Effect of the Rural Interprofessional Faculty Development Initiative on VHA Clinician Turnover by Level of Program Engagement

ABSTRACT

Purpose

VHA's Rural Interprofessional Faculty Development Initiative (RIFDI) is a structured program designed to develop teaching and leadership skills among clinical educators in rural settings. While prior literature indicates engagement in professional development programs increases retention of staff in health systems, no prior work has quantitatively evaluated the relationship between level of engagement and the associated effect on retention. We evaluated the impact of RIFDI on employee turnover by level of program engagement among VHA clinicians nationwide.

Methods

We employed a quasi‐experimental study design using RIFDI program records and VHA administrative data from 2019 to 2024. Turnover was defined as leaving employment within 1 year after completing RIFDI. We used latent class methods to identify sub‐groups among 134 RIFDI participants based on completion rate of RIFDI curriculum activities, allowing us to classify participants by level of program engagement. Using a propensity weighting approach, we compared predicted probabilities of turnover between each engagement group and 150,045 comparable non‐RIFDI clinicians.

Findings

We identified two sub‐groups among RIFDI participants: lower engagement participants (31%); and higher engagement participants (69%). In adjusted models, turnover was 12.8% among lower engagers, <1.0% among higher engagers, and 11.3% among non‐RIFDI participants. Turnover was approximately 11 times lower for higher engagers compared with non‐RIFDI participants (p‐value < 0.001).

Conclusion

The impact of RIFDI on VHA clinician turnover varied by level of program engagement, with highly engaged participants experiencing significantly lower turnover rates. Health systems should consider programs like RIFDI as a potential strategy for improving rural clinician retention.

1. Background

Of the 20 million Veterans in the United States, nearly one‐quarter (4.4 million) live in rural areas [1]. The Veterans Health Administration (VHA) is uniquely positioned to serve rural Veterans through its large integrated network of medical centers, community‐based outpatient clinics, and telehealth service options [2, 3]. VHA also employs providers trained to treat military‐specific health conditions such as post‐traumatic stress and traumatic brain injury, which are often more prevalent among Veterans [4]. However, more than 60% of healthcare provider shortage areas for primary care, dental, and mental health services are located in rural areas [1, 5], creating a critical gap in care for the millions of rural Veterans who rely on VHA for their health services. Addressing this gap requires not only recruiting new providers to rural settings but also retaining those already in place.

Retention of rural clinicians is influenced by a range of individual and organizational factors, and professional development programs that target clinician educators offer one promising avenue for improvement. Providing career development opportunities has been demonstrated to improve job satisfaction and commitment among clinicians, thereby reducing turnover [6, 7]. Beyond their direct retention benefits, such programs can also strengthen the capacity and quality of rural training sites—improving the learning environment for trainees and, in turn, increasing the likelihood that those trainees choose to practice in rural settings themselves after completing their program [8]. In this way, investment in the existing rural clinical workforce may generate a positive feedback loop, helping rural healthcare facilities address staffing challenges from both ends: retaining experienced clinicians while also attracting the next generation.

The VHA's Rural Interprofessional Faculty Development Initiative (RIFDI) is a structured professional development program designed to realize these benefits, with the overall mission of enhancing teaching and training skills for clinician educators in rural VHA settings [9, 10]. While prior literature indicates engagement in professional development programs reduces turnover of staff in health systems [6], little if any research has examined this relationship in the rural healthcare context. Moreover, to our knowledge, no prior work has quantitatively evaluated the potential association between level of engagement and turnover outcomes. To address this gap in knowledge, we evaluated the impact of RIFDI on VHA clinician turnover by level of program engagement. Findings from our study provide guidance to VHA and other rural healthcare delivery systems as they develop and execute workforce initiatives to better attract and retain clinical staff.

2. Methods

2.1. RIFDI Overview

2.1.1. Program Composition

RIFDI is a collaborative effort between the VHA's Office of Rural Health and Office of Academic Affiliations [9, 10]. The program is delivered over approximately 18 months to a cohort of accepted participants. RIFDI's curriculum comprises six core activities, including: intensive educational conferences; interactive webinars; an online asynchronous clinical education training program; interprofessional peer group meetings; a faculty development workshop for RIFDI participants’ medical sites; and a workplace‐based experiential project (Figure 1) [9]. RIFDI launched in October 2019, and five cohorts have completed the program to date. RIFDI was originally conceived as a blended learning program that combines face‐to‐face and online learning activities. However, due to changes stemming from the COVID‐19 pandemic, RIFDI has primarily been implemented using a virtual format.

