ACADEMIA AND CLINIC Career Differences between Primary Care and Traditional Trainees in Internal Medicine and Pediatrics John Noble, MD; Robert H. Friedman, MD; Barbara Starfield, MD; Arlene Ash, PhD; and Charlyn Black, MD, ScD Objective: To assess the relation of Primary Care Residency Training to career choice, board certification, and practice location of internists and pediatricians. Design: Cohort study with up to 8 years of follow-up. Setting: The United States. Participants: The 17 933 residents trained in all internal medicine (13 750) and pediatrics (4183) residency programs between 1977 and 1982 were studied using information from the National Resident Matching Program, the AMA Physician Masterfile, the Area Resource File, and a telephone survey. Measurements: Career choice, board certification, and practice location were studied in relation to five explanatory variables: type of residency (primary care or traditional track), gender, year of medical school graduation, educational orientation of the teaching hospital, and medical school prestige. Main Results: Graduates of primary care residency training programs chose careers in generalist primary care significantly more often than did graduates of traditional tracks in both internal medicine (72% compared with 54%) and pediatrics (88% and 81%, respectively; P < 0.001 for both values). Board certification rates in internal medicine were statistically higher for graduates of primary care training programs (80%) than for graduates of traditional programs (76%, P = 0.002) but were not statistically significant for both groups of pediatric graduates. Graduates of primary care programs in pediatrics and internal medicine practiced in medically less served communities more often than did graduates of traditional programs. Conclusion: Graduates of primary care residency training programs in internal medicine and pediatrics differ from graduates of traditional residency programs in career choices, board certification rates, and practice locations. Annals of Internal Medicine. 1992;116:482-487. From Boston University School of Medicine and Boston City Hospital, Boston, Massachusetts; and the Johns Hopkins School of Hygiene and Public Health, Baltimore, Maryland. For current author addresses, see end of text. A national study of all residency training programs in internal medicine and pediatrics was conducted to assess differences in career choices between primary care trainees and other graduates within each discipline. We compared the graduates of federally funded primary care residencies, nonfederally funded primary care programs, and traditional residency programs. Outcomes of training included the choice of a generalist-primary care career, board certification, and the location of practice in medically underserved areas. The purpose of this study was to assess the outcomes of primary care training and the relevance of these programs to reforms in the traditional residency curricula of internal medicine and pediatrics to better meet the nation's need for primary care physicians. Methods Participants We studied physicians who participated in the National Residency Matching Program (NRMP) during the years 1977-1982 and who entered practice in internal medicine or pediatrics by 1985. The Trainee File, obtained from the NRMP, contained records on 17 933 such physicians; 13 750 trained in internal medicine and 4183 trained in pediatrics. This file was merged with data abstracted from the American Medical Association's (AMA) Physician Masterfile on practicing internists and pediatricians. The AMA file contained data on physicians who received any residency training in either of the two specialties between 1975 and 1985 and who had provided information in 1985 regarding their practice specialties, board certification status, and practice location. For each person in the file, available information included: gender, medical school and year of graduation, type of residency program (internal medicine or pediatrics), sponsoring institution (teaching hospital), and 1985 information regarding the physician's practice specialty, board certification status, and practice location. Most trainees were U.S. medical graduates; only 2% had attended medical schools outside of the United States. This preliminary Trainee File was prepared for analysis by removing records on persons who left internal medicine or pediatrics after 1 year to study other medical specialties (usually after completing a 1-year preliminary residency program). Records with missing values for any of the explanatory or outcome variables used in the multivariate analyses were deleted. Sensitivity analyses showed that these choices did not affect study conclusions (1). Outcome Variables Three outcome variables were studied: career choice, board certification status, and practice location. Career choice was classified as either generalist or subspecialist based on the self-designation of practice specialty by physicians responding to the 1985 AMA Physician Questionnaire. In pediatrics, only two subspecialties (cardiology and allergy) were included in the 1985 Masterfile. Some pediatric subspecialists were therefore classified as generalists in this study. Those who passed the general certifying examination in internal medicine or pediatrics were considered board certified. Practice location was a 482 1992 American College of Physicians
