In an era of expected primary care shortages,1–4 declining interest in primary care among medical students,5 and increased focus on the “triple aim” of decreasing health care costs while improving quality and outcomes,6 medical schools have been criticized for not fulfilling their mission to train practitioners who will address society’s workforce needs.7–9 Although many studies have examined the factors that affect student specialty choice, questions remain about the role that schools play beyond selecting students for individual characteristics (e.g., being female, older, having a rural background) that have correlated positively with primary care careers.10–12
Past research provides evidence that some school-level factors—including being a public school, requiring a family practice clerkship, and receiving lower levels of federal funding for biomedical research—have positive associations with medical students’ interest in primary care.11,13–15 Although medical schools are criticized for creating environments in which primary care is disparaged,16 research on the role that this “hidden curriculum” plays in specialty choice is limited and dated.17–20 Furthermore, many of the previous studies on the topic of specialty choice have had weaknesses, such as focusing on a small number of schools,21–23 or not controlling for a student’s preference for primary care prior to admission,5,24 the competitiveness of the student upon graduation,25 or whether primary care is even a fit with the student’s interest. These weaknesses raise questions about how much a student’s specialty choice is the result of the school versus individual preferences or aptitude.
To examine the role that school culture plays in shaping student preferences for primary care, while also addressing the limitations described above, we conducted a large, multischool study of fourth-year medical students that included both individual- and school-level variables. We focused in particular on two aspects of school culture: (1) whether a negative prevailing opinion on primary care among peers, residents, faculty, and administrators at the institution discouraged student interest in primary care; and (2) whether student exposure to positive aspects of a primary care career during internal medicine (IM), family medicine (FM), and pediatrics (PD) clerkships increased interest in primary care. We used multilevel modeling26 to examine school-level variation in an individual’s likelihood of practicing primary care while controlling for known student characteristics associated with a preference for primary care. Results of this study may help schools better understand the extent to which their institution’s culture surrounding primary care may shape student specialty choice, and offer strategies for encouraging primary care careers among students whose interests make them a good fit.
In spring 2010, we administered an online survey to examine factors in specialty choice decisions to all fourth-year medical students (N = 2,604) at a stratified random sample of 20 U.S. MD-granting medical schools. Nonrespondents received up to three e-mail reminders to complete the survey. Students were informed that individual- and school-level results would be kept confidential and that one respondent from each school would be randomly selected to receive an iPad ($500 value). Before fielding the survey, focus groups were conducted with more than 50 fourth-year medical students at six medical schools to help develop the survey instrument. Schools were stratified into primary care production quartiles based on the percentage of the school’s 1995–1999 graduates who were identified as practicing primary care physicians in the 2008 American Medical Association Physician Masterfile. The 39-question survey included questions about specialty interests, residency plans, and medical school experiences (Supplemental Digital Appendix 1, http://links.lww.com/ACADMED/A164). Deidentified United States Medical Licensing Examination [USMLE] Step 1 scores were merged onto the deidentified analysis file using Association of American Medical Colleges (AAMC) research IDs. The institutional review board of the American Institutes of Research approved this study (EX00175).
We based the outcome variable (likely to practice primary care) on two criteria: entry into an FM, IM, PD, or internal medicine/pediatrics (IMPD) residency, and a response of “very likely” to the question: “After completion of your graduate medical education, how likely are you to become a primary care physician?”
The independent variables were factors that prior research has suggested may be associated with specialty choice. These included self-reported individual characteristics such as sex, age (operationalized as age > 30, based on examining the age distribution), race/ethnicity (being African American, Hispanic, or Native American), rural background (having grown up in a rural setting), competitiveness of the student (operationalized as the student’s USMLE Step 1 score being above the median score for the sample), and educational debt (operationalized as total education debt > $250,000, the 90th percentile for our sample). We also incorporated a variable for students’ primary care preference at matriculation, measured by whether the student indicated that when applying to medical school, he or she actively looked for schools that had a reputation for producing large numbers of primary care physicians. Additionally, we built in a measure for whether the students reported that at least one primary care clerkship made him or her more interested in a career in primary care. We also included a measure for the student’s perception of “badmouthing” of primary care among peers, residents, faculty, and deans at his or her school. We defined badmouthing as a prevailing opinion that “primary care is a waste of a mind and/or a fall-back position.” Finally, we included student self-reports of whether public health and community outreach activities during medical school and income expectations were a strong influence on their specialty choice, and students’ reports of how well the primary care specialty fit with their interests.
