Changes in Career Satisfaction Among Primary Care and Specialist Physicians, 1997-2001 FREE

A confluence of forces has changed the practice of medicine in unprecedented ways during the past decade. There is increasing focus on, and concern about, the quality of medical care,^1- 2 financial constraints are being applied more frequently and stringently, and the practice and management of medical care have become increasingly centralized and dominated by managed care and physician organizations.^3- 5 Whereas physicians once practiced primarily alone or in small autonomous groups, they now are more likely to practice in large groups and are increasingly subjected to profiling, administrative requirements, and preapproval for procedures and treatments. Evidence suggests that some physicians are becoming unhappy in this environment.⁶

Understanding trends in physician career satisfaction and how changes in the practice environments of physicians affect their career satisfaction is important for several reasons. First, physician satisfaction is associated with quality of care, particularly as measured by patient satisfaction.^7- 8 Second, dissatisfied physicians are more likely to leave the profession and discourage others from entering.^9- 11 This is disruptive to patient-physician relationships, costly to physician practices, and could ultimately diminish the overall quality of care if potentially outstanding physicians are dissuaded from choosing this career.¹² Finally, career dissatisfaction might be one manifestation of physicians' perceptions of problems in delivering high-quality care to their patients.¹³

To examine trends in career satisfaction, we administered surveys 2 years apart to a nationally representative sample of primary care physicians (PCPs) and specialist physicians drawn from local health care markets throughout the country.^13- 14 The 60 nationally representative sites include both large and small metropolitan areas, rural counties, and areas with both high and low penetration of managed care. In addition, we also used a longitudinal approach that overcomes some inherent limitations of cross-sectional studies. We had 3 primary questions. First, is overall career satisfaction changing over time? Second, how do changes in physician satisfaction vary by market and by specialty? Third, to what extent are changes in physicians' practice environments and local markets associated with changes in their satisfaction?

Data for this study are from the first 3 rounds of the Community Tracking Study (CTS) Physician Survey, a series of 3 nationally representative telephone surveys of physicians conducted in 1996-1997 (round 1: 12 385 respondents; 65% response rate), 1998-1999 (round 2: 12 280 respondents; 61% response rate), and 2000-2001 (round 3: 12 389 respondents; 59% response rate) for the Center for Studying Health System Change.¹⁴ The survey samples were designed to be representative of physicians directly providing patient care in the continental United States, as well as in selected communities. Each survey round used a complex sample clustered in 60 randomly selected sites and a small, independently drawn, unclustered national sample comprising about 10% of all interviews used to increase the precision of national estimates.¹⁵

In each site, we sampled physicians from the master files of the American Medical Association and the American Osteopathic Association. The sample included active nonfederal office- and hospital-based physicians who spent at least 20 hours per week in direct patient care, but excluded specialties such as radiology, pathology, and anesthesiology. We oversampled PCPs, including those in family practice, general practice, general internal medicine, internal medicine/pediatrics, pediatrics, and geriatric medicine. The second and third rounds of the survey included portions of physicians sampled in the previous round, as well as new physicians. This approach created 2 panels of physicians, the first interviewed in both rounds 1 and 2 and the other interviewed in both rounds 2 and 3.

To create the first panel (round 1-round 2), 9353 of the 12 385 respondents (76%) to the first round were contacted in the second round. Of these, a total of 915 had become ineligible because they had retired, worked fewer than 20 hours per week, or were unlocatable. From the remaining 8438 we received 6569 responses, for a panel response rate of 78%. A further 650 were eliminated because they changed sites or PCP status between rounds or were not practicing in the 60 sites, leaving a final sample of 5919 (3476 PCPs and 2443 specialists). To create the second panel (round 2-round 3), 11 394 of the 12 280 respondents to the second-round survey were contacted in the third round. Of these, a total of 671 had become ineligible. From the remaining 10 723 we received 7723 responses, for a panel response rate of 72%. A further 976 were eliminated because they changed sites or PCP status between rounds or were not practicing in the 60 sites, leaving a final sample of 6747 (4257 PCPs and 2490 specialists). Some physicians were sampled in both panels. Item nonresponse for both rounds of the survey was very low, typically less than 3%, and less than 0.5% for the dependent variables. A few independent variables were imputed using hot-deck procedures.¹⁶ Greater detail about the surveys is available elsewhere.^14- 17

In the first section of the survey, physicians were asked: "Thinking very generally about your satisfaction with your overall career in medicine, would you say that you are currently very satisfied, somewhat satisfied, somewhat dissatisfied, very dissatisfied, or neither satisfied nor dissatisfied." These responses were coded on a scale from 5 to 1 with "neither satisfied nor dissatisfied" being coded as 3. In addition to reporting overall satisfaction levels, we also report satisfaction levels by market for 12 of the markets that were selected randomly for more intensive study. These sites had survey samples large enough (approximately 500 physicians per round) to draw conclusions about changes over time and were the subject of in-depth site visits that were conducted concurrently with each round of the survey.

