Lawson, Sonya R.; Hoban, J Dennis; Mazmanian, Paul E.
Although training the correct number of primary care physicians in the United States continues to be a challenge, the residency choice literature is remarkable for inconsistencies, methodological weaknesses,1,2 and a paucity of theory.3 To address these shortcomings, Bland and colleagues3 offered a comprehensive medical career decision model with three components: (1) student characteristics, (2) medical school characteristics, and (3) students’ perceptions of the medical specialty.
In a recent review of the student characteristics component, Lawson and Hoban4 found 17 studies that evaluated relationships between (1) demographics, (2) undergraduate major, (3) perceptions of clinical clerkships, (4) expected income, (5) debt, (6) interest in technology, and (7) interest in diverse patients and health problems, and residency choice using descriptive or univariate statistics. Since these studies did not allow one to determine the combined effect of each individual variable with all other predictor variables, they provide limited predictive value.4 However, they do produce factors that can be examined in multivariate analyses. Since medical residency choice appears to involve several related factors, multivariate statistical techniques are the most appropriate for developing residency choice predictive models.4,5
A review of six multivariate studies5–10 appeared subsequent to the published work of Bland and colleagues.4 The predictive variables found in these studies were gender (female), age (older), marital status (married), ethnicity (underserved minority), expected income (lower), lower MCAT physical science and biology scores, postundergraduate work, parents with lower income, rural hometown, less interest in research, and lower ratio of education debt to expected income.4 Although these studies presented comprehensive, methodologically, and statistically sound research, they did not test a theoretical model or control for demographic variables.
Consolidating family medicine (FM), internal medicine (IM), and pediatrics (PED) into one dependent variable (PC residency choice) limits the value of findings because studies show that physicians within these three fields are not a homogenous group.7,8,11 Subtle differences in significant predictor variables between the three groups are lost when combined into one dependent variable labeled PC. Findings are limited to those factors that are so powerful for one specialty that they remain significant even when weakened by combining the three areas into one dependent variable.12 Analyzing these three career choices separately provides a more complete description of the decision making process of the choices within PC.
Primary care residency choice is marked by a dearth of empirical research that involved a theoretical framework, identified a multivariate predictive model, provided separate analyses for the three fields within PC, and entered sets of variables hierarchically in the model building process. The present study addresses this limitation by testing selected variables of the Bland-Meurer Model. Specific research questions are:
1. “Do variables in the Bland-Meurer Model predict medical student's choice of PC residency?”
2. “Do variables in the Bland-Meurer Model predict medical students’ choice of IM, FM, or PED residency?”
The population of interest was graduates of the Virginia Commonwealth University School of Medicine (VCUSOM) from 1998–2002. The unit of analysis was the individual student. Variables were extracted from the VCUSOM database, and the AAMC Graduate Student Questionnaire (AAMCGSQ). Select variables from these sources were merged into a research database and all identifying information was removed.
Using the student section of the Bland-Meurer Model, the following independent variables were selected and entered into statistical analyses in three stages:
Stage 1: Student characteristics—VCU database
* Demographics (gender/age at graduation/marital status at matriculation/ethnicity)
* Undergraduate major—Science/Nonscience
* Highest degree held—Bachelors/Masters/Doctorate/Law
* Post baccalaureate work—Yes/No
* Physical science Medical College Admission Test (MCAT) scores—0–15.
* Biology MCAT scores—0–15
Stage 2: Student experiences during medical school—AAMCGSQ
* Perceptions of clinical education quality—Excellent(1)/Good(2)/Fair(3)/Poor(4)
* Research volunteer activities in medical school—Yes/No
* Community volunteer activities in medical school—Yes/No
* Influence of debt on career choice—None/minor/moderate/strong
* Debt (educational/noneducational)—$
Stage 3: Student experiences anticipated after medical school—AAMCGSQ
* Extent of research in career—Exclusively/Significantly/Somewhat/Involved in limited way/Not involved
* Practice in underserved area—Yes/No/Undecided
* Perceptions of medical practice environment (AAMCGSQ #29)—Strongly Agree/Agree/No Opinion/Disagree/Strongly Disagree
The dependent variable was residency choice with two groups: PC (FM, IM, and PED) or NPC (all other specialties), identified using National Residency Match Program data. VCU Institutional Review Board approval was obtained. SPSS 11.0 was used to generate both descriptive and inferential statistics.
