In 1983, the Association of Program Directors of Internal Medicine (APDIM) studied the demographics of residency program directors in internal medicine.1 Program directors, on average, had been in their positions for five years; 43% had held their positions for less than three years. Recently, a representative from the Accreditation Council for Graduate Medical Education's (ACGME's) Residency Review Committee (RRC) for Internal Medicine noted a large turnover in the address list for program directors and demonstrated concern at the APDIM convention.2
The learning curve for a program director to become optimally effective is several years long. Maturity and experience within the culture of an institution likely enhance one's success as an administrator, role model, and mentor. Curriculum development, implementation, and successive cycles of improvement require several years.3 There may be characteristics of programs and program directors that predispose them to high or low job turnover, but the correlates of job turnover among program directors have not been comprehensively studied.
The purposes of this study were (1) to determine the turnover rate for program directors of U.S. internal medicine residency programs; and (2) to identify characteristics of residency programs and program directors that are associated with job turnover.
To develop the questionnaire, we searched Medline databases for studies published from 1966 to 1996 about residency program directors, physicians' job satisfaction, and career moves. We held discussions with current and former program directors to identify factors potentially associated with program director job satisfaction and turnover. And, we asked authors of previous program director studies to provide us with their survey instruments.1,4
We designed a questionnaire with four main sections. The first two sections elicited respondents' demographics and specifics about the residency program (size, locale, etc.). The third section contained a four-question global job-satisfaction measure used in previous studies5,6 and a 20-item job satisfaction measure (PD-Sat) that was facet-specific and was designed for this study.7 The global measure gauged respondents' general affective feelings about their jobs8 and the facet-specific measure measured the respondents' satisfaction with components of their jobs such as pay, resources, and items amenable to change. In the third section, we also asked respondents to describe their degrees of “burnout,” defined as “the loss of enjoyment or enthusiasm for a job, so that an individual is no longer able to devote emotional energy to its accomplishment.”
The fourth section of the questionnaire asked how long the respondents had been in their current positions as program directors. It also asked whether the respondents were “not considering,” “considering,” “actively seeking,” or “definitely making” a career move. The subset of respondents who were considering, actively seeking, or definitely making a career move were asked to indicate the importance of 13 potential reasons on a five-point Likert scale, and these respondents were also asked to write the “top three” reasons that they were considering, actively seeking, or definitely making a move.
We obtained a mailing list from the ACGME of all non-military internal medicine residency program directors in the continental United States in October 1996.9 We excluded military and non-continental program directors because their lengths of job stay were likely affected by different variables. Our final list contained 386 potential participants.
The survey was administered from November 1996 through March 1997. The questionnaire was accompanied by a letter of support, on Society of General Internal Medicine stationery, from prominent academic internists, including the president and several past presidents of the society. Non-respondents received second and third requests with duplicate surveys at six-to-eight-week intervals; postcard reminders were sent between mailings. The survey was confidential, and respondents were tracked by numerical codes.
In October 1999, we obtained an updated list of the program directors from the ACGME. The ACGME's program requirements state that institutions must “notify the RRC promptly of… a change in the program director.”10 Therefore, the ACGME's address list should be an objective and accurate measure of program director turnover. We compared it with the list from 1996, and the programs that listed different program directors were contacted to (1) confirm there had been a change, and (2) find the month and year that the program director had left the job. The turnover data were added to the dataset in such a way as to ensure the confidentiality of the questionnaire's respondents.
Each survey was double-data entered. We used a standard statistical software package for statistical analysis. We combined response categories for variables when sparsely selected responses were identified. We examined continuous variables for evidence of skewness, outliers, and non-normality. Mean substitution for missing data was used for the PD-Sat and the global job-satisfaction measures if 75% or more of the items were completed, and for the individual facets of PD-Sat if 60% or more of the items were completed. Continuous variables were described using distributions, means, medians, standard deviations, and ranges. To facilitate presentation and interpretation of the bivariate and multivariate analyses, we recoded continuous variables into three or more categories (usually “low,” “medium,” and “high”).
Independent variables were demographics, professional background, job and program characteristics, and program directors' attitudes. The length of job stay during the three-year study period was the main outcome variable. For the bivariate analyses, we calculated proportional hazard ratios and 95% confidence intervals (CIs) for the associations of the individual independent variables with the length of job stay during the study period. An “event” was defined as occurring when there was a change in an institution's program director listing by the ACGME. If the program director was still in his or her position at the time of follow-up, the length of stay was calculated to the date of follow-up (October 1999) and the program director was censored.
We used multivariate Cox proportional-hazards regression to identify variables that were independently associated with the outcome variable. We considered all the variables used in the bivariate analyses as potential candidates. To guard against multicollinearity, all possibly related independent variables were examined in a series of bivariate analyses.
For multivariate Cox proportional-hazards regression modeling, we employed a backward elimination procedure using the change in chi-square statistic as the criterion for model inclusion. Modeling was repeated using a forward stepwise procedure to confirm the model from the backward procedure. We arrived at the same model with both methods.
