Association of Time to Attrition in Surgical Residency With Individual Resident and Programmatic Factors

Key Points

Question  Which residents drop out of general surgery training, and when?

Findings  In a large national cohort study of 836 US surgical residents with long-term follow-up, attrition from general surgery residency was highest during the internship year. Attrition persisted into later years of training among women and Hispanic residents and residents in large programs.

Meaning  Interventions to improve graduate surgical education could address diversity by focusing on the residents at risk of late attrition.

Importance  Attrition in general surgery residency remains high, and attrition that occurs in the later years is the most worrisome. Although several studies have retrospectively investigated the timing of attrition, no study to date has prospectively evaluated a national cohort of residents to understand which residents are at risk for attrition and at what point during residency.

Objective  To prospectively evaluate individual resident and programmatic factors associated with the timing of attrition during general surgery residency.

Design, Setting, and Participants  This longitudinal, national cohort study administered a survey to all categorical general surgery interns from the class of 2007-2008 during their first 30 days of residency and linked the data with 9-year follow-up data assessing program completion. Data were collected from June 1, 2007, through June 30, 2016.

Main Outcomes and Measures  Kaplan-Meier curves evaluating time to attrition during the 9 years after the start of residency.

Results  Among our sample of 836 residents (306 women [36.6%] and 528 men [63.2%]; gender unknown in 2), cumulative survival analysis demonstrated overall attrition for the cohort of 20.8% (n = 164). Attrition was highest in the first postgraduate year (67.6% [n = 111]; absolute rate, 13.3%) but continued during the next 6 years, albeit at a lower rate. Beginning in the first year, survival analysis demonstrated higher attrition among Hispanic compared with non-Hispanic residents (21.1% vs 12.4%; P = .04) and at military programs compared with academic or community programs after year 1 (32.3% vs 11.0% or 13.5%; P = .01). Beginning in year 4 of residency, higher attrition was encountered among women compared with men (23.3% vs 17.4%; P = .05); at year 5, at large compared with small programs (26.0% vs 18.4%; P = .04). Race and program location were not associated with attrition.

Conclusions and Relevance  Although attrition was highest during the internship year, late attrition persists, particularly among women and among residents in large programs. These results provide a framework for timing of interventions in graduate surgical training that target residents most at risk for late attrition.

The rate of attrition from general surgery residency training programs continues to be higher than that of many other medical and surgical specialties, including internal medicine, orthopedics, neurosurgery, and otolaryngology.^1-6 Although some attrition early in residency (years 1 and 2) might be expected, late attrition among senior and chief residents who train for years only to leave before graduation is a major concern because this represents loss of time and investment for the resident and the program. During the past 30 years, several reform measures were implemented by the American Board of Surgery (ABS) and the Accreditation Council for Graduate Medical Education to reduce the rate of resident attrition. These reforms included elimination of the pyramidal training system in 1982 and introduction of Accreditation Council for Graduate Medical Education work hour regulations in 2003 and 2011. Nevertheless, attrition from general surgery residencies remains high.^1,2,7-10

To date, no prospective national cohort study has tracked general surgery residents from the start of their training through board certification; as a result, the timing of attrition in graduate surgical education is poorly understood.^11-14 Single-institution and cross-sectional studies have suggested that the highest attrition is seen in the first 2 years of residency.^2,8

With support from the ABS and the Association of Program Directors in Surgery, the National Study of Expectations and Attitudes of Residents in Surgery (NEARS) was launched in 2007. The study aimed to provide longitudinal yearly follow-up data for categorical general surgery interns from the class of 2007-2008; data from NEARS offer the opportunity to analyze the potential association between individual resident and residency program variables and attrition for the duration of the residency. This analysis focuses on late attrition, that is, attrition after the first 2 years. Prior work has shown that attrition varies by gender and race/ethnicity.¹⁵ We hypothesized that women and nonwhite residents would experience a distinct time course of attrition because of the unique obstacles that they potentially face.

