Internal medicine resides at the interface between generalist and specialty medicine. Although internal medicine is the largest primary care specialty, about two thirds of internists have chosen to subspecialize during the past several decades [7]. Medical subspecialties seem to be strongly affected by changing market conditions; the workforce in many specialty areas currently seems to be highly oversupplied. For example, 3.3% to 6.2% of physicians completing training in 1994 who pursued careers in medical subspecialties had not found full-time jobs in their specialties in 1995 (excluding geriatric medicine, for which an oversupply did not exist) compared with only 1.5% of those practicing general internal medicine [8]. The number of recruitment advertisements for internal medicine specialists decreased by 75% between 1990 and 1995, the highest decline of any career category [9].

In response to the growing number of internal medicine subspecialists, the Federated Council for Internal Medicine, representing a variety of internal medicine organizations, has advocated the goal that 50% of internal medicine graduates enter general medicine practice [10, 11]. Internal medicine residency programs have expanded primary care tracks to encourage trainees to pursue generalist careers, but no empirical evidence has shown whether the desired 50% goal is being achieved.

The changing trend in internal medicine subspecialization seems to be an important marker in the wider generalist-specialist question, and several previous studies have identified factors associated with generalist versus specialist career decisions. One recent survey of third-year internal medicine residents suggests that those in training programs located in markets with higher managed care penetration were less likely to pursue careers as specialists [12]. Personal characteristics (age, ethnicity, and sex) [13, 14] and undergraduate medical school characteristics (public compared with private ownership and research intensiveness) [15] have also been associated with generalist versus specialist career choice.

We expected that the market environment in the area where general internal medicine training took place would influence residents' career decisions. Many market characteristics may influence career choice, but we focused on one-local managed care presence-that we believed was particularly important. We thus sought to determine whether the increased prevalence of managed care has catalyzed a shift away from specialty medicine toward generalism by influencing the career decisions of individual medical trainees. We hypothesized that general internal medicine residents in markets with increasing managed care penetration would be less likely to subspecialize. To gain insight about individual residents, we analyzed a longitudinal census of individual graduate medical education records maintained by the AAMC. The present analysis extends the one previous demonstration of the influence of managed care on generalist versus specialist career choice by controlling for U.S. Census division, several individual characteristics, and characteristics of undergraduate medical institutions that have previously been associated with the decision to pursue a generalist career.

Methods

Top Methods Results Discussion Author & Article Info References

Study Sample

Information about graduate medical education was obtained from the AAMC's Graduate Medical Education Tracking Census, which is collected by an annual survey of directors of medical education at all U.S. graduate medical education training locations [16]. These data document the entire graduate medical education training history of all residents and fellows enrolled in graduate medical education in the United States, including U.S. and international medical graduates. Completion of general internal medicine training was determined by tracking each resident's graduate medical education data until at least 3 years of general internal medicine training were completed and specialty training was begun or graduate medical education was discontinued (presumably so that the resident could enter practice). The activity after the completion of general internal medicine training was the outcome.

We tested the influence of local managed care penetration on subspecialization decisions, controlling for region, individual characteristics, and undergraduate medical school characteristics. It was therefore necessary to restrict the sample to only those residents for whom data on all of these variables were available. In 1993, 4922 residents completed general internal medicine residency training at U.S. training sites. Residents who completed graduate medical education training in areas where data on managed care penetration were not available (81 in Puerto Rico or rural areas; 53 in metropolitan statistical areas not available in the market database), and 1763 graduates of international, Canadian, or U.S. osteopathic undergraduate medical schools (for whom individual characteristics and information on undergraduate medical education were not available) were eliminated from the sample. We excluded an additional 424 residents because data on National Institutes of Health (NIH) awards to the undergraduate medical school were missing, and we excluded 2 residents because of missing information on ethnicity. To allow adequate and uniform follow-up, we excluded 181 general internal medicine residents who did not complete graduate medical education in consecutive years and 61 residents who completed more than 4 years of general internal medicine training. Finally, 94 residents were excluded because they enrolled for training in specialties other than internal medicine (for example, dermatology and emergency medicine). Analyses included the remaining 2263 residents who completed general internal medicine training in 1993.