We set significance for chi-square tests, t-tests, Cox proportional hazards, Pearson and Spearman correlations, and regression analyses at a probability of less than .05 using two-sided tests. The Johns Hopkins Bayview Medical Center Institutional Review Board approved the study.
A total of 301 (78%) of the 386 potential participants completed the questionnaire. Respondents and non-respondents did not differ in gender or age. Respondents (31%) were more likely than were non-respondents (7%, p < .05) to hold the titles of both department chair and program director.6
Since we were interested in programs' and program directors' characteristics that might influence a program director's career decision, we removed from the analysis the respondents from the 14 programs that closed during the study period. We also removed the 25 respondents who indicated they were assistant program directors because they were not our primary subjects and we had no way of tracking them through the ACGME's address lists. This left a final cohort of 262 program directors.
Figure 1 diagrams the outcomes for the respondents and non-respondents. A total of 29.4% of the cohort were no longer listed as the program directors by the ACGME at the end of the three years, a proportion that was not significantly different for non-respondents (28.2%). The mean age of the cohort was 47.4 years (SD = 8.0), 16% were women, and 7% were minority. The majority of the respondents were subspecialists (60%).
Potential Reasons for Career Moves
The subgroup of program directors (34%) who were considering, actively seeking, or definitely making career moves indicated the importance of 13 potential reasons for doing so. The reasons are listed in descending order of importance in Table 1. “Administrative hassles” was rated as most important, followed by “expand my horizons,” and the “institution does not support the residency program.”
We also asked the program directors to write their top three reasons for considering, actively seeking, or definitely making career moves. These qualitative responses mirrored Table 1 with one exception: “Frustration with the Residency Review Committee rules and regulations” was the fifth most-cited reason for considering, seeking, or making a career change.
Measures of Turnover
At the beginning of the study period, program directors had been in their positions for a mean of 5.3 years (see Figure 2). Forty-nine percent had been on the job three years or less. However, the range was wide (maximum = 39 years).
From the beginning of the cohort study (October 1996) to the end (October 1999), the mean length of job stay was 2.4 years (SD = 0.8 years), with a median of 2.7 years. Figure 3 shows a Kaplan-Meier curve of the time until job turnover, measured in days. The turnover rate was 29.4% (95% CI = 23.5–35.3) over the three-year period—nearly 10% every year.
Characteristics Associated with Job Turnover
Table 2 lists the characteristics of the program directors and their residency programs alongside their crude hazard ratios for job turnover. Items in program directors' demographics and background training that trended toward an association with job turnover, but were not statistically significant, were a younger age, formal training in time management, and formal training in dealing with problem residents. Gender, rank, and subspecialty training showed minimal if any associations with turnover.
The only job characteristic to be significantly associated with turnover was spending a large percentage of time in administration, although being on the job less than two years, being concomitantly a program director and chair or chief of a department, and the percentage of work time spent attending on the wards all trended toward associations. Salary was not associated with turnover.
One program characteristic, a moderately high percentage of international medical graduates (IMGs) in the residency program, was significantly associated with job turnover. The category with the highest percentage IMGs was not associated with turnover. The setting, size, and number of support personnel in the program had negligible associations with turnover.
Finally, most of the job satisfaction and attitude facets trended toward an association with turnover, but only satisfaction with colleague relationships was strongly significant. Global job satisfaction and PD-Sat (the sum of the facet-specific job satisfaction measures) were both significantly associated with turnover.
Because the global job satisfaction measure was a broad-based attitudinal scale, it provided no specific information about programs, program directors, or job characteristics that could be targeted for change. Therefore, it was excluded from multivariate modeling.
Cox proportional-hazards regression analysis demonstrated that four variables were independently associated with job turnover: low satisfaction with colleague relationships (hazard ratio = 3.2, 95% CI = 1.6–6.4), high percentage of administrative work time (HR = 2.9, 95% CI = 1.4–6.2), perceiving the job as a steppingstone (HR = 1.8, 95% CI = 1.0–3.2), and having had formal training to deal with problem residents (HR = 0.6, 95% CI = 0.4–1.1). Although the latter variable was of only borderline statistical significance, it significantly improved the regression model (see Table 3).
Items from the questionnaire that composed the colleague-relationships satisfaction scale (1 = strongly disagree, 7 = strongly agree) were “My superiors value my work,” “The people with whom I work share a sense of mission,” “I feel valued by my colleagues,” “Colleagues follow my administrative leading,” and “I have freedom to make important decisions in the residency program.”