Quiz Ref IDThis study provides follow-up to June 30, 2016, 9 years after the inception of a prospective cohort on June 1, 2007, consisting of all categorical general surgery interns across the United States who participated in the NEARS Intern Survey in the first 30 days of the 2007-2008 academic year (eFigure in the Supplement).² The initial results of the NEARS cohort were cross-sectional. Herein, we report the long-term longitudinal attrition of the interns included in that cohort. This study was approved by the institutional review board at Yale University School of Medicine, New Haven, Connecticut, and subsequently approved and transferred to the Weill Cornell Medical College institutional review board, New York, New York.

The ABS was not given access to any individual resident or program responses from the NEARS survey, as stipulated in our initial agreement with the ABS and as stated in the informed consent. Only residents with survey information were included for analysis. Participants with missing survey information were excluded from the bivariate analyses when necessary. To ensure that survey respondents were not a biased sample, the ABS confirmed that attrition in the nonrespondent group was similar to that of the respondent group, although this information was available for only the first year of follow-up. All residents with survey information were queried against ABS records; matched data provided by the ABS included ABS In-Training Examination (ABSITE) dates, documentation of completion of surgical training, board eligibility, and board certification. Demographic data collected on the initial NEARS survey included resident gender, race/ethnicity, marital status, and medical school location. Additional information regarding training programs was provided by the ABS and included the program geographic location (Northeast, South, Midwest, or West), the size of the chief resident class, and classification as an academic, community, or military program. Program size was dichotomized based on ABS groupings, with large programs defined as having at least 6 chief residents and small programs defined as having fewer than 6 chief residents.

Demographic and program characteristics were compared between participants who the ABS documented as having completed training and those who did not. Program completion was defined as documented graduation according to the ABS. For those who were confirmed to have started their intern year but did not complete residency, attrition was defined as occurring during the end of the last year for which there was documented administration of the ABSITE. For those who were confirmed to have started their internship year but who did not complete residency training and did not have a single ABSITE score, attrition was assigned to have occurred in the middle of the first (internship) year.

Kaplan-Meier survival analysis was used to describe attrition over time. Estimated attrition at each calendar year was calculated and stratified by resident gender, race/ethnicity, and other sociodemographic characteristics, as well as training program variables. We compared the cumulative risk of attrition over time and determined how many years after beginning residency the difference became significant. To achieve this measure, attrition was examined with different length of follow-up (from <1 to all years) and the log-rank test was performed for each follow-up length.

A multivariable Cox proportional hazards model was used to assess time to attrition and risk factors among participants who had complete information for gender, race/ethnicity, and the type and size of the program in which they were enrolled. Covariates included in the model were participant gender, race/ethnicity, family status, whether the intern attended a US or a Canadian medical school, whether they had family living nearby (defined as within driving distance) and/or family members in the medical profession, and the location, size, and type of the training program. A missing category was created for missing data for variables other than the 4 risk factors of interest. A random effect of program cluster was tested and determined to be insignificant. Therefore, a simple model without random effects was used in the main analysis. Multicollinearity among gender, race/ethnicity, and program size, location, and type was evaluated using variance inflation factors. No significant collinearity was found. Interactions between these 6 covariates were explored and were not significant. Proportional hazards assumptions were examined with Schoenfeld residual plots and by fitting an interaction variable between the selected risk factor and time. Gender was found to have a trend toward violating the proportional hazard assumption (P = .09), and therefore a time-dependent gender covariate was tested. All P values are 2-sided, and P < .05 was regarded as significant. All analyses were performed using SAS software (version 9.3; SAS Institute Inc).

