aAll hospitalizations to acute care hospitals in 2012-2014 among continuously enrolled traditional Medicare beneficiaries 65 years and older. Major teaching hospitals were defined as those with membership in the Council of Teaching Hospitals (COTH). Minor teaching hospitals had medical school affiliation but no COTH membership. Nonteaching hospitals had neither COTH membership nor medical school affiliation.
eTable 1. Analytical Sample With Exclusions
eTable 2. Comparison of Seven-Day Mortality for All Hospitalizations and Composite Mortality for Selected Medical and Surgical Conditions, 2012-2014
eTable 3. Seven-Day Mortality for Selected Medical and Surgical Conditions Adjusted for Patient and Hospital Characteristics, 2012-2014
eTable 4. Comparison of Ninety-Day Mortality for all Hospitalizations and Composite Mortality for Selected Medical and Surgical Conditions, 2012-2014
eTable 5. Ninety-Day Mortality for Medical and Surgical Conditions Adjusted for Patient and Hospital Characteristics, 2012-2014
eTable 6. Comparison of Thirty and Ninety-Day Mortality for All Hospitalizations and Composite Mortality for Selected Medical and Surgical Conditions, Excluding Transfers, 2012-2014
eTable 7. Comparison of Mortality for All Hospitalizations and Composite Mortality for Selected Medical and Surgical Conditions, Using Elixhauser Conditions to Adjust for Patient Comorbidities, 2012-2014
eTable 8. Coefficient for Intern/Resident to Bed Ratio as a Measure of Teaching Intensity, 2012-2014
eMethods. Description of the Linear Probability Model Applied to Binary Outcomes
Customize your JAMA Network experience by selecting one or more topics from the list below.
Burke LG, Frakt AB, Khullar D, Orav EJ, Jha AK. Association Between Teaching Status and Mortality in US Hospitals. JAMA. 2017;317(20):2105–2113. doi:10.1001/jama.2017.5702
Is there a difference in mortality rates at US teaching hospitals compared with other hospitals?
In an observational study of approximately 21 million hospitalizations of Medicare beneficiaries, adjusted 30-day mortality rates were significantly lower at 250 major teaching hospitals compared with 894 minor teaching and 3339 nonteaching hospitals overall (8.3% vs 9.2% and 9.5%) as well as for several individual common medical and surgical conditions.
Major teaching hospital status was associated with lower mortality rates for common conditions.
Few studies have analyzed contemporary data on outcomes at US teaching hospitals vs nonteaching hospitals.
To examine risk-adjusted outcomes for patients admitted to teaching vs nonteaching hospitals across a broad range of medical and surgical conditions.
Design, Setting, and Participants
Use of national Medicare data to compare mortality rates in US teaching and nonteaching hospitals for all hospitalizations and for common medical and surgical conditions among Medicare beneficiaries 65 years and older.
Hospital teaching status: major teaching hospitals (members of the Council of Teaching Hospitals), minor teaching hospitals (other hospitals with medical school affiliation), and nonteaching hospitals (remaining hospitals).
Main Outcomes and Measures
Primary outcome was 30-day mortality rate for all hospitalizations and for 15 common medical and 6 surgical conditions. Secondary outcomes included 30-day mortality stratified by hospital size and 7-day mortality and 90-day mortality for all hospitalizations as well as for individual medical and surgical conditions.
The sample consisted of 21 451 824 total hospitalizations at 4483 hospitals, of which 250 (5.6%) were major teaching, 894 (19.9%) were minor teaching, and 3339 (74.3%) were nonteaching hospitals. Unadjusted 30-day mortality was 8.1% at major teaching hospitals, 9.2% at minor teaching hospitals, and 9.6% at nonteaching hospitals, with a 1.5% (95% CI, 1.3%-1.7%; P < .001) mortality difference between major teaching hospitals and nonteaching hospitals. After adjusting for patient and hospital characteristics, the same pattern persisted (8.3% mortality at major teaching vs 9.2% at minor teaching and 9.5% at nonteaching), but the difference in mortality between major and nonteaching hospitals was smaller (1.2% [95% CI, 1.0%-1.4%]; P < .001). After stratifying by hospital size, 187 large (≥400 beds) major teaching hospitals had lower adjusted overall 30-day mortality relative to 76 large nonteaching hospitals (8.1% vs 9.4%; 1.2% difference [95% CI, 0.9%-1.5%]; P < .001). This same pattern of lower overall 30-day mortality at teaching hospitals was observed for medium-sized (100-399 beds) hospitals (8.6% vs 9.3% and 9.4%; 0.8% difference between 61 major and 1207 nonteaching hospitals [95% CI, 0.4%-1.3%]; P = .003). Among small (≤99 beds) hospitals, 187 minor teaching hospitals had lower overall 30-day mortality relative to 2056 nonteaching hospitals (9.5% vs 9.9%; 0.4% difference [95% CI, 0.1%-0.7%]; P = .01).