In the 1983 APDIM study of internal medicine residency program directors, the average program director had been in his or her position for five years, while 43% had been in their positions for less than three years.1 A 1992 survey of pediatrics program directors demonstrated a similar pattern, with the majority of respondents in their positions for six years or less.4 The respondents in our study averaged 5.3 years on the job, 39% had been on the job for less than three years, and 70% had been on the job for six years or less. Turnover rates seem to be similar across a gamut of academic administrators, including department of medicine chairs,11 family medicine chairs,12 and physical medicine and rehabilitation chairs.13 Some job turnover may be desirable, if those leaving are poorly matched to succeed in their jobs. The departure of an effective administrator, however, is likely to be detrimental to a program. Our findings suggest that a significant proportion of internal medicine program directors and other academic administrators may not remain in their positions long enough to be maximally effective in implementing and institutionalizing changes that address their programs' needs.
Given the potential negative impact of job turnover, associated factors are of interest. To our knowledge, this is the first study of the correlates of job turnover in a large group of academic medical administrators. We demonstrated four variables among internal medicine residency directors that were independently associated with turnover: low satisfaction with colleague relationships, a high percentage of administrative work time, perceiving the job as a steppingstone, and having had formal training to deal with problem residents.
Our findings echo those of other studies. A survey of Canadian physicians demonstrated that a major source of job satisfaction was the respondents' relationships with their colleagues.14 In one nursing study, nurses with few opportunities to meet with colleagues scored higher on emotional exhaustion and lower on personal-accomplishment scales.15 In another study, characteristics associated with a successful academic medicine career included, a “positive group climate,” “frequent communication,” and “accessible human resources.”16 The findings of these studies and ours highlight the importance and power of interpersonal connectedness in a program director's professional life. Sharing a sense of mission and feeling valued by colleagues and superiors were both strong negative correlates of job turnover.
Having to spend a high percentage of time in administration independently and negatively correlated with the turnover rate among program directors. Administrative hassles and meeting the requirements of the RRC, reasons cited by respondents for considering, actively seeking, or definitely making career moves, are inherent parts of a program director's job. Frustrations with these aspects of the job may relate more to a program's director's interests and organizational skills than to program structure, because the number and type of support staff did not correlate significantly with measures of job turnover.
Program directors are, for the most part, mid-level administrators. It is not surprising, therefore, that 19% of the respondents perceived their jobs as steppingstones, and that this was one of the strongest independent factors associated with job turnover. It is not clear whether the characteristic of seeing the position of residency program director as a steppingstone in one's career advancement has adverse consequences for a residency program. Other factors that could help assess the impact of this variable, such as job effectiveness, were not measured in this study.
Although having had formal training in dealing with problem residents was not statistically significant in the multivariate model, as an independent variable it improved the model and seemed to provide protection against job turnover (HR = 0.6). Yao and Wright17 recently surveyed internal medicine residency program directors and found a 6.9% prevalence of problem residents. Chief residents, clinical faculty, and program directors were the most frequent identifiers of problem residents. Program directors believed frequent feedback and mentoring were the most critical components of promoting behavioral change in this difficult group. If one of every 14 residents could be identified as a problem resident, a program director not prepared for this work would be encumbered.
It is encouraging that several of the characteristics we studied were not associated with job turnover. Program directors who changed jobs were no different in rank, specialty status, size and setting of their programs, and the amount of support personnel from those who stayed in their jobs.
This study has several strengths. It was a national cohort study with a high response rate (78%) to the initial questionnaire, which helped to ensure a representative sample of subjects. Also, respondents to the questionnaire verified whether they were the program director. This ensured that the people followed through the ACGME's address lists for three years were the same as the respondents to the confidential questionnaire. Follow-up was sufficiently long, allowing us to track the trend and demonstrate that it was steady. Because of the study's design, we were able to follow up with every respondent. We used objective and unbiased outcome criteria (the ACGME's address list) and verified the changes by phone calls. We formerly demonstrated that the facet-specific job-satisfaction measure (PD-Sat) was reliable and valid.7 Multivariate analyses allowed us to adjust for important prognostic factors. The outcome measure was a discrete event, and the dates (month and year) of turnover were precisely provided by the departments in which the program directors had been employed. The power of the study was adequate to provide a confidence interval on the turnover with a spread of only 11.8%.
The study's limitations should be considered in interpreting the findings. In a cohort study, it is best to have the participants be at similar points in the “course of their disease.”18 In our study, 49% of the program directors had been on the job for three years or less. However, 30% had been on the job for more than seven years—one for 39 years! We adjusted for this variable in multivariate analyses. It played only a small role in overall turnover. Also, we had no data on job efficacy, which could surely impact job turnover.
In conclusion, job turnover for internal medicine residency program directors is substantial. We have documented some significant differences between internal medicine residency program directors who make job changes and those who do not. While the findings of the study can not be assumed to apply to program directors in other specialties, the method of this study could be used by national organizations, such as the ACGME, to study these trends among other specialties. Our study may prove useful to the APDIM as a needs assessment to direct career development programs. The findings should also be of use to those who are recruiting residency program directors, to those who are seeking to retain program directors and foster long-term program development, and to those who aspire to become program directors.