We identified a total of 1048 categorical general surgery interns who started their training in 2007. Survey linkage information was available for 836 of these (79.8%) from 206 US general surgery residency programs (Figure 1). The final cohort of 836 residents included 306 women (36.6%) and 528 men (63.2%), with gender unknown in 2 (Table 1). We found no differences between men and women regarding race or ethnicity. Women were more likely than men to enter residency unmarried and without a partner (127 of 305 [41.6%] vs 168 of 527 [31.9%]; P < .01). Women were also more likely to train at a program in the Northeast (101 of 306 [33.0%] vs 148 of 528 [28.0%]; P < .01). A small minority of all residents were of Hispanic ethnicity (76 [9.1%]) (Table 2). Residents of Hispanic ethnicity were more likely than non-Hispanic residents to enter residency while married with children (15 of 76 [19.7%] vs 81 of 759 [10.7%]; P = .02). Hispanic residents were less likely than non-Hispanic residents to train at a large residency program (12 of 76 [15.8%] vs 209 of 759 [27.5%]; P = .03).

Quiz Ref IDThe overall attrition for the study cohort was 20.8% (n = 164). Most of the attrition (67.7%) occurred in the first postgraduate year (absolute rate, 13.3% per year) but continued over subsequent years, albeit at a slower rate (absolute rate, 0.6%-2.7% per year). When examining the time course of attrition during the course of residency training, significant observations were made around 4 variables (resident gender, Hispanic vs non-Hispanic ethnicity, program size, and program type), but these differences affected residents at different times along the training continuum. Quiz Ref IDMen and women experienced similar attrition in the first year (13.7% vs 12.7%; P = .66). However, after postgraduate year 1, attrition diverged by gender, with women experiencing higher attrition compared with men (Figure 2A). After 4 years of residency, this risk of attrition was significantly different (21.9% for women vs 16.3% for men; P = .05). Quiz Ref IDHispanic ethnicity was associated with a significantly higher risk of attrition compared with non-Hispanic ethnicity (Figure 2B), and this difference was significantly different after the first year (21.1% for Hispanic vs 12.4% for non-Hispanic residents; P = .04). There was no difference in attrition based on resident race (Figure 2C).

Residents at large training programs experienced higher attrition compared with residents at small programs (Figure 2D). This difference was only significant after year 5 (23.6% vs 17.5%; P = .05). After the first year, residents at military programs (Figure 2E) experienced significantly higher attrition compared with those at community and academic training programs (32.3% vs 11.0% and 13.5%, respectively; P = .01). Quiz Ref IDThere was no difference in the time course of attrition between academic and community training programs and no differences in the observed course of attrition during residency based on the geographic location of the training program within the United States (Figure 2F).

Given the distinct nature of military training, a sensitivity analysis of time to attrition was performed (eTable 1 in the Supplement). Excluding military residents did not change our findings regarding differences in time to attrition based on gender, race, or ethnicity.

A Cox multivariable proportional hazards model was constructed to confirm these findings (eTable 2 in the Supplement). After adjustment, cumulative attrition was noted to be higher for women (odds ratio [OR], 1.40; 95% CI, 1.02-1.94), residents of Hispanic ethnicity (OR, 1.71; 95% CI, 1.06-2.76), and residents in military programs (compared with academic programs; OR, 2.68; 95% CI, 1.36-5.29).

This study is the first longitudinal prospective study, to our knowledge, to evaluate the time course of resident survival in categorical general surgery residency programs. Overall, attrition in a longitudinal cohort at a national level was 20%, consistent with the initial cross-sectional results of this cohort as well as findings from other studies.^1,2,7,8 Most attrition (67.7%) took place in the first postgraduate year. Although men and women had similar cumulative attrition during the first 3 years of residency, beyond that point, women had a significantly higher cumulative attrition. Although Hispanic ethnicity was associated with higher attrition compared with non-Hispanic ethnicity, this difference was noted earlier (after 1 year) than the difference between men and women. Other factors associated with an increased risk of attrition were training at a military program (1 year after beginning residency) and training at a large program with at least 6 chief residents (5 years after beginning residency).