Conclusions and Relevance
Among hospitalizations for US Medicare beneficiaries, major teaching hospital status was associated with lower mortality rates for common conditions compared with nonteaching hospitals. Further study is needed to understand the reasons for these differences.
Promoting value is a central US health policy goal, and to this end, payers and policy makers are increasingly promoting efforts that steer patients away from higher-cost clinicians and hospitals.1 Academic medical centers (AMCs) are often considered more expensive than community hospitals2,3 and some insurers have excluded AMCs from their networks in an attempt to control costs,4 assuming that quality is comparable.
Because evaluating the value of medical care requires consideration of quality as well as cost, understanding whether teaching hospitals provide better care is critical. The seminal studies5-7 on this topic are 18 to 25 years old, and it is unclear whether those findings persist in the contemporary health care environment. Health care delivery has changed substantially from 2000 to 2017, driven by efforts such as quality improvement initiatives,8 digitization of the medical record,9 and changes in resident duty hour requirements.10,11 Identifying how patient outcomes differ between teaching and nonteaching hospitals in the current era is essential to understanding the value of health care provided at US teaching institutions.
In this study, contemporary national Medicare data were used to answer 3 key questions. First, to what degree do overall outcomes differ in teaching hospitals compared with nonteaching hospitals? Second, are the benefits of receiving care at a teaching hospital, if any, focused on a small number of conditions or are they present more broadly across multiple types of conditions and procedures? Third, are differences present even among large hospitals, where high volume could potentially mitigate any advantage of being a teaching institution?
This study was approved by the Office of Human Research Administration at the Harvard T.H. Chan School of Public Health. Informed consent was not obtained, as the data were obtained from previously collected, deidentified administrative data.
Hospitalizations were identified from the 100% Medicare inpatient file for years 2012 through 2014. Beneficiary characteristics and death date were obtained from the Medicare Beneficiary Summary file. Medicaid eligibility was determined using the State Buy-In Coverage Count variable. Any beneficiary with at least 1 month of state buy-in (Medicare premium paid by the state) was considered Medicaid eligible. Information on hospital characteristics was obtained from the American Hospital Association (AHA) annual survey and Medicare Impact File. Admissions to non–acute care hospitals, federal hospitals, and those outside of the 50 states and the District of Columbia were excluded. Additionally, admissions to hospitals without corresponding data in the AHA annual survey (2.2% of total) were excluded, as it was not possible to determine the primary predictor of interest, teaching status, for these admissions.
The primary exposure variable of interest was hospital teaching status. Consistent with other research,12-14 all hospitals were placed into 1 of 3 categories based on their response to the AHA survey: major teaching hospitals (those that are members of the Council of Teaching Hospitals [COTH]), minor teaching hospitals (non-COTH members that had a medical school affiliation reported to the American Medical Association), and nonteaching hospitals (all other institutions). For each hospital, data were obtained on its teaching status, size, geographic region, ownership (for-profit, private nonprofit, or public), rural vs urban location, and presence or absence of a medical and cardiac intensive care unit.
The study sample included beneficiaries who were 65 years or older and enrolled in the traditional fee-for-service program continuously for the entire year. For each hospitalization, the patient’s age, sex, race, Medicaid eligibility, and chronic conditions were obtained. Beneficiary race/ethnicity in the Medicare data were self-reported based on fixed categories.15 Beneficiary race/ethnicity was included to determine if differences in outcomes between teaching and nonteaching hospitals were associated with potential differences in racial and ethnic composition of their patients. Data for chronic conditions were obtained using software from the Centers for Medicare & Medicaid Services that allows for the creating of Hierarchical Condition Categories based on conditions coded in inpatient claims for that calendar year.