FIGURE 1.

Curriculum activities.

2.1.2. Participant Eligibility

To participate in RIFDI, VHA educators must be affiliated with a VHA site that serves a patient population of at least 40% rural patients, have no or very limited graduate medical education capacity (<20 full‐time equivalent (FTE) trainee positions annually), and/or operate a large rural community‐based outpatient clinic [9]. Medical directors at eligible sites can support up to two participants per cohort [9]. Each participant is guaranteed 10% protected clinical release time for the duration of the program. Participants must be VHA‐paid employees with a minimum 0.5 FTE appointment and must complete a participant agreement committing to attend all program activities and curricular components [9].

2.1.3. Data Sources and Study Population

This was a quasi‐experimental study using information collected from RIFDI program records and administrative data from the VHA Corporate Data Warehouse (CDW) from 2019 to 2024. CDW is a centralized database that consolidates and standardizes clinical, financial, and administrative data from various VHA systems [11]. We collected data from clinical educators across 140 VHA medical sites nationwide, including 134 RIFDI participants (from the first three consecutive program cohorts) and 150,045 VHA clinicians who did not participate in RIFDI. We restricted non‐RIFDI clinicians to those affiliated with VHA sites with prior RIFDI participation and with VHA employment dates that span the 2019–2022 period, aligning with completion of the first program cohort and one full year post‐completion.

2.1.4. Key Measures

Our outcome measure was clinician turnover, treated dichotomously. Turnover status was determined based on whether an individual separated from VHA employment within 1 year of their RIFDI completion date. Employment status was verified based on payroll data, with employees classified as separated if they had not received VHA payments for at least 3 months. Employee retirements were included in our turnover definition.

Our key explanatory variable was the level of participant engagement with RIFDI. Engagement levels were based on the proportion of required RIFDI curriculum activities completed by each participant. Per program requirements for graduation, participants needed to: attend two educational conferences; complete 12 online asynchronous education training courses; attend 75% of offered peer group sessions; organize one workshop at a participant site; and receive an experiential project score of two or more (highest possible score of three). Webinar attendance data was not available for all RIFDI cohorts during the study period and was not included in analysis. We included participant‐level completion rates for each of these activities to create a composite latent sub‐group measure capturing patterns of RIFDI engagement. The statistical method used to determine engagement sub‐groups is described in detail below. All RIFDI participants with available program records were included in engagement sub‐groups, regardless of whether they completed the full program curriculum. Participants who did not complete all required activities were retained in the analysis and assigned to latent classes based on their observed patterns of activity completion; only participants who separated from VHA employment due to death prior to the 1‐year follow‐up window were excluded from our sample (n = 1).

Other participant‐level covariates included measures of sex (male; female), age (in years), VHA tenure (in years, by mean quartile grouping), provider type (physicians and advanced practice providers (APPs), including nurse practitioners and physician assistants; nursing staff; mental health providers; all other educator types), and rurality (proportion of total rural Veterans served by affiliated VHA site). These covariates were included to control for potential confounding that might have influenced the clinician turnover outcome.

2.1.5. Analytic Approach

We used latent class analysis (LCA) to identify distinct sub‐groups among RIFDI participants based on their patterns of engagement (i.e., RIFDI curriculum activity completion) and other observable characteristics. LCA uses probabilistic modeling to cluster participants and estimate the probability of each individual belonging to a particular latent class [12, 13]. A key advantage of LCA over traditional clustering approaches is its reliance on rigorous statistical tests to determine clustering criteria, making it more objective than other approaches. We specified multiple LCA models with varying numbers of latent classes and assessed goodness‐of‐fit using the Bayesian Information Criterion (BIC) to determine the model with the optimal number of classes, with lower values indicating a better fit [14]. (Refer to Appendix A for a complete review of specified LCA models and corresponding BIC values.) We adjusted for additional clinician covariates to increase model precision. All LCA modeling was performed using STATA/MP Version 18 [15].