We determined changes in satisfaction for the panel respondents by calculating the differences in reported satisfaction for participants in consecutive rounds of the survey, with the resulting variable ranging from +4 to −4.

The literature suggests that changes in physician career satisfaction may be influenced by 3 general factors: changes in the physician's practice environment, including the income they derive from practice and hours worked; changes in physicians' autonomy as measured by perceptions of their ability to provide high-quality care to their patients and to obtain needed medical services for them; and changes in the the local market.

Independent variables used to assess changes in physician satisfaction for the panel respondents were generally constructed either as changes in a continuous variable or as sets of categorical variables. For continuous variables, the preceding value (from either round 1 or round 2) was subtracted from the later value (from either round 2 or round 3). For categorical variables, dummy variables were constructed indicating an increase or decrease in the variable, with those physicians with no (or small) change serving as the reference group.

We measured changes in several attributes of physicians' practice environments, including practice size, ownership, and involvement with managed care. The survey assessed size (dichotomized to <10 or ≥10 physicians) and type of practice (physician group, staff/group model health maintenance organization [HMO], medical school–based practice, hospital-owned practice, or another setting). After exploring alternative specifications, we created dummy variables indicating whether physicians changed from a larger to smaller practice setting or vice versa.

Changes in ownership were captured by 2 dummy variables indicating whether the physician was a full or partial owner who became a nonowner or vice versa. A dichotomous variable indicated whether the physician's practice had been acquired in the preceeding 2 years.

Changes in the physicians' patient population were assessed with 2 continuous variables measuring changes in the percentage of practice revenue from Medicaid and Medicare.

Changes in the practice's overall exposure to financial influences of managed care were examined using continuous variables measuring changes in the proportion of total practice revenue derived from managed care and changes in the proportion of managed care revenues received in the form of capitation.

To measure effects of various administrative care management strategies, physicians were asked, on a 6-point scale ranging from "very large" to "no effect," the extent to which practice profiling, patient satisfaction surveys, and guidelines affected their practice of medicine. Because responses to these questions varied little over time, and because these variables were not significant in preliminary analyses, they were not included in the final analysis.

Changes in work effort were assessed by categorical variables, indicating whether the number of hours worked the previous week changed from fewer than 60 hours to more than 60 hours, or vice versa. Change in income was included as a categorical variable indicating that income increased or decreased by at least 5% after accounting for inflation. A measure of income per work effort based on hours worked in the prior week was correlated with the income measure, and was therefore not included in final analyses.

Professional autonomy refers to individuals' ability to control the terms and content of their work. Based on the work of the Society of General Internal Medicine Task Force on Physician Satisfaction, we identified 2 general influences on clinical autonomy: control over work and ability to provide needed services.^13,18- 19 To measure clinical autonomy, we created a summary score based on responses to a set of 5 questions assessing these factors. Respondents first were asked their level of agreement with a general statement regarding the ability to provide high-quality care and with 3 statements related to aspects of the process of care delivery: whether they had adequate time with patients, freedom to make clinical decisions, and the ability to form continuing relationships with patients over time that promote the delivery of high-quality care. Response categories included agree strongly, agree somewhat, neither agree nor disagree, disagree somewhat, and disagree strongly. An exploratory factor analysis indicated that there was a single underlying construct to the questions. Since the factor loadings were almost equal, we created the index by summing up the responses. The resulting index was highly internally consistent (Cronbach α, .82).

Primary care physicians and specialists were also asked to rate the appropriateness of the level of patient complexity that they were expected to address prior to or at the time of referral, respectively.²⁰ We constructed categorical variables indicating increases or decreases in each of these variables between the 2 rounds of the survey.