We also included school characteristics in our models: school ownership type (obtained from the AAMC’s Organizational Characteristics Database) and two school-level primary care culture variables derived from survey responses. The first was a dichotomous variable for whether the percentage of students at a school that reported one or more groups—peers, residents, faculty, and deans—badmouthing primary care was greater than the median percentage for all schools. The second was a dichotomous variable for whether the mean number of positive primary care clerkship experiences reported by students at a school was greater than the median number of positive primary care clerkship experiences across all schools. Positive experiences encompassed exposure to a well-run practice, team-based care, mentors who encouraged careers in primary care, evidence of strong doctor–patient relationships, primary care physicians with good work–life balance, and physicians for whom primary care was a calling.
We used chi-square tests to determine whether students likely to practice primary care differed from other students across a range of individual and school-level characteristics and to compare key student experiences across the four primary care production quartiles of schools. For continuous variables, we compared means using a t test or analysis of variance. All descriptive analyses used SPSS version 17.0 (SPSS Inc., Chicago, Illinois).
Multilevel regression analysis
We used multilevel logistic regression modeling27 to investigate two distinct questions: (1) Is there an association between a given school variable (e.g., private/public, badmouthing) and students’ likelihood of practicing primary care, after controlling for other individual factors in the regression model, such as demographic characteristics and student experiences? and (2) What proportion of the variance in an individual’s likelihood of practicing primary care can we attribute to the fact that students attend different schools? We used MLwiN 2.23 (Centre for Multilevel Modelling, University of Bristol, Bristol, United Kingdom) for the analyses.28
To build the regression models, we started with factors that were important based on theory. We tested bivariate associations with the dependent variable using chi-square tests or correlations, and then tested each independent variable separately in a bivariate multilevel logistic regression model with the dependent variable. Our final regression models included only those variables that were statistically significant at both stages (P < .01), with the exception of educational debt and race/ethnicity, which, from a theoretical standpoint, we deemed necessary to include. We also excluded from the final models independent variables that were highly correlated with each other (e.g., region with school ownership type) or that did not have a significant bivariate relationship with the outcome variable (e.g., research intensity of the school, region of school, attending a regional medical campus, completing an FM/IM/PD clerkship in a community-based setting).
We used a model in which the intercept can vary among schools and therefore the variation in likelihood of practicing primary care can be examined across schools. The null model (Model 1) allowed us to partition the total variance in the likelihood that a student will practice primary care into variation among schools and variation among individuals within schools. In Model 2, we examined the extent to which the included individual-level characteristics contributed to the school-level variation in a student’s likelihood of practicing primary care. Finally, we specified two separate multivariable models that included both individual and school characteristics (Models 3a and 3b). Model 3a incorporates a school-level measure for badmouthing and Model 3b incorporates a school-level measure for positive clerkship experiences to show the effects of each aspect of a school’s primary care culture on student interest in primary care. The two school-level primary care culture variables—badmouthing and positive clerkship experiences—could not be modeled together because of multicollinearity. By including the school-level variables, we were able to examine how much of the residual school-level variance was explained by the school-level variables.
We calculated the intraclass correlation coefficient (ICC)29 using the latent variable method to measure the proportion of total variance in a student’s likelihood of practicing primary care that was due to school-level versus individual-level characteristics. The ICC has a minimum value of 0 (if all students were independent and there was no clustering effect within any medical school) and a maximum value of 1 (if all students within each medical school had exactly the same outcome).