The primary outcome was death at 30 days from the admission date. Thirty-day mortality rates were calculated initially for all eligible hospitalizations and then for hospitalizations for individual medical and surgical conditions. The 15 most common medical causes of hospitalizations (using diagnosis related groups) were chosen, as well as 6 common costly surgical procedures across a variety of surgical specialties that have been previously used in studies of surgical quality.16,17 For secondary outcomes, 7- and 90-day mortality rates were also calculated for all hospitalizations and for the individual medical and surgical conditions.
After identifying hospitalizations among eligible beneficiaries in 2012-2014, patient and hospital characteristics were examined for each admission by hospital teaching status. Hospitalizations that ended in transfer were attributed to the original hospital. To illustrate the timing of mortality after hospital admission by teaching status, 3 Kaplan-Meier survival curves (for major teaching, minor teaching, and nonteaching hospitals) were constructed, with censoring of patients still alive at 90 days. Unadjusted overall mortality rates were calculated by specifying a linear regression model (eMethods in the Supplement) with each hospital’s overall 30-day mortality rate as the outcome and teaching status as the primary predictor. To account for regionally mediated differences in care, all models included state fixed effects, allowing for the effective comparison of teaching and nonteaching hospitals within the same state. Patient clustering within hospitals was accounted for using generalized estimating equations. To account for differences in patient severity, the model adjusted for principal discharge diagnosis related group weight and the following patient characteristics: age, sex, Medicaid eligibility, and Centers for Medicare & Medicaid Services Hierarchical Condition Categories. The Hierarchical Condition Categories model is used by the Centers for Medicare & Medicaid Services to publicly report hospital performance and for pay-for-performance programs. The final model also incorporated hospital volume, urban vs rural location, and profit status. This model was developed to assess the independent association with teaching status and the degree to which outcomes varied between teaching and nonteaching hospitals when other factors such as size and ownership were held constant.
Unadjusted and 2 adjusted models were also constructed with 30-day mortality rate as the outcome and teaching status as the exposure variable for each of the selected 15 medical conditions and 6 surgical procedures as well as the composite mortality for these selected conditions.
To examine if differences in mortality by teaching status persisted across hospitals of different sizes, the analysis was repeated stratifying by small (≤99 beds), medium (100-399 beds), and large (≥400 beds) hospital size. Given that there were only 2 small major teaching hospitals and 1 was a specialty cancer center, the analysis of small hospitals was limited to minor and nonteaching institutions. Linear regression was performed with 30-day mortality as the outcome and teaching status as the exposure variable with state fixed effects, as well as the patient and hospital characteristics described above.
P < .05 (2-sided) was considered statistically significant for the analyses of overall hospitalizations as well as the medical and surgical composite. A Bonferroni correction was applied to adjust the significance threshold to P < .002 (2-sided) for the analyses of the 21 individual medical and surgical conditions. Analyses were conducted using SAS version 9.4 (SAS Institute Inc).
To examine differences in early and longer-term mortality by teaching status, the previously described models were constructed with 7- and 90-day mortality as the outcomes.
Outcomes of transferred patients were assigned to the original hospital in the primary analysis, consistent with public reporting of mortality rates as well as other studies using mortality as a quality indicator.18,19 However, there is some evidence that transferred patients may have higher mortality relative to patients who are not transferred.20,21 Given that teaching hospitals tend to receive a greater number of transfer patients,14,22 attributing patients to the original hospital could bias the results against the hospitals in which these more complex episodes tend to originate. Thus, the analysis was repeated after completely excluding transfers.
To determine if the results were sensitive to the method of classifying patient comorbidities, the main models were repeated using Elixhauser conditions instead of Hierarchical Condition Categories.
To further evaluate the relationship between teaching intensity and outcomes, teaching intensity was examined as a continuous variable using intern/resident to bed ratio. Linear regression models were specified with 7-, 30-, and 90-day mortality as the outcomes, intern/resident to bed ratio as the predictor, and patient age, sex, Elixhauser conditions, Medicaid eligibility, and the same hospital characteristics described above as covariates.