Outcome Variable

Graduate Medical Education Tracking Census data records were compiled into longitudinal profiles that contained historical information about the graduate medical education fields in which residents were enrolled each year. Residents who enrolled in internal medicine subspecialty training programs after completing 3 or 4 consecutive years of general internal medicine training were categorized as electing to subspecialize. Residents who were not enrolled in graduate medical education the year after they completed general internal medicine training were categorized as entering practice, presumably as generalists. The outcome variable in this study, subspecialization, was coded as 1 for residents who pursued subspecialty training and as 0 for residents who entered practice.

Explanatory Variables

Managed care penetration was indexed by a health maintenance organization (HMO) penetration variable derived from the Interstudy Competitive Edge Database, 6.1 Regional Market Analysis [17]. We calculated 1993 HMO penetration by dividing the HMO enrollment in a metropolitan statistical area by the population in that area, after adjusting the 1995 values to 1993 levels using change variables present in the Interstudy data set (1993 Interstudy data are not available at the metropolitan statistical area level). The HMO penetration index ranges from 0 to 1; values closer to 1 indicate greater HMO penetration (the actual range in our study was 0 to 0.64). The Interstudy HMO penetration variable is superior to other, similar calculations because it adjusts the numerator for HMO cross-coverage in contiguous metropolitan statistical areas. The HMO penetration variable was linked to the metropolitan statistical area where each resident in the study sample completed his or her general internal medicine residency.

Sites where general internal medicine training was completed were grouped into nine U.S. Census divisions [18]. Individual characteristics for graduates of U.S. medical schools were taken from the AAMC's Student and Applicant Information Management System (SAIMS) database, a repository of information about all applicants, matriculants, and graduates of U.S. allopathic medical schools. Individual characteristics included sex, ethnic group (white or other), age at first general internal medicine year, and average Medical College Admission Test (MCAT) subject test score.

The SAIMS database also provided information about ownership (public or private) of U.S. medical schools. Research intensiveness was indicated by the amount of NIH research funds awarded to each medical school, averaged over the years 1988 to 1991. The NIH award variable was log-transformed to correct for extreme values. This information was taken from the Information for Management, Planning, Analysis, and Coordination (IMPAC) system maintained by the Division of Research Grants at the NIH [19]. School information was linked to the last medical school attended by U.S. allopathic medical graduates in the study sample.

Statistical Analysis

We used logistic regression analysis to test the effect of HMO penetration on the odds of subspecialization, controlling for Census division, individual characteristics, and undergraduate medical school characteristics. As stated previously, individual characteristics included sex, age, and average MCAT score; medical school characteristics included average amount of NIH award for 1988 to 1991 and type of ownership (private or public). We fit a series of nested logistic regression models to the data to determine whether HMO penetration in the metropolitan statistical area where residents completed general internal medicine training influenced the decision to subspecialize above and beyond the effects of regional, individual, and medical school variables. Models were estimated by using the "logistic" procedure of Intercooled Stata for Windows 95, Release 5.0 [20]. Standard errors and P values were adjusted for clustering on metropolitan statistical area by using the "cluster" option. Residents for whom data on any variable in the most complex model were missing were excluded from all logistic regression analyses. This ensured that the same cases were included in all models in the testing hierarchy.

Results

Top Methods Results Discussion Author & Article Info References

Table 1 shows descriptive information for the variables included in the analysis. The lower portion of Table 1 reports the percentage of residents who subspecialized in each HMO penetration-level quartile, based on the 2263 residents in the sample. The pattern suggests that the relation between the likelihood of subspecialization and HMO penetration was curvilinear. The pattern was similar when the data for graduates of non-U.S. allopathic medical schools who met the other study criteria were included (n = 4205).

