeling more stressed or anxious. These results are consistent with the interpretation that students are healthier prior to beginning medical school and that less healthy behaviors become evident by the time they begin residency. However, this is not a longitudinal study, and no inferences about causation may be made from these data.

Matriculating medical students were the most satisfied with their choice of career, followed by residents and then by attendings, suggesting that satisfaction may decline with experience in the career. It is important, however, to note that overall satisfaction was high for all groups. Of course, matriculating students could only report on satisfaction based on expectations because they had not yet had any experience being a physician, and so their responses were theoretical. Residents and attending physicians were able to report satisfaction based on actual experiences. Students should (and did) score especially high because they were excited about beginning their medical career.

Matriculating medical students reported lower levels of depression and were more likely to have someone with whom to share major life events, which is encouraging because the social demographics in the United States are showing an alarming decrease in the number of confidants for adults in general, but also is consistent with data showing that confidant network size decreases with age (18). Matriculants also reported being less stressed. This is predictable given that they completed the Empathy, Spirituality, and Wellness in Medicine survey at entry into training, so the major stressors that accompany medical school had not yet begun for them. However, this study is not longitudinal and does not address the question of when depressed feelings become more common or cause dysfunction. Nonetheless, these results suggest that it would be advantageous to provide easy access to counseling for students who become depressed or anxious or who lose social contacts during training. Other studies have shown that depression and suicide are uncharacteristically high for medical students (19). Therefore, our findings that depressed feelings are even higher among residents and attendings suggest that this is a very important area for continued study.

Both tobacco and alcohol use were more common among entering medical students than among older and more experienced medical persons. This trend is consistent with the overall U.S. trend showing that tobacco use and binge drinking decline with age (20). Thus, although medical training is stressful, it does not appear that residents and attendings commonly turn to drinking alcoholic beverages or to smoking to relieve that stress.

Health-supporting behaviors were most common for matriculating medical students and least common among residents, probably because matriculating students had more time for such things as sleep, exercise, and social connections at the time that they completed the surveys. It should be noted that medical students’ responsibilities for studying (though not for taking call until later in their training) and their increased propensity for engaging in alcohol use may contribute to less time for sleep. Sufficient time for these behaviors decreases dramatically in residency and remains low after entering practice. These data were collected prior to the advent of the 80-hour work rule for residents, which was implemented in 2003. It is possible that residents since that time have more personal time for engaging in social activities and healthier behaviors than those residents completing this survey. We also found that these behaviors were related to decreased depression and anxiety, suggesting that residents and attendings may be at greater risk for these psychological problems in part as a function of their decreased time for healthy behaviors. There is evidence that residents are particularly receptive to training about these issues (21). It is also possible that people who are less depressed and anxious make time for these behaviors, and that residents and attendings are more depressed and anxious as a function of their job.

Wellness behaviors and perceptions may have a direct impact on the quality of care that physicians provide their patients. Physicians who live healthier lifestyles have been found to be more likely to discuss these lifestyles with their patients and to encourage their patients to behave in healthier ways (3, 22). Also, Frank and colleagues (23) found that matriculating medical students reported being more likely to believe that it was important to counsel future patients about exercise if they themselves were actively involved in an exercise regimen. Thus, wellness activities among medical professionals clearly are important both for themselves and for their patients.

From stratified analyses, it appears that female residents may be at risk for lowest overall wellness, although males are at higher risk for tobacco and alcohol use than females, with the highest-risk group being male medical school matriculants. Female residents, followed by attending physicians of both genders, are at risk for more frequent episodes of feeling depressed or sad (Figure 1 ).

Limitations

Although nearly 75% of matriculating medical students completed the Empathy, Spirituality, and Wellness in Medicine survey, only about 50% of attending physicians and less than 20% of residents did so. It is possible that the most anxious, depressed, or unhealthy residents and physicians chose not to complete the questionnaire, although their inclusion would be expected to yield larger group differences than we already see. In addition, the respondents across all groups were primarily white, making it possible that these results will not apply to other populations. Because the survey data are self-reported, these data are subject to both recall and rater bias. The medical students and residents were affiliated with a single medical school, and it would be illuminating to determine whether similar results arise for students from different geographical and learning backgrounds. Relationships between health-promoting behaviors, health-risk behaviors, and age or maturity may confound the cohort-based analysis of both individual items and of the Wellness Scale itself. Because we did not analyze nonrespondents, the potential that there are group differences between those who did and those who did not complete the survey is possible. The percentage of male respondents was higher among attending physicians than among the other groups, perhaps reflecting a cohort effect based on more recent entry of women into the physician workforce in larger numbers. Finally, these data are cross-sectional rather than longitudinal and no inferences may be made about causation or time of onset for health-promoting or health-risking behaviors from this analysis.

Implications

The findings from this study demonstrate that matriculating medical students overall engage in healthier behaviors than do residents or attending physicians, except for slightly increased tobacco use and binge-type drinking. It is possible that the rigors of medical school training prevent students from continuing these healthy behaviors, so that by the time they are residents they exercise less, sleep less, and spend less time in social activities outside of medical school. If physicians do not engage in these healthy behaviors, they are less likely to encourage such behaviors in their patients (3, 21), patients are less likely to listen to them even if they do talk about it (24), and students they train are more likely to exhibit anger, depression and anxiety (25), perpetuating the cycle. Addressing the highest-risk groups, especially those illustrated in the stratified analysis (Figure 1 ), may be particularly useful in targeting interventions with the highest chance for successful behavior change. For the sake of both doctors and their patients, it is essential to address the importance of wellness, prevention, and self-care measures for physicians during medical training and throughout their careers.

Funding for this study was provided in part by an award from the National Institute for Healthcare Research.