The analytic sample consisted of 21 451 824 total hospitalizations at 4483 hospitals (Table 1; eTable 1 in the Supplement); 482 799 hospitalizations (2.2%) were excluded because of missing data in the AHA annual survey. Of the 4483 hospitals, 250 (5.6%) were major teaching hospitals and accounted for 16.7% of the admissions in the sample, 894 (19.9%) were minor teaching hospitals and accounted for 33.6% of admissions, and 3339 (74.5%) were nonteaching hospitals and accounted for 49.7% of admissions (Table 1). Patient characteristics for hospitalizations among major teaching, minor teaching, and nonteaching hospitals are presented in Table 1, as are key characteristics of the hospitals within each group.
In the unadjusted analyses of overall 30-day mortality, the mortality rates for major teaching, minor teaching, and nonteaching hospitals were 8.1%, 9.2%, and 9.6%, respectively, with major teaching hospitals having a 1.5% lower mortality (95% CI, 1.3% to 1.7%; P < .001) relative to nonteaching hospitals. This pattern persisted after adjusting for patient characteristics (8.0% mortality at major vs 9.1% at minor and 9.7% at nonteaching; 1.7% difference [95% CI, 1.5% to 1.9% ] between major teaching and nonteaching hospitals; P < .001) (Table 2). After accounting for hospital characteristics, major teaching hospitals had lower mortality rates relative to nonteaching hospitals, although the difference was smaller (1.2% difference [95% CI, 1.0% to 1.4%]; P < .001). Kaplan-Meier survival curves, presented in the Figure, show the trajectory of mortality at major teaching, minor teaching, and nonteaching hospitals.
Unadjusted overall 30-day mortality was 11.1% at major teaching hospitals and 11.8% at minor teaching and nonteaching hospitals (0.7% difference [95% CI, 0.4% to 0.9%] between major and nonteaching hospitals; P < .001), and this pattern persisted after adjusting for patient and hospital characteristics (Table 2). Major teaching hospitals had lower mortality than nonteaching hospitals for 11 of the 15 individual medical conditions examined (Table 3).
For the 6 surgical procedures, unadjusted mortality rates for major teaching, minor teaching, and nonteaching hospitals were 3.0%, 3.7%, and 4.3%, respectively, with a 1.2% difference (95% CI, 1.0% to 1.4%) between major and nonteaching hospitals (P < .001). This finding of lower mortality at major teaching hospitals compared with nonteaching hospitals persisted after adjusting for patient and hospital characteristics (Table 2). Major teaching hospitals had lower adjusted mortality rates than nonteaching hospitals for 2 of the 6 major procedures examined (Table 3), with lower mortality for open abdominal aortic aneurysm (AAA) repair (12.2% vs 16.9%; 4.7% difference [98.8% CI, 1.1% to 8.3%]; P < .001) and colectomy (7.0% vs 7.8%; 0.8% difference [98.8% CI, 0.2% to 1.5%]; P < .001).
In the analysis stratified by hospital size, there were significant differences by teaching status across each of the size groups (Table 4). Among large hospitals, the overall 30-day mortality rate was 8.1% for 187 major teaching hospitals, 8.9% for 185 minor teaching hospitals, and 9.4% for 76 nonteaching hospitals, with a 1.2% difference (95% CI, 0.9% to 1.5%) between major and nonteaching hospitals (P < .001). There was a similar pattern for overall medical 30-day mortality (11.0% vs 11.6% vs 12.0%; 1.0% difference between major and nonteaching hospitals [95% CI, 0.6% to 1.4%]; P < .001) and surgical 30-day mortality (3.2% vs 3.6% vs 3.8%; 0.7% difference between large major and large nonteaching hospitals [95% CI, 0.4% to 0.9%]; P < .001). Among medium-sized hospitals, 61 major teaching institutions had lower mortality than 1207 nonteaching hospitals for overall 30-day mortality (8.6% vs 9.4%; 0.8% difference [95% CI, 0.4% to 1.3%]; P = .003) and for 30-day surgical mortality (3.6% at major teaching hospitals vs 4.2% at nonteaching hospitals; 0.6% difference [95% CI, 0.2% to 0.9%]; P = .01), but there were no differences by teaching status for medical conditions in this size category (Table 4). Among small hospitals, 187 minor teaching hospitals had lower 30-day mortality overall compared with 2056 nonteaching hospitals (9.5% vs 9.9%; 0.4% difference [95% CI, 0.1% to 0.7%]; P = .01) and for medical conditions (11.3% vs 11.8%; 0.5% difference [95% CI, 0.1% to 0.9%]; P = .01). There was no statistically significant difference in mortality for surgical procedures between small minor teaching hospitals and small nonteaching hospitals.