Table 2 shows the adjusted odds ratios for the five models in the testing hierarchy, and Table 3 shows the chi-square tests of difference for the nested model comparisons. We started with a baseline model (model A) containing only Census division variables and then added undergraduate medical school characteristics variables (log of average amount of NIH awards for 1988 to 1991 and private or public ownership; model B); comparison 1 showed that this substantially improved model fit. We then added the individual characteristics (age, sex, ethnicity, and MCAT scores) as a group (model C). Comparison 2 in Table 3 shows that the addition of the individual characteristics further improved the fit of the model. We then added the HMO penetration variable to the model (model D). Comparison 3 shows that the addition of the HMO penetration variable also improved the fit of the model, indicating that managed care penetration in the market where general internal medicine training occurred affected the odds of subspecialization above and beyond individual and medical school characteristics. Finally, to test the nonlinear relation suggested by the results shown in Table 1, model E was created by adding HMO penetration squared to model D. Comparison 4 shows that this addition also improved the fit of the model, and the Hosmer-Lemeshow goodness-of-fit test indicated that model E fit the data adequately (Hosmer-Lemeshow chi-square [8 degrees of freedom] = 15.04; P < 0.06).

We conducted an ancillary analysis to test the effect of indebtedness, another variable that has been associated with specialty choice in previous research. Self-reported information on indebtedness at graduation from undergraduate medical school was available for 1653 of the 2263 residents in our study. Indebtedness, when added to the fully specified model (model E), was not related to the likelihood of subspecialization.

To test the relative explanatory power of individual, medical school, and HMO penetration information, we added each group of variables, in turn, to a baseline model that included only Census division variables. The increase in the chi-square value of the model yielded by each group of variables was used as an index of comparison: that is, individual variables were added to the baseline model and the chi-square value of the difference was calculated, individual variables were removed and the HMO penetration variables were added, and so forth. Individual characteristics were by far the most powerful predictors of subspecialization (chi-square [4 degrees of freedom] = 97.57). Information on HMO penetration, although clearly having a measurable effect, was less than one third as informative as individual characteristics (chi-square [2 degrees of freedom] = 30.02). The two undergraduate medical school variables were the worst predictors of subspecialization (chi-square [2 degrees of freedom] = 12.33).

Table 2 shows the predictor variable odds ratios for all logistic regression models. The odds ratio for MCAT score is the increase in odds of subspecialization for each one-point increment in average MCAT score, and the odds ratio for HMO penetration is the increase in odds of subspecialization for each 0.1 (or 10%) increase in HMO penetration level. Because the NIH support variable was represented in the model as the base-10 log of the dollar amount, the odds ratio is for a tenfold increase in the amount of NIH funds. Increases in the odds of subspecialization were associated with younger age and male sex. Graduates of privately owned undergraduate medical schools and schools with higher NIH award amounts were more likely to pursue subspecialization.

We examined the estimated probabilities generated by model E to determine the nature of the nonlinear relation between the odds of subspecialization and HMO penetration level. Estimated probabilities plotted against HMO penetration rates indicated that the odds of subspecialization increased as HMO penetration increased from 0 to 0.15 (roughly the median value), then decreased as HMO penetration increased above 0.15. This interpretation was supported by a logistic regression model that substituted a design variable indicating HMO penetration values greater than 0.15 for continuous HMO penetration in model D; the odds ratio for HMO penetration greater than 0.15 was 0.88 (compared with 1.0 for HMO values 0.15). Other information from this analysis, shown in Table 4, also supports the conclusion that the odds of subspecialization declined after HMO penetration increased past a threshold of 0.15. After we controlled for individual and undergraduate medical school effects, increased managed care penetration above the median was associated with a decrease in the likelihood of subspecialization.

Discussion

Top Methods Results Discussion Author & Article Info References

Our study provides a rigorous look at how market forces driven by managed care may be affecting internal medicine subspecialization rates. Our findings suggest that graduates of U.S. medical schools completing general internal medicine training in areas with high HMO penetration were significantly less likely to subspecialize, although the effect of managed care penetration was relatively small compared with the influence of individual characteristics.