We measured ability to provide needed services with a series of 5-point questions that were combined to construct 3 indices by averaging component variables: ability to obtain outpatient services ("index 1," including 3 items related to obtaining referrals to specialists, ancillary services, and diagnostic imaging), ability to obtain hospitalizations ("index 2," including 2 items related to obtaining hospitalizations and length of stay), and ability to obtain mental health services ("index 3," including 2 items related to obtaining outpatient mental health treatment and inpatient mental health hospitalizations). The mental health questions were only asked of PCPs, psychiatrists, and obstetricians/gynecologists. Change scores were then calculated for each index. A dummy variable was created to account for physicians who did not routinely need the applicable services (eg, some specialists rarely hospitalize patients).

To assess changes in measures of managed care penetration, we created variables by subtracting site-level averages of the percentage of physician revenue from managed care, and also the percentage of physician revenue that was capitated.

For descriptive analyses, we used t tests to examine changes in continuous variables and the χ² test for dichotomous and categorical variables. We computed descriptive information on overall satisfaction using the cross-sectional information from the 3 rounds of the survey. Because the characteristics of the sample changed somewhat over time, we computed regression-adjusted mean satisfaction levels for rounds 2 and 3 based on the composition of the first round of the survey.

Longitudinal Analyses. We investigated determinants of career satisfaction by evaluating changes in satisfaction for the longitudinal panels of physicians. Cross-sectional studies of physician career satisfaction can be biased because unobserved physician preferences may affect their practice choices (eg, involvement with managed care, size of practice) as well as their career satisfaction. In a longitudinal first-differences model, in which the dependent and independent variables are mostly expressed in terms of their changes over time, unobserved physician preferences, as well as observable physician characteristics, are held constant because we are only looking at changes affecting a given physician over time. The inclusion of baseline career satisfaction in the regression models also allows us to control for possible bias due to floor and ceiling effects as well as for the tendency of individual physicians to rate their satisfaction more or less highly.

We used ordinary least-squares regressions to investigate multivariable relationships. For these models, we combined both panels into a single data set because our model suggests that the determinants of satisfaction would not change appreciably over time. We estimated separate models for PCPs and specialist physicians because we thought that changes in the practice environment and autonomy might be expected to affect generalists and specialists differently. We first examined bivariate relationships. Variables that were not significantly related to changes in satisfaction at P<.10 in either bivariate or multivariable models were eliminated from the models. This did not result in significant changes for any of the remaining coefficients. Because they were a primary interest of this study, variables related to exposure to managed care at both the individual and market level were also included in the final model. All analyses were done using SUDAAN v8.01 (RTI International, Research Triangle Park, NC) to control for the complex sample design and use weights created for the panel.¹⁴ The SUDAAN software also accounts for the nonindependence of observations when the same physician is included in both panels.

Each category of characteristics (for instance, changes in practice environment or changes in physician autonomy) was entered individually and then in combination to examine both direct and indirect effects. For instance, we speculated that changes in satisfaction related to managed care exposure might be mediated by changes in professional autonomy, so we examined these groups of variables individually and in combination. For each subgroup, variables were first examined for collinearity. Correlations were below levels at which multicollinearity would be considered likely (r<0.30). We report results from the model including all categories of independent variables because individual coefficients remained robust when alternative specifications were tested.

We also performed additional exploratory multivariable analyses at the site level, using the combined panel data sets. For these analyses, we created site-level means of physician satisfaction as well as independent variables related to managed care exposure and distribution of patients across public and private payers. We also used information from Interstudy²¹ and the Bureau of Health Professions Area Resource File²² to examine physician, hospital, and health plan concentration.

The characteristics of the respondents across the 3 rounds of the survey are presented in Table 1. Among primary care physicians, 23.6% were women in 1997 and this increased to almost 31% in 2001 (P<.01). Fourteen percent of specialists were women in 1997, as were 18.7% in 2001 (P<.01). A slightly higher proportion of PCPs were international medical graduates (21.3%-24.9% [1997 and 2001, respectively], compared with 18.5%-18.8% for specialists). Among PCPs in 1997, 33.1% were internists, 21.0% were pediatricians, and the remainder were either family physicians or general practitioners. By 2001, the proportion that were internists had increased to 37.3%, with an accompanying decrease in the proportion of family physicians to general practitioners. Among specialists, the share of medical specialties increased from 40.3% in round 1 to 47.6% in round 3, while the surgical specialties decreased from 36.7% to 31.3% (P<.01).