Of the 2,604 students who were sent the survey, 1,661 (64%) responded. The 20 schools in our sample showed no difference from U.S. MD-granting medical schools overall in terms of school type or percentage of graduates entering primary care, and the differences by region were not statistically significant (chi-square = 3.60, P > .30). Survey respondents appeared representative of all fourth-year medical students. As Table 1 shows, sampled students were somewhat more rural than all 2010 U.S. MD graduates, although that may have been because this was self-reported in our sample, whereas for all graduates, rural status was determined by high school county. Of the respondents, 1,554 (94%) students had complete data on all variables included in the multilevel logistic regression models. These students constituted our analytic sample.
Of the analytic sample, 616 students (40%) stated that they would be entering a residency in general IM, FM, PD, and/or IMPD. Only 207 students (13%), however, indicated that they were “very likely” to become a primary care physician after completing residency training. Of these 207, 108 (52%) were entering FM residencies, 49 (24%) PD, 36 (17%) IM, and 14 (7%) IMPD.
Table 2 compares students who are likely to practice primary care with students who are not. The biggest differences between the two groups were in the percentages of students with the following characteristics: had sought a medical school with a primary care reputation (P < .001); at least one of the FM, IM, or PD clerkships made them more interested in a primary care career (P < .001); indicated that public health and community outreach activities during medical school (P < .001) or income expectations (P < .001) were a strong influence; reported that FM or PD was a poor fit with their interests (P < .001); and attended a school that was above median on students’ positive primary care clerkship experiences (P < .001).
Tables 3 and 4 show the variation in primary care culture variables by schools’ historical primary care production quartiles. Overall, respondents were more likely to say that peers (23%, n = 362) and/or residents (23%, n = 351) badmouthed primary care compared with faculty (10%, n = 150) and deans and other administrators (5%, n = 74) (Cochran Q = 419.7, P < .0001). The higher the quartile of primary care production was, the lower the reported prevalence of primary care badmouthing was across all groups (P < .001) (Table 3) and the higher the number of positive primary care clerkship experiences was (P < .001) (Table 4).
Multilevel regression analysis
Table 5 shows the results from our multilevel regression analysis and the relative roles that individual and medical school characteristics play in specialty choice. Student-level factors leading to greater odds of practicing primary care (Model 3a) included the demographic characteristics of being female (1.4; CI 1.0–2.0), which was marginally significant, and having a rural background (1.7; CI 1.1–2.7). Other student characteristics associated with increased likelihood of practicing primary care included having sought a medical school with a primary care reputation (1.9; CI 1.2–2.8); stating at least one primary care clerkship made the student more interested in a career in that field (2.9; CI 1.7–4.8); reporting at least one group at the student’s medical school as badmouthing primary care (2.1; CI 1.5–3.0); and indicating public health or community outreach activities during medical school were a strong influence on specialty choice (2.8; CI 1.8–4.4).
Student attributes leading to reduced odds of practicing primary care included having a USMLE Step 1 score above the survey respondents’ median score (0.6; CI 0.4–0.8); stating that income expectations were a strong influence on specialty choice (0.1; CI 0.0–0.3); and indicating that FM (0.2; CI 0.1–0.3) or PD (0.6; CI 0.4–0.9) was a poor fit with interests.
School-level variables associated with students’ likelihood of practicing primary care included two variables measuring primary care culture. Having an above-median level of badmouthing primary care at a school led to lower odds of practicing primary care (0.6; CI 0.4–0.9), whereas having an above-median level of positive primary care clerkship experiences led to greater odds of practicing primary care (1.6; CI 1.1–2.3).
Factors not associated with the increased likelihood of practicing primary care after other characteristics were accounted for included education debt > $250,000, race/ethnicity, age, stating that IM was a poor fit with interests, and school ownership type (public/private).
Variation among schools
In the null model (Model 1), 8% of the total variation in a student’s likelihood of entering primary care practice was attributable to the context students shared at their respective schools. This means that the vast majority (92%) of the variation in the likelihood of practicing primary care is attributable to factors other than a shared school environment, like individual characteristics. After including the student-level characteristics (Model 2), the residual variation in outcomes across schools was reduced by 86% (0.303 to 0.042). With the addition of the school-level variables, the residual variation in outcomes across schools was virtually eliminated (0.042 in Model 2 to 0.000 in both Models 3a and 3b). This indicates that the factors included in the model account for all of the small amount of variance (8%) observed at the school level, and that a school’s primary care culture plays a small but meaningful role in shaping a student’s decision to pursue a career in primary care.