Table 1. Frequency of Generalist Career Choice by Graduates of Internal Medicine and Pediatrics Residency Training Programs Type of Residency Program Internal Medicine Pediatrics Number Generalists, % Number Generalists, % Traditional All primary care Federally funded primary care Nonfederally funded primary care 12 594 1156 786 370 54 72* 72 72t 3411 772 458 314 81 88* 88 89f * Differences between traditional and primary care programs are significant at P < 0.001 for both internal medicine and pediatrics. t Differences between federally and nonfederally funded programs are not significant in either specialty. continuous variable defined as the density of physicians in the county in which the physician practiced (number per 100 000 population derived from the Area Resource File of the United States Department of Health and Human Services [USD- HHS]). To introduce physician density within the county of medical practice to the Trainee File, zip codes in the Area Resource File were matched with those of physician addresses in the Trainee File. Unfortunately, the federal Health Manpower Shortage Area (HMSA) designation could not be used because it does not conform to census, zip code, or other national data sets of either population or geography. Explanatory Variables Five explanatory variables were studied: the type of residency program (traditional or primary care, and if primary care, federally or nonfederally funded), gender, the year of medical school graduation, the educational orientation of the principal teaching hospital, and the prestige of the medical school affiliated with that hospital. The purpose of the study was to investigate the association between the type of residency training and the three outcome variables. The other four explanatory variables were included because they have been shown to affect one or more of the study outcomes (2). The classification of a residency program as traditional or primary care was determined by written responses to questionnaires mailed to all residency program directors in internal medicine and pediatrics in 1986 and was confirmed by telephone interviews. For training programs with only a single track (either traditional or primary care) and programs with both tracks but separate NRMP match numbers for each track, the training program type for each physician in the Trainee File was determined using the physician's NRMP match number. For dualtrack programs that used only one NRMP match number for both primary care and traditional trainees, residents were identified in each track using the names of all primary care residents obtained from the program directors. To capture the educational orientation of the hospital with which each resident's training program was affiliated, we used the Kelly Typology (3). Hospitals were categorized as follows: 1 = nonteaching hospital; 2 = regional hospital; 3 = major academic affiliated hospital; and 4 = one of the 100 largest teaching centers. The prestige of the medical school associated with each training program was obtained from Cole and Upton's national study of the opinions of full-time medical school faculty (4). This Prestige Rank Scale was coded on an ordinal scale of 1 to 4 in which 4 indicated a teaching hospital affiliated primarily with a medical school in the highest prestige quartile. Prestige rank scores were attached to the Trainee File by linking each hospital with the score of the medical school most closely affiliated with it, as determined from the 1985 NRMP Directory. Statistical Analysis Analyses were done separately for internal medicine and pediatrics. Within each discipline, analyses were done for all trainees, comparing graduates of primary care programs with graduates of traditional programs. Separate analyses examined only graduates of primary care programs, and compared those in federally-funded programs with those in programs without such funding. Associations between type of training (primary care compared with traditional and federally funded compared with nonfederally funded) and choice of practice specialty (generalist or subspecialist), and the relations between training type and board certification (certified or not) were tested using chisquare analysis. The relation between the type of training and the number of physicians per capita in the physician's county of practice were tested using the Student Mest. The joint effects of the five explanatory