Physician career satisfaction levels declined marginally between 1997 and 2001 (Table 2), although there did not appear to be a clear trend over that time period. In addition, modest declines noted between 1997 and 1999 appeared to moderate by 2001. There were no significant differences between levels in 1999 and 2001. For instance, 42.4% of PCPs and 43.3% of specialists were very satisfied in 1997, compared with 38.5% and 41.4%, respectively, in 1999 (P<.001 for PCPs and P<.01 for specialists). For PCPs, however, this was accompanied by an offsetting increase in the percentage of physicians who were somewhat satisfied (P<.001). The proportion of specialists who were somewhat dissatisfied also increased between 1997 and 1999, from 12.9% to 14.8% (P<.05).

We also examined changes in career satisfaction for physicians included in the 2 panel samples (Table 3). Overall, 25.7% of physicians reported decreased satisfaction between 1997 and 1999, and this proportion was similar for both PCPs and specialists. Only 17.9% reported increased satisfaction. In contrast, between 1999-2001, approximately equal numbers of physicians in the panel reported decreased (21.4%) and increased (19.7%) satisfaction.

These overall means mask significant site-level variation. Across the 12 high-intensity sites, the percentage of physicians who were somewhat or very dissatisfied ranged from a low of 8.8% of physicians in Lansing, Mich, in round 2 to a high of 34.2% of physicians in Miami, Fla, in round 1 (Figure 1), and these were reflective of the range of variation seen across the full sample of 60 sites. More physicians in 7 of these 12 markets were dissatisfied in 2001 vs 1997, with the biggest change occurring in Syracuse, NY, where the percentage of somewhat or very dissatisfied physicians increased from 14% to 21% during the time period of the study. Among the markets with the highest proportion of dissatisfied physicians in 1997 (including Newark, NJ [24%], Orange County, Calif [27%], and Miami, Fla [34%]), dissatisfaction levels appeared to moderate by 2001. Multivariable analyses to understand further changes in satisfaction at the market level revealed only a single significant predictor. Increasing proportion of capitated managed care revenue in a market was associated with decreased satisfaction (P = .01) among physicians in that market.

Changes in Practice Environment. Multivariable linear regression results for the 7733 PCPs and 4933 specialists included in the panel analyses are presented in Table 4. Of note, there were few significant interactions between the individual predictors and a dummy variable indicating the panel the physician belonged to, suggesting it was reasonable to combine them into a single regression. Primary care physicians and specialists who became owners of their practices were more likely to report increased satisfaction (P<.01 and P<.001, respectively) while PCPs who reported increases in the number of hours they worked, from less than 60 to more than 60 hours per week, reported decreased satisfaction (P<.01). Income was not related to changes in specialist satisfaction, but changes in satisfaction among PCPs were significantly related to income (P<.01). Changes in exposure to managed care were generally not related to changes in satisfaction.

Changes in Professional Autonomy. The most consistent predictors of changes in both PCP and specialist career satisfaction were changes in measures related to clinical autonomy. Primary care physicians who reported increases in the level of patient complexity that they were expected to address without referral were more likely to report decreased satisfaction (P<.01). Similarly, both PCPs and specialists who reported more difficulty obtaining high-quality outpatient services (P<.01) and inpatient services (P<.05) were also more likely to be report decreased satisfaction with their careers. Finally, changes in the clinical autonomy scale, which measures physicians' ability to provide high-quality care, maintain continuing patient relationships, adequacy of time with patients, and freedom to make clinical decisions, were strongly related to changes in satisfaction for both PCPs and specialists (P<.001 for both).

Changes in Market-Level Characteristics. Primary care physicians in markets with increased proportions of managed care revenue in the form of capitation were significantly less satisfied (P<.01), while specialists' satisfaction was not related to changes in the market-level measures of managed care. Lastly, we also examined predictors of changes in satisfaction at the market level. Both increased market-level managed care penetration and higher proportions of capitated managed care revenue were associated with decreases in market-level physician satisfaction.

Despite a lack of systematic data, many observers perceived a decline in physician career satisfaction. As physician practice continues to evolve, it becomes increasingly important to understand how satisfaction is changing and whether managers and policy makers at the local or national level can influence these changes.