To determine latent factors associated with AAMCGSQ question 29 (16 unique subitems probing students’ perception of the medical practice environment), a principal component analysis was conducted. Since there were no published studies using these AAMCGSQ data, this exploratory process was necessary to determine whether these questions were measuring similar dimensions of the medical practice environment.
To address research question one, logistic regression analysis was used to develop a model of PC residency choice.13 Independent variables were entered in a logistic regression analysis in three stages based on the pathway depicted in the student characteristics section of the Bland-Meurer Model. First, the dependent variable, PC versus NPC, was regressed separately on each variable in Stage 1. Next, a series of logistic regression procedures was performed using the backward stepwise technique (α = .10). Because the results of logistic regression are reported as probability outcomes, classification cut-point assignment is critical in evaluating the success of the model.14 The assigned cut-point represented the proportion of students that chose a PC residency in this dataset. Variables adequately predicting PC residency choice were combined with variables at Stage 2. Lastly, variables shown to adequately predict PC residency choice at Stage 2 were combined with all variables at stage 3 and a final predictive model was identified for PC residency choice.
To address research question 2, three series of logistic regression procedures were conducted using the same independent variables. The dependent variables associated with these analyses were (1) choice of IM versus choice of all NPC residencies, (2) choice of FM versus choice of all NPC residencies, and (3) choice of PED versus choice of all NPC residencies.
Our sample represents 67% of the students in five classes (555/832). One-third (33%) were omitted because AAMCGSQ information was unavailable (“nonresponders”). Using a medical school student database for nonresponders with all identifying information removed, independent samples t-tests and chi-square tests revealed no differences in student demographics, aptitude, and residency choice between AAMCGSQ responders and nonresponders. Our sample size varies by stage of analysis because only three years of data were available for three AAMCGSQ questions: (1) question 29, (2) influence of debt on career choice, and (3) the extent of research anticipated in their career. As expected, a comparison of logistic regression models, one with 555 participants and one with 373 participants, revealed comparable residency choice outcomes.13 Therefore, we conclude these samples represented this population of medical students.
Fifty-one percent of the 555 medical students chose PC residencies (n = 283). Of the 283 students that chose a PC residency, 28% chose FM (n = 79), 50% chose IM (n = 142), and 22% (n = 62) chose PED. Overall, this sample comprised mostly white (67%) male (59%) students that ranged in age at graduation from 26–30 (66%). The mean MCAT physical science and biology scores were 9.6 (SD = 1.8) and 9.8 (SD = 1.6), respectively and scores ranged from 4–15.
Based on principle component analysis, we determined that the 16 questions taken from the AAMCGSQ question 29 consisted of two factors, which explained 32% of the total variance. Factor 1, “The changing health care system has a negative effect on physicians,” (α = .76) consisted of six items. Two items with the highest factor loadings were “Changes in the health care system impair physician independence” (.757) and “Legal liabilities and high cost of malpractice insurance is a problem” (.678). Factor 2, “Access to medical care and disease prevention is a problem in the U.S.”, (α = .60) consisted of three items, and two items with the highest factor loadings were “Everyone should receive medical care regardless of ability to pay” (.716) and “Access to medical care continues to be a major problem” (.624). Average item raw scores for each factor served as independent variables in subsequent logistic regression analyses.
The logistic regression model for choice of PC residency suggests that being female, giving lower ratings of psychiatry and surgery clerkships and higher ratings of IM clerkship, choosing not to participate in a research project, disagreeing with Factor 1, agreeing with Factor 2, and planning to practice in an underserved area increase a student's likelihood of choosing a PC residency (see Table 1). The Hosmer and Lemeshow goodness of fit test indicates good model fit between the predicted and observed values of PC residency choice and the model chi-square indicates that the independent variables, as a set, reliably distinguish between PC and NPC residency choice.