Adjusted 7-day mortality was 3.3% at major teaching hospitals (3.3%), 3.6% at minor teaching hospitals, and 3.6% at nonteaching hospitals, with major teaching hospitals having 0.3% (95% CI, 0.2% to 0.5%; P < .001) lower mortality relative to nonteaching hospitals for all hospitalizations (eTable 2 in the Supplement), for the selected 15 medical conditions in aggregate as well as the selected surgical procedures in aggregate (eTable 3 in the Supplement). This pattern of lower mortality at teaching hospitals was observed for 7 of 15 medical conditions and for open AAA repair (eTable 3 in the Supplement). Adjusted 7-day mortality for open AAA repair was lower at major teaching hospitals compared with nonteaching hospitals (7.9% vs 11.6%; 3.7% difference [98.8% CI, 0.6% to 6.8%]; P = .002).
Ninety-day mortality was l3.8% for all hospitalizations at major teaching hospitals, 15.0% at minor teaching hospitals, and 15.5% at nonteaching hospitals. Mortality was lower at major teaching hospitals relative to nonteaching hospitals for hospitalizations overall (1.6% difference [95% CI, 1.3% to 1.9%]; P < .001) as well as for the 15 selected medical conditions in aggregate and the 6 selected surgical procedures in aggregate (eTable 4 in the Supplement). This pattern was observed for 13 of 15 medical conditions, with no differences by teaching status for stroke or sepsis (eTable 5 in the Supplement).
When these analyses were repeated excluding the 454 296 hospitalizations that ended in transfer, major teaching hospitals had lower adjusted mortality relative to nonteaching hospitals at 7, 30, and 90 days for all hospitalizations, for the composite mortality for the 15 selected medical conditions, and for the composite mortality for the 6 surgical procedures (eTable 6 in the Supplement). Using the Elixhauser risk-adjustment approach also demonstrated lower mortality for major teaching hospitals at 7, 30 and 90 days (eTable 7 in the Supplement). In addition, on examining teaching status as a continuous variable (using the intern/resident to bed ratio), every increase of 0.1 in intern/resident to bed ratio was associated with a 0.23% decrease in overall 30-day mortality (95% CI, −0.27% to −0.19%; P < .001) and a 0.31% decrease in overall 90-day mortality (95% CI, −0.37% to −0.26%; P < .001) (eTable 8 in the Supplement). Increasing intern/resident to bed ratio was associated with lower mortality for composite medical and surgical mortality as well as most individual medical conditions at 7, 30, and 90 days. This relationship was observed for most surgical conditions at 30 and 90 days but only for open AAA repair at 7 days (eTable 8 in the Supplement).
In this analysis of 21.4 million Medicare discharges during 2012-2014, admission to a major teaching hospital was associated with lower overall 30-day mortality compared with admissions to nonteaching hospital. These differences were observed overall and across a majority of conditions examined and persisted after adjustment for hospital characteristics, including volume. Lower mortality rates among teaching hospitals were present at both 7 days and 90 days and after excluding transfers, using an alternative adjustment model, and measuring teaching intensity as a continuous variable.
It is not clear why teaching status was associated with lower mortality. This difference in outcomes by teaching status may be related to greater experience treating particular conditions, but accounting for hospital volume did not substantially explain the differences. Teaching hospitals also tend to be early adopters of certain technologies,23 which could yield better outcomes for conditions that are more technologically intensive or require specialized knowledge. However, these results suggest better outcomes for a broad range of conditions, including pneumonia and heart failure, for which advanced technologies are helpful for only a minority of patients. A recent study found that teaching intensity was associated with higher performance on process measures for several conditions, suggesting that superior processes may explain the lower average mortality found at teaching hospitals in the present study.23 Further understanding of the mechanisms behind this association is important to determine whether these outcomes may be replicated at community hospitals.
These findings may be relevant to the recent changes in the broader health care delivery system. Narrow insurance networks have become more commonplace, and some have excluded teaching hospitals4 out of concern that they may be high cost. Some policy makers have also tried to steer patients with common conditions away from teaching institutions, citing higher costs without better outcomes.24 Additionally, all 3 of the national pay-for-performance programs established by the Affordable Care Act (the Value-Based Purchasing Program, Hospital Readmissions Reduction Program, and the Hospital Acquired Conditions Reduction Program) disproportionately penalize teaching hospitals.25,26 However, the findings of this study suggest that teaching hospitals have better outcomes, calling into question whether the national approach to measuring and rewarding on performance is working effectively.