In this national study of satisfaction among a large representative sample of physicians, we find several notable results. First, our findings demonstrate that overall levels of career satisfaction among physicians over this time period did not change dramatically, although there appeared to be a slight decline between 1997 and 1999. Second, national-level data mask significant variation in local markets. These findings suggest that changes in physician satisfaction may be significantly influenced by local phenomena. Third, we found that changes in physicians' clinical autonomy, including their ability to obtain needed outpatient and inpatient services, were the most consistent and powerful predictors of changes in their levels of satisfaction over time. Change in income was a significant predictor only for PCPs, and change in practice ownership was significant for both PCPs and specialists. A number of other possible influences, such as managed care penetration within physicians' practices, had little apparent effect.

Changes in autonomy cannot be attributed simply to the spread of managed care, since managed care variables generally were not important predictors of change in satisfaction, even in models that did not include the change in autonomy measures.²³ Instead, more subtle factors seem to be at work. These may be related to unmeasured differences in managed care organizations in different communities, or in how physicians' organizations respond to managed care. In addition, it could also reflect changes in the intensity of managed care practices over the time period, such as those related to increasing use of certain types of capitation, which we were unable to measure.

Previous studies of the trends in and determinants of physician satisfaction relied on cross-sectional data,^13,24- 25 focused on selected markets or specialties,^26- 27 or failed to identify the full range of influences on satisfaction. The few existing longitudinal studies of career satisfaction use old data, have low response rates, or rely on sequential cross-sectional data, thus failing to fully control for personal characteristics of the physicians.^6,26,28 Furthermore, previous research has not examined whether changes in satisfaction vary across local health care markets. If differences do exist across markets, this might present opportunities for managers and policy makers to understand and target factors important to physician satisfaction in particular markets. This study addresses many of these methodological limitations and takes advantage of the ability to use representative cross-sectional data as well as longitudinal information to analyze the determinants of changes in satisfaction.

Our findings related to variation in local markets merit attention because cross-sectional satisfaction levels across markets and changes in satisfaction levels both varied substantially across markets. To our knowledge, no previous study has documented these marked differences in satisfaction across markets. Many have observed that "all health care is local" and these findings appear consistent with this view.²⁹ We attempted to explain changes in satisfaction at the site level through multivariable models. However, the small number of sites, the small number of respondents in some sites, and resulting limits in statistical power precluded meaningful analysis using the market as the level of analysis. We did find, however, that increased levels of capitation within managed care were associated with decreased satisfaction at the site level, despite there being no relationship at the individual level. The latter finding suggests that increasing capitation at the market level may be a marker for more aggressive managed care practices that can lead to subtle changes in practice norms and that may limit physicians' opportunities to escape these managed care influences by moving to different practices in the same market. This effect may operate through market-level influences on physicians' professional autonomy.

Other measures of market characteristics, such as hospital concentration and physician density, did not prove to be important predictors of changes in physician satisfaction. This might be because local markets differ substantially in ways that are difficult to measure. These range from subtle variations in practice norms and patient expectations to more basic differences in the structure and organization of medical groups and local contracting practices and the rate at which these factors are changing.^30- 31 Findings from recent site visits, however, do suggest some possible explanations for some of the more dramatic changes between 1999-2001. For example, in Syracuse, NY, which experienced a significant increase in dissatisfaction, increased health plans leverage that resulted from the merger of the 2 largest health plans in the market might have had deleterious effects on physician payment and autonomy. In contrast, in Newark, NJ, which experienced a significant decline in dissatisfaction, the state moved aggressively to adopt new more stringent oversight of HMOs and to mandate prompt payment requirements following the high-profile collapses of 2 health plans in 1998.^32- 34

Our data are subject to several limitations. First, we used only a single measure of general satisfaction. Our measure, however, is robust and similar questions have been used in numerous other studies in the past. Second, while to our knowledge our sample is the largest nationally representative longitudinal study of physicians, the sample sizes for many individual markets were relatively small. Consequently, with the exception of the 12 high-intensity sites, we were unable to make precise estimates of changes at the individual market level and lacked statistical power to identify actionable influences. This is, however, the first study to document these important differences across markets and this area should be a high priority for future research.

In summary, our data suggest that declining physician career satisfaction is real but variable in both time and space. Physician satisfaction seems to have declined from 1997 to 1999, but not appreciably between 1999 and 2001. However, this was not true for all markets. Constraints on physicians' clinical autonomy seem to have been the most important influences on changes in career satisfaction among physicians, but other factors, such as increases in hours worked and changes in practice ownership status, have also played a role. Future research needs to explore the determinants of physician satisfaction in different markets and should highlight local interventions that might be pursued to maintain or improve physician satisfaction.