Logistic regression predictive models generated for choice of IM, FM, and PED residency demonstrate good model fit and the independent variables, as a set, reliably distinguished between IM, FM, or PED and NPC residency choice (see Table 1).
The overall percentage of correctly classified cases for the PC (72.4%), IM (70.6%), FM (83.3%), and PED (77.2%) models was acceptable. A higher percentage of cases were correctly classified at stage three than at stage one for all four models (see Table 2).
Although prior studies of PC residency choice provided mixed predictive results, we found no predictive value of age, marital status, undergraduate background, MCAT scores, and debt level on FM, IM, or PED residency. Since earlier studies did not apply a common theoretical framework, we called upon constructs of the Bland-Meurer framework to guide the selection of variables for the initial stage of model building and for hierarchical design of the regression analysis. The Bland-Meurer Model should continue to be tested empirically to enable further specification of relations among variables that explain and predict PC decisions.
Of the demographic variables in this study, only gender emerged as predictive of residency choice. Female students were approximately four times more likely to choose a PED residency than were male students. This finding appears to corroborate the results of previous studies.5–7,10,15–21 Further research is needed to determine why women are more likely to choose a PED residency. Surprisingly, there was no relationship between gender and FM or IM residencies. The finding that gender was a significant predictor of PC residency choice is most likely accounted for by the strong relationship with PED residency choice. This further demonstrates the need for multivariate studies that analyze these three residency choices separately.
VCU students more likely to choose a PC career are less likely to believe the changing health care system negatively effects them, but more likely to agree that access to medical care and disease preventive services is a problem in the United States. Perhaps medical educators should consider how these values are shaped and influenced by clinical clerkship experiences, as well as interactions with faculty, patients, and peers.
More factor analyses are necessary to assess the validity of the two factors associated with students’ perceptions of medicine and the profession. Since this study included students at only one medical school, this analysis should be replicated with different populations of medical students to verify the two-factor model. Year-to-year comparisons might be useful to changing curriculum, admissions policies, and/or budgetary resources.
Previous multivariate studies reported unsuccessful attempts at correctly predicting the percentage of students who will choose FM, IM, or PED residencies.7,8,22 In this study, the percentages of correctly classified cases for the PC (72.4%), IM (70.6%), FM (83.3%), and PED (77.2%) models were acceptable, and suggest that it is appropriate to generate separate predictive models of these three fields. In addition, the final four residency choice models (Stage 3) showed improvement over the first stage models for ability to classify known cases. This suggests that (1) researchers interested in variables that predict a specific field within PC should not merely examine PC models, but design studies with a specific PC field as a dependent variable; and (2) it is unwise to rely solely on admission variables to determine whether a student might choose a PC field.
Despite the rigor of logistic regression, one should not infer causal relationships. Also, using National Residency Match Program data is a limitation in this study. A PC residency match does not assure a PC career path, since subspecialty options are available in IM and PED. Longitudinal studies are needed to examine predictor models for eventual career paths after initial residency choice. Clearly, medical school officials should recognize that associations exist, yet be cautious in interpreting the relationships.
Finally, making decisions about residency choice is complex. In light of growing concern about whether the United States is producing physicians in sufficient numbers and specialties for serving in diverse geographic regions,23–27 the Bland-Meurer Model should be tested for its usefulness in managing this complexity for predicting other medical specialties. This model may provide health care policymakers and medical educators with a clearer picture of residency choice and a better understanding of how to select and develop physicians for selected patient populations.
1. Davis W, Bouhuijs PJ, Dauphinee WD, et al. Medical career choice: current status of research literature. Teach Learn Med. 1990;2:130–8.
2. Mowbray R. Research in choice of medical specialty: a review of the literature 1977-1987. Aust N Z J Med. 1989;19:389–98.
3. Bland CJ, Meurer LN, Maldonado G. Determinants of primary care specialty choice: a non-statistical meta-analysis of the literature. Acad Med. 1995;70:620–41.