variables on the three outcomes of interest were explored using linear and logistic regression analyses with stepwise fitting. To detect changes within the study period, we used another explanatory variable: "early" or "late" graduation from medical school. Individuals who graduated between 1977 and 1979 were classified as "early" graduates, and those who graduated during or after 1980 were considered "late" graduates. In analyses restricted only to primary care training program trainees, trainee type distinguished those in federally-funded primary care programs from those in nonfederally funded programs. All reported P values are two-tailed. For all analyses, P values less than 0.05 were considered statistically significant. Variables associated with P values less than 0.15 were retained in stepwise regression analyses. Results Choice of a Primary Care Generalist Career Between 1977 and 1983, graduates of primary care residency training programs in both internal medicine and pediatrics chose generalist primary care careers more often than did graduates of traditional programs (Table 1). In internal medicine, primary care careers were chosen by 54% of the graduates of traditional residencies compared with 72% of graduates from primary care programs. In pediatrics, generalist careers were chosen by 88% of the primary care graduates, compared with 81% of those in traditional programs (P < 0.001 for all values). In contrast, no difference in career choice was found between the graduates of federally funded and nonfederally funded primary care training programs in either specialty. Multivariate analyses were done using five explanatory variables for career choices in internal medicine and pediatrics (Table 2). In each specialty, all five variables were significantly associated with the choice of a generalist, as compared with a subspecialist, career. Primary care generalist career choices were more common among those who had received primary care residency training, who had trained in less academic-oriented hospitals, and who had graduated from programs affiliated with less prestigious medical schools. Internists and pediatricians with generalist careers were 15 March 1992 Annals of Internal Medicine Volume 116 Number 6 483
Table 2. Factors Associated with a "Generalist" Career for Graduates of Internal Medicine and Pediatric Residency Training Programs* Variable Internal Medicine Pediatrics Ordert of Odds Ratio Ordert of Odds Ratio Entry into Model (95% CI)* Entry into Model (95% CI)* All residents Primary care training program 3 2.10 (1.83 to 2.40) 4 1.49 (1.10 to 1.89) Hospital type 2 0.77 (0.74 to 0.80) 5 0.85 (0.74 to 0.98) Medical school prestigell 5 0.40 (0.87 to 0.94) 3 0.86 (0.79 to 0.93) Female sex 4 1.52 (1.38 to 1.68) 2 1.72 (1.44 to 2.05) Recent medical school graduation** 1 2.20 (2.04 to 2.38) 1 2.48 (2.08 to 2.96) Primary care residents only Federally funded training programs 1.16 (0.82 to 1.65) 0.94 (0.60 to 1.48) Hospital type 3 0.78 (0.64 to 0.95) 0.99 (0.77 to 1.28) Medical school prestigell 0.96 (0.84 to 1.10) 1.03 (0.85 to 1.25) Female sex 2 1.92 (1.35 to 2.80) 2 1.77(1.11 to 2.83) Recent medical school graduation** 1 1.99 (1.52 to 2.62) 1 2.34 (1.46 to 3.74) * Sample sizes for analyses of all residents: internal medicine = 13 750; pediatrics = 4183. Sample sizes for analyses of primary residents only: internal medicine = 1156; pediatrics = 772. Determined by stepwise logistic regression. f Order refers to the order of entry into a stepwise logistic regression model whose variables met a P = 0.15 significance level. $ Odds ratios and 95% CIs were taken from regressions in which all listed variables were entered in a nonstepwise procedure. P < 0.05. Hospital type (1 = nonteaching to 4 = major teaching hospital). 11 Medical school prestige (1 = lowest quartile to 4 = highest quartile). ** Recent Medical School Graduation (0 = before 1980; 1 = after 1980). more likely to be women and more recent medical school graduates. Recent graduation from medical school was the first variable to enter the stepwise regression model in both internal medicine and pediatrics. Multivariate analyses were also done for the subset of trainees in either federally-funded or nonfederally funded primary care programs (see Table 2). The same five explanatory variables were used to compare career choices among primary care trainees; however, in these analyses the training program marker distinguished those in federally funded programs from other primary care resident trainees. In both internal medicine and pediatrics, women and more recent medical school graduates were more likely to be generalists. Among internists, but not pediatricians, those training in less teaching oriented hospitals were more likely to choose generalist careers. Board Certification We noted a small but statistically significant difference in board certification rates in internal