Numerous studies have examined the role that medical student characteristics play in specialty choice decisions; however, few have examined the role that medical school characteristics play in shaping career choices. Our study shows that although individual students’ characteristics and preferences drive specialty choice decisions, the prevailing primary care culture at a school also plays a role. After controlling for individual characteristics, including a preference for primary care upon entry to medical school and competitiveness of the student, students who attend schools with a higher prevalence of disparaging comments about primary care (badmouthing) are less likely to become primary care physicians. Similarly, students who attend schools where they are more exposed to positive experiences during their primary care clerkships (e.g., team-based care, primary care physicians with good work–life balance) are more likely to graduate medical school intending to pursue a career in primary care.
Paradoxically, while students who attend schools where the prevalence of badmouthing primary care is above the median are less likely to become primary care physicians, students pursuing a career in primary care are more likely to report that primary care is badmouthed at their institution. This suggests that students on a primary care career path may be more likely to be targeted with comments disparaging primary care and may also be more attuned and sensitive to them.
Deans and administrators were cited the least by students for disparaging primary care, but they can still play a role in helping to shape the culture around primary care at their institution, and some are already moving in this direction.30 Efforts to minimize the badmouthing of primary care—or any specialty—could have other benefits besides potentially increasing the number of graduates in primary care. The future U.S. health care system is going to be highly reliant on the primary care workforce; mutual respect will likely engender greater collaboration and coordination among specialties and professions as health care increa singly moves toward integrated care models.
Several of the individual student characteristics found to be significant in our study can be influenced by the school. Schools can recruit more students from rural backgrounds. They can offer more opportunities and incentives for students to participate in public health and community outreach activities, and they can identify preceptors and clerkship opportunities where students are exposed to thriving primary care practices.31 New care delivery models such as patient-centered medical homes and accountable care organizations may open up even more clerkship training sites that reinforce the value of primary care and present primary care as a professionally rewarding career choice.
Our study does have limitations. Although the vast majority of students who graduate with a stated interest in primary care do become primary care physicians,32 we do not know the ultimate practice outcomes of these students. Furthermore, individuals’ abilities to accurately and retrospectively identify the factors that influenced a particular decision may be limited.33 Additionally, we were unable to run separate models for IM, FM, and PD because of small sample sizes. Finally, we only surveyed MD students, and the total number of schools included, though larger than most other studies, is still small. Despite these limitations, we believe our findings identify actionable steps that medical schools can take to increase student interest in primary care careers at a time when the United States is facing a significant shortage of primary care physicians.
Acknowledgments: The authors thank the deans of the 20 medical schools that participated in the survey and the six that hosted focus groups; Henry Sondheimer, MD, for his assistance in facilitating the study; Shevaun Neupert, PhD, for her methodological guidance; Stacey Schulman, MA, and Zoe Berman, MS, for their assistance with the focus groups and research; Kelly Mahon, MA, for her help with the initial literature review; and Casey Tilton for administrative assistance.