Attrition early in training may be appropriate because learners potentially determine that their expectations for residency are not well aligned with their actual experiences. Dodson and Webb⁸ evaluated attrition rates during a 13-year period at a single academic general surgery residency program and demonstrated that more than half of the residents (56%) who voluntarily left their training programs did so in the first and second years, with the remainder (44%) leaving in the third and fourth years.⁸ Prior cross-sectional work from the initial NEARS study^2,16 demonstrated that across all postgraduate years (including time dedicated to research or other related pursuits), the cumulative risk of resignation was 19.5%. Attrition was highest in the first year (5.9%), second year (4.3%), and dedicated research year(s) (3.9%).^2,16 Gifford et al¹¹ conducted a cross-sectional survey of 288 surgery residents at 13 residency programs in the United States and evaluated which residents had considered leaving training. They found that being a first-year (45.8%), second-year (41.4%), or research (35.4%) resident was associated with considering leaving training, although training level was not a significant risk factor for having considered leaving residency on multivariable analysis. Our study differs from the analysis by Gifford et al¹¹ because it measured behavior and actual attrition rather than just consideration of dropping out.

Late attrition is more worrisome than early attrition, because it implies that significantly more resources were potentially wasted at the program level and by the resident and their family and support structure. Evidence suggests that financial debt is greater among groups with higher late attrition. The median debt for graduating medical students in 2016 was $190 000, and some evidence suggests that underrepresented minority students have higher debt; in 1 national study of educational debt from 2013,¹⁷ a higher proportion of black medical students had debt exceeding $150 000 compared with white and Asian medical students (77% vs 65% and 50%, respectively; P < .01). Debt is high among general surgery and surgical subspecialty trainees; the median debt after completing residency for junior surgery faculty in 2012 was approximately $100 000.^18-20 The extent to which debt may contribute to late attrition remains unclear, and this should be a focus of future research.

Of particular concern, the residents at highest risk for late attrition were women and minorities, who are already underrepresented along the training pipeline.²¹ For context, the difference observed between men and women in general surgery is similar in scale to the cumulative overall 5-year attrition rate seen in otolaryngology training programs at a national level.⁵ This finding of higher attrition among women is consistent with the research of Gifford et al,¹¹ which demonstrated that women were more likely to continue to have serious thoughts of leaving residency in later years compared with men. Some of this late attrition may be due at least in part to the effect that starting a family has on women as they move through their residency training, because traditionally women have taken greater responsibility for childcare. A recent national study²² demonstrated the perception among surgery program directors that having a child has a negative effect on training and well-being of women residents.

An important consideration that may help explain late attrition is related to the time that residents spend away from their clinical responsibilities pursuing dedicated research, because this has been shown to be a particularly vulnerable time for residents.^16,23 It typically occurs after the second or third postgraduate year. Based on the NEARS cross-sectional reports,²⁴ research residents were less satisfied with surgical training, suggesting that they may be vulnerable to leaving surgery during this period. Research time is also when many residents decide to start a family, and changes in priorities may place a strain on work-life balance and contribute to burnout.

Late attrition from large programs is likely multifactorial but may be driven by the nature and strength of interpersonal relationships. Residents at such programs may not feel as personally invested or may have weaker personal relationships with faculty and coresidents. Similarly, attrition during research years may reflect isolation and a loss of protective personal relationships. This feeling may be amplified in the larger programs, which generally encourage or require that their residents perform research and have higher overall levels of attrition. These programs have the potential to be more isolating if a resident does not fit in well. Previous work from NEARS²³ has demonstrated the importance of social interaction and fit, with residents at large programs reporting fewer social interactions, on average, with attending physicians.