medicine Table 3. Graduates of Internal Medicine and Pediatric Residency Training Programs Who Are Board Certified* Type of Training Percent of Graduates Board Program Certified by 1985 Internal Medicine Pediatrics Traditionalt 76 80 All primary care 80 76 Federally funded primary caret 85 75 Nonfederally funded primary care 71 78 * Sample sizes are listed in Table 1. t Difference between traditional and primary care programs is significant for internal medicine (P = 0.002) but not for pediatrics (P = 0.089). $ Difference between federally funded and nonfederally funded programs is significant for internal medicine (P < 0.001) but not for pediatrics (P > 0.2). between the graduates of primary care training programs (80%) and the graduates of traditional residencies (76%) (Table 3). A larger difference in rates of board certification was seen between internal medicine graduates of federally funded primary care programs (85%) and those of nonfederally funded primary care programs (71%). Differences in board certification rates between graduates of primary care programs and traditional programs in pediatrics and between graduates of federally and nonfederally funded pediatric programs were not significant. A multivariate logistic analysis was done for board certification with the same five explanatory variables used to study career choice (Table 4). Male graduates from medical schools in the earlier period and trainees at centers with more teaching and affiliation with more prestigious medical schools were more likely to be board certified. In internal medicine, primary care graduates were more likely to achieve certification than were traditional program graduates. When the analysis was confined to graduates of primary care programs in each of the specialties, the teaching orientation of the hospitals and the period of medical school graduation remained important explanatory variables, but gender did not. No difference in board certification was seen when we compared men and women in primary care training programs. Among primary care graduates, training in a federally funded program and affiliation with a more prestigious medical school were significantly associated with board certification for internal medicine but not for pediatrics. Practice Location Graduates of primary care training programs tended to practice in areas with fewer physicians per capita than did graduates of traditional training programs (Table 5). This difference was significant in pediatrics 484 15 March 1992 Annals of Internal Medicine Volume 116 Number 6
Table 4. Factors Associated with Achievement of Board Certification (by 1985) by Graduates of Internal Medicine and Pediatric Residency Training Programs* Variable Internal Medicine Pediatrics Order of Odds Ratio Order of Odds Ratio Entry into Model (95% CI) Entry into Model (95% CI) All residents Primary care training program 5 1.35t (1.15 to 1.58) 1.02(0.65 to 1.21) Hospital type 1 1.55t (1.49 to 1.61) 2 1.40f(1.25 to 1.58) Medical school prestige 2 1.22t (1.17 to 1.27) 3 1.211" (1.11 to 1.31) Female sex 4 0.72 (0.65 to 0.79) 4 0.80t (0.70 to 0.92) Recent medical school graduation 3 0.69t (0.64 to 0.75) 1 0.82t (0.71 to 0.93) Primary care residents only Federally funded training program 1 1.67t (1.15 to 2.42) 0.99(0.71 to 1.39) Hospital type 3 1.32 (1.09 to 1.61) 2 1.46t (1.20 to 1.77) Medical school prestige 4 1.19t (1.03 to 1.36) 1.07 (0.94 to 1.22) Female sex 0.87 (0.60 to 1.26) 0.96(0.68 to 1.34) Recent medical school graduation 2 0.67t (0.50 to 0.90) 1 0.81 (0.09 to 0.19) * Sample sizes are given in Table 1. Coding of variables (determined by stepwise logistic regression) is described in Table 2. Order of entry into the model is described in Table 2. t P < 0.05. (P = 0.007) but not in internal medicine (P > 0.2). For each specialty, graduates of nonfederally funded primary care programs tended to practice in counties with lower proportions of physicians per capita (P = 0.013 for all values). For internal medicine, all five explanatory variables were associated with the number of physicians per capita in the communities where the graduates practiced (Table 6). In the regression analysis, the independent variable was the physician-to-population ratio in the 1985 practice location of the former trainee. A lower density of physicians in the practice area was associated with a primary care residency, with residencies attached to less prestigious medical schools, and with programs located in less teaching-oriented hospitals. A lower density of physicians in the practice area was also associated with male sex and with more recent graduation. Similar findings pertained to pediatrics, except that trainee type (primary care compared with traditional) and year of medical school graduation did not enter the model. In analyses limited to graduates of primary care training programs, less medical school prestige, less teaching orientation