3. Petterson SM, Liaw WR, Phillips RL Jr, Rabin DL, Meyers DS, Bazemore AW. Projecting US primary care physician workforce needs: 2010–2025. Ann Fam Med. 2012;10:503–509
4. Kirch DG, Henderson MK, Dill MJ. Physician workforce projections in an era of health care reform. Annu Rev Med. 2012;63:435–445
5. Jeffe DB, Whelan AJ, Andriole DA. Primary care specialty choices of United States medical graduates, 1997–2006. Acad Med. 2010;85:947–958
6. Iglehart JK. Primary care update—light at the end of the tunnel? N Engl J Med. 2012;366:2144–2146
7. Freeman J, Ferrer RL, Greiner KA. Viewpoint: Developing a physician workforce for America’s disadvantaged. Acad Med. 2007;82:133–138
9. Mullan F, Chen C, Petterson S, Kolsky G, Spagnola M. The social mission of medical education: Ranking the schools. Ann Intern Med. 2010;152:804–811
10. Bland CJ, Meurer LN, Maldonado G. Determinants of primary care specialty choice: A non-statistical meta-analysis of the literature. Acad Med. 1995;70:620–641
11. Lawson SR, Hoban JD. Predicting career decisions in primary care medicine: A theoretical analysis. J Contin Educ Health Prof. 2003;23:68–80
12. Senf JH, Campos-Outcalt D, Kutob R. Factors related to the choice of family medicine: A reassessment and literature review. J Am Board Fam Pract. 2003;16:502–512
13. Campos-Outcalt D, Senf J, Watkins AJ, Bastacky S. The effects of medical school curricula, faculty role models, and biomedical research support on choice of generalist physician careers: A review and quality assessment of the literature. Acad Med. 1995;70:611–619
14. Meurer LN. Influence of medical school curriculum on primary care specialty choice: Analysis and synthesis of the literature. Acad Med. 1995;70:388–397
15. Bennett KL, Phillips JP. Finding, recruiting, and sustaining the future primary care physician workforce: A new theoretical model of specialty choice process. Acad Med. 2010;85(10 suppl):S81–S88
17. Block SD, Clark-Chiarelli N, Peters AS, Singer JD. Academia’s chilly climate for primary care. JAMA. 1996;276:677–682
18. Block SD, Clark-Chiarelli N, Singer JD. Mixed messages about primary care in the culture of U.S. medical schools. Acad Med. 1998;73:1087–1094
19. Hearst N, Shore WB, Hudes ES, French L. Family practice bashing as perceived by students at a university medical center. Fam Med. 1995;27:366–370
20. Hunt DD, Scott C, Zhong S, Goldstein E. Frequency and effect of negative comments (“badmouthing”) on medical students’ career choices. Acad Med. 1996;71:665–669
21. Lawson SR, Hoban JD, Mazmanian PE. Understanding primary care residency choices: A test of selected variables in the Bland–Meurer model. Acad Med. 2004;79(10 suppl):S36–S39
22. Newton DA, Grayson MS, Thompson LF. The variable influence of lifestyle and income on medical students’ career specialty choices: Data from two U.S. medical schools, 1998–2004. Acad Med. 2005;80:809–814
23. Phillips J, Weismantel D, Gold K, Schwenk T. How do medical students view the work life of primary care and specialty physicians? Fam Med. 2012;44:7–13
24. Hauer KE, Durning SJ, Kernan WN, et al. Factors associated with medical students’ career choices regarding internal medicine. JAMA. 2008;300:1154–1164
25. Phillips RL, Dodoo MS, Petterson S, et al. Specialty and Geographic Distribution of the Physician Workforce: What Influences Medical Student and Resident Choices?. 2009 Washington, DC Robert Graham Center
26. Raudenbush S, Bryk A Hierarchical Linear Models: Applications and Data Analysis Methods. 2002 Thousand Oaks, Calif Sage Publications
27. Snijders TAB, Bosker RJ Multilevel Analysis: An Introduction to Basic and Advanced Multilevel Modeling. 1999 Thousand Oaks, Calif Sage Publications
28. Rashbash J, Steele F, Browne W, et al. A User’s Guide to MLwiN. 2000 London, England University of London
29. Merlo J, Chaix B, Ohlsson H, et al. A brief conceptual tutorial of multilevel analysis in social epidemiology: Using measures of clustering in multilevel logistic regression to investigate contextual phenomena. J Epidemiol Community Health. 2006;60:290–297
31. Keirns CC, Bosk CL. Perspective: The unintended consequences of training residents in dysfunctional outpatient settings. Acad Med. 2008;83:498–502
32. Jones KC, Erikson CE, Shipman SA. AM Last Page: Medical students' plans at graduation and their relationship with actual practice. Acad Med. 2013;88:1950
© 2013 by the Association of American Medical Colleges
33. Pathman DE, Agnew CR. Querying physicians’ beliefs in career choice studies: The limitations of introspective causal reports. Fam Med. 1993;25:203–207