Several studies in the 1990s and early 2000s found better outcomes at AMCs.6,7,27 More recent studies on outcomes at AMCs have focused primarily on a small number of conditions28,29 and often on in-hospital mortality (not accounting for the postdischarge care that can affect patient outcomes).29-31 Three recent studies that examined 30-day mortality, focusing on acute myocardial infarction, congestive heart failure, and pneumonia,14,32,33 found lower mortality among teaching institutions. One also found that hospital volume mediated some, but not all, of the benefit of teaching status for these 3 conditions.32 However, given the substantial policy attention to these 3 conditions in recent years,34,35 performance on these conditions may reflect targeted efforts. The present study extends this recent work by examining a broad range of clinical conditions and finds better outcomes at teaching hospitals as compared with nonteaching hospitals.
For 2 medical conditions, sepsis and stroke, outcomes were no better at teaching hospitals compared with nonteaching hospitals. Several studies have found recent changes in coding for sepsis such that patients with pneumonia and systemic signs of infection are far more likely to be coded as having sepsis with respiratory infection than they would have been a decade ago.36 Whether this occurs more systematically at nonteaching hospitals or for generally healthier patients is unclear, but differential changes in coding could explain why this study did not find lower mortality at academic centers. Some studies have suggested that teaching hospitals tend to code less aggressively and thus may be falsely penalized when administrative data are used for risk-adjusted quality metrics.37 Undercoding by teaching hospitals would tend to bias against finding lower mortality for AMCs. There is some evidence that teaching hospitals may be disproportionately misclassified as poor performers on stroke mortality using traditional adjustment models that do not adjust for stroke severity,38,39 although additional research is needed to further evaluate the degree to which types of stroke differ among teaching and nonteaching hospitals.
This study has several limitations. First, this study examined mortality rates for the Medicare fee-for-service population, and thus it was not possible to determine whether these findings are generalizable to nonelderly populations. Second, this study examined only mortality as an indicator of the quality of hospital care. It is not clear if other measures important to patients, such as functional status, differ significantly between teaching and nonteaching hospitals. Third, this study did not account for the patient preferences in end-of-life care. Lower mortality could, in theory, reflect underuse of palliative care for appropriate patients at teaching hospitals. The evidence regarding this idea is mixed, with teaching hospitals both providing more aggressive end-of-life cancer care but also referring to hospice at greater rates.40 However, even if differences in end-of-life care explained some of the short-term findings, the association would be expected to dissipate by 90 days, which was not the case in this study. Another potential limitation is that the measure of Medicaid eligibility, state buy-in coverage count, may undercount the number of Medicaid-eligible Medicare beneficiaries in some states. Additionally, because of the observational design, the differences in outcomes could represent unmeasured confounding. However, teaching hospitals appear to have higher mean case mix indices compared with nonteaching hospitals, which would bias outcomes data against a finding of lower mortality at teaching hospitals.23
Among hospitalizations for US Medicare beneficiaries, major teaching hospital status was associated with lower mortality rates for common conditions compared with nonteaching hospitals. Further research is needed to understand the reasons for these differences.
Corresponding Author: Ashish K. Jha, MD, MPH, Harvard T.H. Chan School of Public Health, 42 Church St, Cambridge, MA 02139 (email@example.com).
Accepted for Publication: April 25, 2017.
Author Contributions: Dr Burke had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Concept and design: Burke, Khullar, Orav, Jha.
Acquisition, analysis, or interpretation of data: Burke, Frakt, Orav, Jha.
Drafting of the manuscript: Burke.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Burke, Orav, Jha.
Obtained funding: Burke, Jha.
Administrative, technical, or material support: Khullar, Jha.
Conflict of Interest Disclosures: All authors have completed and submitted the ICMJE Form for Disclosure of Potential Conflicts of Interest and none were reported.
Funding/Support: This study was funded by Association of American Medical Colleges.
Role of the Funder/Sponsor: The Association of American Medical Colleges had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, and approval of the manuscript; or decision to submit the manuscript for publication.