of the hospital, and male sex were associated with internists and pediatricians practicing in areas with fewer physicians (Table 6). Discussion In this national study about three quarters of physicians who entered primary care internal medicine training programs chose generalist careers. This finding confirms previous reports of individual programs (5-7) and contrasts sharply with both the 37% of internists who chose generalist careers in the immediate preprimary care training era (8) and with the 54% of residents in traditional residency programs who chose these careers identified in our study between 1980 and 1985. The increase in generalist career choices by traditional program trainees observed between 1972 and 1985 may reflect the secular trend toward primary care prevalent in the internal medicine during the late 1970s and 1980s. Primary care was the central mission of pediatric residency training until 1987 when pediatric subspecialty training began to receive more emphasis (9, 10). Pediatric departments continue to support primary care training, as reflected in the 81% of graduates from traditional pediatric residencies who described themselves in our study as general pediatricians. However, among pediatricians who trained in primary care tracks, a higher percent (88%) chose primary care careers. In an earlier study, Shelov and colleagues (11) reported that 97% of the graduates of federally funded pediatric primary care training programs were practicing general pediatrics. As noted, the limited designation of pediatric subspecialties in the 1985 AMA Masterfile undoubtedly increased the number of pediatricians reporting generalist careers in our study. Although the association of primary care training and a subsequent primary care career choice has been noted in preliminary surveys of the career choices of primary care and traditional residents in both internal medicine (12) and pediatrics (13), our survey indicates that more recent graduates of primary care residencies have chosen primary care careers more often than have earlier graduates. Table 5. Active Physicians per 100 000 Population Practicing in the County of Medical Practice of Graduates of Internal Medicine and Pediatrics Residency Training Programs* Type of Training Program Physicians per 100 000 Population Internal Medicine Pediatrics Mean ± SD Traditionalt All primary care 285 ± 200 278 ± 185 273 ± 185 253 ± 166 Federally funded primary care* 299 ± 190 266 ± 177 Nonfederally funded primary care 233 ± 166 236 ± 147 * Sample sizes are listed in Table 1. t Difference between traditional and primary care program graduates is significant for pediatrics (P = 0.007) but not for internal medicine {P > 0.2). $ Difference between federally funded and nonfederally funded primary care programs is significant for both internal medicine (P - 0.0001) and for pediatrics (P = 0.013). 15 March 1992 Annals of Internal Medicine Volume 116 Number 6 485
The extent to which the differences in career choice between primary care and traditional residents can be attributed to primary care residency training is unclear. This is because of recruitment bias; that is, trainees entering primary care residencies may have had predispositions towards different careers than did traditional trainees at the start of their residencies. Primary care programs have provided three of the four factors identified by Ernst and Yett (14) to explain the career choices of physicians: provision of a defined curricula, preferential recruitment of interested individuals, and exposure to faculty role models. Despite lower incomes of generalists (the fourth factor), most primary care program graduates went on to choose primary care careers. The increased likelihood of recent graduates to choose generalist careers may reflect the maturation of primary care training programs, or it may be a reporting artifact associated with an earlier stage in a physician's career. However, no evidence of attrition or switching to subspecialty careers of primary care graduates was apparent, in our data, in more recent surveys (5-7) or during reviews of graduate placements at the 12 site visits to primary care programs conducted as part of our study (1). The vital importance of faculty who served as role models was emphasized both during the site visits and in other studies (15-17). Practice styles of primary care graduates were found during the site visits to differ from those of their traditionally trained peers. Similar findings were reported by Weil and Schleiter (18), who noted that primary care program graduates provided more principal care than did subspecialty graduates in the 1976-1977 cohort of residents in internal medicine. More recent graduates of primary care training programs have reported that they provide a greater breadth of service and more psychosocial support and that they use community agencies more extensively than do their