4. Lawson SR, Hoban JD. Predicting career decisions in primary care medicine: a theoretical analysis. J Contin Educ Health Prof. 2003;23:68–80.
5. Gorenflo DW, Ruffin MT, Sheets KJ. A multivariate model for specialty preference by medical students. J Fam Pract. 1994;39:570–7.
6. Martini CM, Veloski J, Barzansky B, Xu G, Fields SK. Medical school and student characteristics that influence choosing a generalist career. JAMA 1994;272:661–5.
7. Kassebaum D, Szenas PL, Schuchert MK. Determinants of the generalist career intentions of 1995 graduating medical students. Acad Med. 1996;71:197–209.
8. Colquitt WL, Zeh MC, Killian CD, Cultice JM. Effect of debt on U.S. medical school graduates’ preferences for family medicine, general internal medicine, and general pediatrics. Acad Med. 1996;71:399–411.
9. Newton DA, Grayson MS, Whitley TW. What predicts medical student career choice. J Gen Intern Med. 1998;13:200–3.
10. Owen JA, Hayden GF, Connors AF. Can medical school admission committee members predict which applicants will choose primary care careers? Acad Med. 2002;77:344–9.
11. Senf JH, Campos-Outcalt D, Watkins AJ, Bastacky S, Killian C. A systematic analysis of how medical school characteristics relate to graduates’ choices of primary care specialties. Acad Med. 1997;72:524–33.
12. Senf JH, Campos-Outcalt D, Kutob RM. Lessons not learned from the generalist initiatives. Acad Med. 2002;77:774–5.
13. Lawson SR. Predictors of primary care residency choice and perceptions of medical practice of Virginia Commonwealth University School of Medicine Graduates. Dissertat Abstr, 2003;64(O2B):1–341.
14. Hosmer DW, Lemeshow S. Applied Logistic Regression. 2nd ed. New York: John Wiley & Sons, 2000.
15. McGrath E, Zimet CN. Female and male medical students: differences in specialty choice selection and personality. J Med Educ. 1997;52:293–300.
16. Babbott D, Baldwin DC, Killian CD, Weaver SO. Racial-ethnic background and specialty choice: a study of U. S. medical school graduates in 1987. Acad Med. 1989;64:595–9.
17. Osborn EHS. Factors influencing students’ choices of primary care or other specialties. Acad Med. 1993;68:572–4.
18. Xu G, Rattner SL, Veloski JJ, Hojat M, Fields SK, Barzansky B. A national study of the factors influencing men and women physicians’ choices of primary care specialties. Acad Med. 1995;70:398–404.
19. Schieberl JL, Covell RM, Berry C, Anderson J. Factors associated with choosing a primary care career. West J Med. 1996;164:492–5.
20. Doucet H, Shah MK, Cummings TL, Kahn MJ. Comparison of internal medicine, pediatric, and medicine/pediatrics applicants and factors influencing career choices. South Med J. 1999;92:296–9.
21. Nieman LA, Gracely EJ. Where nontraditional is the norm: Are sex and age determinants of practicing primary care specialties? J Womens Health Gend Based Med. 1999;8:968–72.
22. Fadem BH, Nicolich MJ, Simring SS, Dauber MH, Bullock LA. Predicting medical specialty choice: a model based on students’ records. J Med Educ. 1984;59:407–15.
23. Cooper RA, Getzen TE, McKee HJ, Laud P. Economic and demographic trends signal an impending physician shortage. Health Aff. 2002;21:140–54.
24. Grumbach K. Primary care in the United States: The best of times, the worst of times. N Engl J Med 1999;341:2008–10.
25. Weiner J. A shortage of physicians or a surplus of assumptions? Health Aff. 2002,21:160–2.
26. Biola H, Green LA, Phillips RL, Guirguis-Blake J, Fryer GE. The U.S. primary care physician workforce: persistently declining interest in primary care medical specialties. Am Fam Physician. 2003;68:1484.