This study provides a framework for how to intervene on a national scale to reduce attrition in general surgery training. For example, the bulk of attrition occurs in the first year of training. Therefore, 1 method for reducing early attrition may be to better educate medical students interested in general surgery and related specialties about what to expect during residency. Other groups who may need additional resources include women, Hispanic students, and trainees in military programs. Although women are more likely to enter residency while unmarried or without a partner, they may start families during the research years and/or as senior residents and, as a result, might experience unique challenges balancing clinical and family responsibilities, which may explain why cumulative attrition for women is different from that of men after 4 years of residency. Therefore, 1 intervention could be to better support women as their personal and family lives evolve. Support for flexible childcare would be a direct way to facilitate this. In contrast, Hispanic residents may need more support during the transition from medical school to the internship year, given their higher risk of attrition after 1 year of residency. One explanation for this observed higher early attrition is that Hispanic residents are more likely to start residency while married with children; as a result, they may have more difficulty balancing family and work during the internship year, when acclimation to a new program and lifestyle is particularly acute. Military requirements might compel some residents to leave training when they wish to continue. This restriction may explain the higher attrition observed in military-based programs compared with community and academic programs, but more research is needed to evaluate factors associated with attrition in the military.²⁵

Limitations of this study include the lack of individual data on childbirth and family, lifestyle changes, and stressors over time. The NEARS internship survey only provided baseline information about these factors at the time that residents entered their internship. However, other data suggest that many women choose to have children and start their families during residency training, and this may place them at higher risk for late attrition. Some interns may have taken longer than 9 years to complete residency, but this could not be captured owing to the duration of our follow-up. Long-term follow-up is only available for a cohort of interns from a single year, which may limit generalizability of our findings. Graduate surgical education has continually evolved since the inception of our cohort. Duty hour changes, increasing recognition of attrition as a problem, and the individual efforts of programs may have reduced attrition in the interim. How the current pace of attrition compares with findings in this cohort, which should be a focus of future work, remains unclear. Finally, because survey questions were only posed to interns and because the ABS has no way of documenting which residents are in research years or how long residents spend in research years, we cannot know which interns pursued dedicated research or whether they needed to repeat a year.

Overall, attrition in surgery is higher than in other specialties. This study highlights differences not just in which residents leave surgery, but when. Although attrition is highest during the internship year, late attrition appears to be a problem, particularly for women, Hispanic residents, and residents in large or military programs. Future interventions should focus on late attrition, because it erodes diversity in surgery and is particularly costly to residents and programs. Such interventions might include better social support, improved child care, longitudinal mentorship, and more flexible scheduling.

Accepted for Publication: December 1, 2017.

Corresponding Author: Heather L. Yeo, MD, MHS, Department of Surgery, New York–Presbyterian and Weill Cornell Medical Center, 525 E 68th St, PO Box 172, New York, NY 10065 (hey9002@med.cornell.edu).

Published Online: February 21, 2018. doi:10.1001/jamasurg.2017.6202

Author Contributions: Drs Yeo and Abelson share first authorship and did equivalent work. Dr Yeo had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study concept and design: Yeo, Abelson, Mao, Bell, Sosa.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Yeo, Abelson, Symer.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Yeo, Symer, Mao.

Obtained funding: Yeo, Sedrakyan.

Administrative, technical, or material support: Yeo, Abelson, Symer, Bell, Sedrakyan.

Study supervision: Yeo, Michelassi, Bell, Sosa.

Conflict of Interest Disclosures: Dr Yeo reports her spouse receives a salary from Bioscrip, Inc, a home infusion company. Dr Bell reports being the owner and president of Discourse LLC, a company that creates educational software for health care. Dr Sosa reports serving as a member of the Data Monitoring Committee of the Medullary Thyroid Cancer Consortium Registry supported by Novo Nordisk, GlaxoSmithKline, AstraZeneca, and Eli Lilly and Company. No other disclosures were reported.

Funding/Support: This study was supported by the Robert Wood Johnson Foundation and by grant T32[HS000066-23] from the Agency for Healthcare Research and Quality (salary support for Drs Abelson and Symer).

Role of the Funder/Sponsor: The sponsors had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.