peers who trained in traditional tracks sponsored by the same residency program (7, 11). Although recruitment bias limits our ability to independently identify the effect of the primary care training programs on outcomes, the data from this national survey, from our site visits, and from the literature show that graduates of primary care training programs in each discipline make substantially different career choices than do graduates of traditional residency programs. Differences in career outcomes were identified among primary care trainees in federally funded and in nonfederally funded residency training programs. Most federally funded primary care training programs in internal medicine are in larger, more academic hospitals, and this may account for the higher rate of board certification among this subset of trainees. Overall, the reports of board certification in our study indicate that participation in a primary care program did not diminish the trainee's ability to master the traditional elements of either the internal medicine or pediatrics curriculum. These results answer questions posed by medical educators ten years ago about the quality of primary care training (19, 20). The results of the practice location analyses are difficult to interpret because the outcome measure "density of physicians" is a weak indicator of a health manpower shortage area. In our study, primary care graduates tended to establish their practices in areas with fewer physicians per capita than did graduates of traditional programs. This trend appeared in every analysis and is consistent with earlier findings that graduates of traditional training programs in internal medicine continue to enter practice in more urban and suburban areas (14, 21, 22). In all analyses among primary care trainees alone (in both disciplines), the graduates of nonfederally funded primary care programs practiced in areas with fewer physicians than did graduates of federally funded programs. This finding may have occurred because a large number of the nonfederally funded primary care residency programs were located in smaller regional hospitals that are situated in less populated areas. Ernst and Yett (14) have reported that up to 63% of physicians Table 6. Factors Associated with the Density of Physicians in Practice Location of Graduates of Internal Medicine and Pediatric Residency Training Programs* Variable Internal Medicine Pediatrics Order of Coefficient! SE Order of Coefficientf SE Entry into Model Coefficient Entry into Model Coefficient All residents Primary care 5 0.14 0.6-0.7 0.7 Hospital type 2 2.0 0.2 2 2.2 0.4 Medical school prestige 1 3.0 0.2 1 3.1 0.3 Female sex 3 3.0 0.4 3 2.9 0.6 Recent medical school graduation 4 1.0 0.3-0.6 0.6 Primary care residents Primary care training program 4 2.0 1.4 4 1.8 1.2 Hospital type 2 2.0 0.8 2 2.1 0.7 Medical school prestige 1 3.0 0.5 1 1.9 0.5 Female sex 3 5.0 1.3 3 3.7 1.2 Recent medical school graduation - 1.0 1.1-2.0 1.8 * Sample sizes are listed in Table 1. Coding of variables (determined by stepwise logistic regression) is described in Table 2. Coding of order of entry is described in Table 2. t Coefficients are expressed in units of 1 physician/10 000 persons (for example, a coefficient of 3.0 for female sex means that female graduates practiced on average in locations with three more physicians per 10 000 persons than did male graduates. 486 15 March 1992 Annals of Internal Medicine Volume 116 Number 6
practice in the state in which their residency was located. Lewis (23) stated in 1986 that it was "time to retool the factory" in his assessment of the recent NRMP results and the decline of applicants to internal medicine and other primary care residencies. Similar calls for action and change have been expressed for both pediatric and internal medical residency training (21, 25, 26). Primary care residency training programs have provided an important expansion of traditional residency training in both disciplines. They have "retooled" the curriculum and the orientation of faculty and academic leaders at teaching hospitals throughout the nation. The challenge for the next decade is to secure a more stable basis on which to finance the costs of primary care training (27) and to incorporate this training more widely in internal medicine and pediatrics. Acknowledgments: The authors thank Anne Crowley, PhD, and Jean Robak, MS, at the American Medical Association and John Graettinger, MD, at the National Resident Matching Program for the data and assistance. Grant Support: In part by Contract No. HRSA 240-85-0048, Division of Medicine, Bureau of Health Professions, Health Resources and Services Administration, Department of Health and Human Services. Fellowship support was provided to Dr. Black by the National Health Research and Development Program of Canada. Requests for Reprints: John Noble, MD, Section of General Internal Medicine, Boston City Hospital, 818 Harrison Avenue, Boston, MA 02118. Current Author Addresses: Dr. Noble: Section of General Internal Medicine, Boston City Hospital, 818 Harrison Avenue, Boston, MA 02118. Drs. Friedman and Ash: Doctor's Office Building, Evans Medical Group, 720 Harrison Avenue, Boston, MA 02118. Dr. Starfield: Department of Health Policy and Management, Johns Hopkins University, Hampton House, 624 North Broadway, Baltimore, MD 21205-1901. Dr. Black: Community Health Sciences, S101-750 Bannalyne Avenue, Winnipeg, Manitoba R3E 0W3, Canada. References 1. Noble J, Friedman RH, Starfield B. Assessment of the development and support of primary care residency training (general internal medicine and pediatrics). Final report. 