Previous evidence that medical students adjust their career paths in response to changes in the market for physician services had some conflicting results. Our results extend and confirm findings of a recent survey of internal medicine residents that showed that residents training in markets with higher managed care penetration were less likely to pursue specialist careers [12]. Similarly, recent data from the National Resident Matching Program show that fewer graduating senior medical students are choosing careers in specialties [21] at a time when the employment outlook for generalists is more promising and the incomes of generalist physicians are increasing faster than those of specialists. In contrast, findings in another study have suggested that the market for physicians has little influence on graduate medical education training, but this study was confined to New York State [22].

Although our results suggest that local market forces affect the decision to subspecialize, the influence of the market is small compared with the influence of individual characteristics, shown in previous research to be related to specialty preference [23, 24]. It should be noted that we used only one variable, HMO penetration, to represent market forces. Further research using more sensitive and thorough indices of the complex market environment may demonstrate greater market effects.

Although our study provides some evidence for the overall impact of the market on residents' career decisions, it does not provide insight into the way market penetration of managed care might influence the career decisions of individual residents. For example, one possible mechanism is that market conditions are transmitted to residents in training through the experiences of friends and former colleagues entering the workforce. Although some preliminary observations suggest that such a process is occurring [25], further research is necessary to explore, identify, and clarify the operation of this and other mechanisms.

Our descriptive analyses provided evidence that the relation between HMO penetration and likelihood of subspecialization is the same regardless of whether graduates of non-U.S. allopathic medical schools are included. Thus, the subspecialization percentages for all 1993 residents (U.S. medical graduates and others) in each HMO penetration level revealed the same curvilinear pattern evident in the data for graduates of U.S. allopathic medical schools only (Table 1). Residents who were not graduates of U.S. allopathic medical schools were excluded from the logistic regression analyses because their individual and medical school information was not available. Ancillary logistic regression analyses testing the relation between HMO penetration and odds of subspecialization without control covariates had results similar to those of the complete analysis. Because uncontrolled analyses were too incomplete to be considered conclusive, our study does not confirm the effect of managed care penetration on residents who were not graduates of U.S. allopathic medical schools, a significant proportion of the internal medicine workforce.

General internal medicine trainees follow a wide variety of paths through graduate medical education (Haynes RA, Killian CD. Percentages of 1987 through 1991 U.S. graduates educated as generalists: changes across and within cohorts between PGY4 and PGY7. Unpublished report. Washington, DC: Association of American Medical Colleges; 1996) and practice. For example, many practicing generalists subsequently acquire subspecialty expertise, and many trained subspecialists have practices similar to those of generalist physicians. Our definition of subspecialist and generalist did not capture these varied career paths. Accordingly, our study should not be interpreted as an analysis of the complete internal medicine workforce. Further research is necessary to gauge trends among these excluded groups.

We included raw NIH award amounts to control for research intensiveness in our model. Several potential denominators can be used to adjust NIH awards for school size, but each requires assumptions and carries implications. Because inferences about NIH awards were not the main focus of this study, we left the measure crude rather than choose a possibly inappropriate adjustment. The reader should keep this limitation in mind, however, when interpreting the NIH award results presented here.

The public-private distinction among medical schools is undoubtedly a generalization; both groups contain a wide variety of schools. We have included the public-private categorization as a control covariate in our analyses because we believe that despite its generality, it contains important information and because it has been commonly used in previous research in this area. Our results are not intended to provide a complete picture of the role of undergraduate medical institutions in influencing decisions to pursue subspecialty training.

The fact that market forces seem to influence internal medicine subspecialization decisions of graduates of U.S. allopathic medical schools has implications for workforce planning. The question of whether the career decisions of residents in training are affected by market conditions has been debated. Our findings indicate that many internal medicine residents are probably being influenced by market characteristics. Although further research is necessary to determine the full extent of the market's impact, those involved in shaping the generalist physician workforce should note that market forces seem to be playing some role in promoting the desired balance in the internal medicine workforce. Because market forces may be helping to generate more favorable generalist-to-specialist physician ratios, the full influence of the market should be investigated as part of an overall strategy for shaping the most appropriate U.S. physician workforce.