1988 Contract No. HRSA 240-85-0048. 2. Rezler AG. Career Choices. In: Rezler AG, Flaherty JA; eds. The Interpersonal Dimension in Medical Education. New York: Springer Publishing Company; 1985. 3. Kelly NL. A four-cell typology to measure hospital teaching status. Inquiry. 1984;21:276-86. 4. Cole JR, Lipton JA. The reputations of american medical schools. Social Forces. 1977;55:662-84. 5. Calkins DR, Noble J. Residency training in general internal medicine: a survey of federally funded programs {Abstract!. Res - c,in 1986;34:812A. 6. McPhee SJ, Mitchell TF, Schroeder SA, Perez-Stable EJ, Bindman AB. Training in a primary care internal medicine residency program. The first ten years. JAMA. 1987;258:1491-5. 7. Witzburg RA, Noble J. Career development among residents completing primary care and traditional residencies in medicine at the Boston City Hospital, 1974-1983. J Gen Intern Med. 1988;3:48-53. 8. Wechsler H, Dorsey J, Bovey JD. A follow-up study of residents in internal medicine, pediatrics, obstetrics and gynecology training programs in Massachusetts. N Engl J Med. 1978;298:15-21. 9. Bergman AB. Pediatric education for what? Pediatrics. 1975;55: 109-13. 10. The future of pediatrics. Implications of the changing environment of medicine. Council on Long Range Planning and Development in cooperation with the American Academy of Pediatrics. JAMA. 1987;258:240-5. 11. Shelov SP, Alpert JJ, Rayman I, Straus JH, Fallon S, Bouflford J. Federally supported primary care training programs and pediatric careers. Am J Dis Child. 1987;141:65-6. 12. Goldenberg DL, Pozen JT, Cohen AS. The effect of a primary-care pathway on internal medicine residents' career plans. Ann Intern Med. 1979;91:271-4. 13. Crain LS, Dienst ER, Malloy MJ. The effects of primary care versus traditional training on career choice in pediatrics. West J Med. 1981;135:248-51. 14. Ernst RL, Yett DE. Physician Location and Specialty Choices. Ann Arbor, Michigan: Health Administration Press; 1985. 15. Adler R, Korsch BM, Negrete VF. Timing and motivation in pediatric career choices. J Med Educ. 1985;60:174-80. 16. Paiva RE, Vu NV, Verhulst SJ. The effect of clinical experience in medical school on specialty choice decisions. J Med Educ. 1982;57: 666-74. 17. Quill TE. Medical resident education. A cross-sectional study of the influence of the ambulatory preceptor as a role model. Arch Intern Med. 1987;147:971-3. 18. Weil PA, Schleiter MK. National Study of Internal Medicine Manpower: VI. Factors predicting preferences of residents for careers in primary care or subspecialty practice of academic medicine. Ann Intern Med. 1981;94:691-703. 19. Thier SO, Berliner RW. Sounding boards: two views of "the doctors' dilemma". Manpower policy: base it on facts, not opinions. N Engl J Med. 1978;299:1305-7. 20. Reitemeier RJ, Benson JA Jr. Sounding boards: two views of "the doctors' dilemna". As the board sees it. N Engl J Med. 1978;299: 1308-9. 21. Schleiter MK, Tarlov AR. A National Study of Internal Medicine. Phase III: analysis of 1976-1977 resident cohort currently in practice. Final report. Division of Medicine Health Resources and Service Administration, Department of Health and Human Services Contract No. HRSA 232-90-0032, 1983. 22. Weil PA, Tarlov AR, Scheiter MR. The National Study of Internal Medicine Manpower: Final Report. Division of Medicine, Health Resources Administration, Department of Health, Education and Welfare Contract # HRA 231-76-0059, 1979. 23. Lewis C. Training in internal medicine: time to retool the factory? Ann Intern Med. 1986;104:570-2. 24. Schroeder SA, Showstack JA, Gerbert B. Residency training in internal medicine: time for a change? Ann Intern Med. 1986;104:554-61. 25. Cleveland WW, Brownell RC. Future training of pediatricians: summary report of a series of conferences sponsored by the American Board of Pediatrics. Pediatrics. 1987;80:451-7. 26. Graettinger JS. Internal medicine in the National Resident Matching Program 1987: the ides of March Ann Intern Med. 1988;108:101-15. 27. Prout DM. Challenges facing general internal medicine in the 99th Congress. J Gen Intern Med. 1986;1:44-8. 15 March 1992 Annals of Internal Medicine Volume 116 Number 6 487