To assess RIFDI's impact on VHA clinician turnover by level of program engagement, we constructed a comparable control group of non‐RIFDI participants. We used inverse probability of treatment weighting (IPTW) to balance covariates across treatment groups and account for potential bias [16]. To generate weights, we first used logistic regression to model the probability (or propensity score) of RIFDI exposure based on observed characteristics of the sample, including VHA tenure and provider type—covariates selected based on their theoretical relevance as predictors of both RIFDI participation and clinician turnover, and for which adequate data were available across the full study sample. Second, weights were calculated as the inverse of the propensity score. Applying these weights to our study sample creates a pseudo‐population where predictors are equally distributed across treatment groups. The IPTW approach provides double robustness against model misspecification by including confounders in both the propensity score and outcome models [17]. We used standardized mean difference (SMD) to assess the balance of covariates between RIFDI latent class sub‐groups and non‐RIFDI participants post‐weighting, with values below 0.10 considered well‐balanced [18]. We then computed the adjusted probability of turnover for latent class sub‐groups and non‐RIFDI participants using predictive margins and tested all pairwise comparisons to identify significant differences between groups using a weighted t‐test approach. To account for multiple comparisons across the three pairwise tests, we applied a Bonferroni correction, adjusting the significant threshold to p < 0.017; findings were consistent under this correction. All IPTW and outcomes modeling were performed using RStudio Version 4.5.0 [19].

3. Results

3.1. Latent Class Identification

Our LCA model identified two sub‐groups of RIFDI participants: participants with lower engagement; and participants with higher engagement (see Figure 2). The lower engagement group included 42 RIFDI participants (31%), with observed activity completion rates ranging from 9% for experimental projects to 71% for site workshops. The higher engagement group included 92 RIFDI participants (69%), with observed activity completion rates ranging from 69% for peer groups to 98% for experiential projects.

FIGURE 2.

Identified sub‐groups of RIFDI program engagement by rate of activity completion.

3.2. Description of Treatment Groups

Before applying IPTW, we observed some significant differences in characteristics across the three treatment groups, including: lower RIFDI engagers, higher RIFDI engagers, and non‐participants (Table 1). Compared with non‐participants, both groups of RIFDI participants were more likely to be physicians/APPs (all RIFDI participants: 50%; non‐participants: 23%) and served a higher average proportion of rural Veterans (all RIFDI participants: 55%; non‐participants: 29%). Compared with lower RIFDI engagers, higher engagers had shorter average VHA tenures (higher engagers: 7.9 years; lower engagers: 10.2 years), were less likely to be physicians/APPs (higher engagers: 49%; lower engagers: 57%), and served a lower average proportion of rural Veterans (higher engagers: 50%; lower engagers: 67%). After applying IPTW, SMDs for all weighting covariates were reduced to below 0.10, indicating adequate balance across treatment groups.

TABLE 1.

Descriptive characteristics by treatment group.

Characteristic	RIFDI: Lower engagement (N = 42)	RIFDI: Higher engagement (N = 92)	Non‐RIFDI (N = 150,045)	Test a	Unadjusted SMD b	Adjusted SMD b
Occupation, n (%)
Mental health provider	6 (14%)	22 (24%)	22,077 (15%)	<0.001	0.0900	0.0211
Nursing staff	6 (14%)	7 (8%)	56,463 (38%)		0.2713	0.0606
Physician/advance practice provider	24 (57%)	45 (49%)	34,962 (23%)		0.3634	0.0371
Other	6 (14%)	18 (20%)	36,285 (24%)		0.1613	0.0994
VHA tenure quartiles, n (%)
Quartile 1 (0–6 years)	17 (41%)	43 (49%)	47,256 (32%)	0.002	0.1748	0.0502
Quartile 2 (7–10 years)	8 (19%)	23 (26%)	33,810 (23%)		0.0576	0.0367
Quartile 3 (11–16 years)	8 (19%)	14 (16%)	33,927 (23%)		0.0737	0.0275
Quartile 4 (17–37 years)	9 (21%)	7 (8%)	35,052 (23%)		0.1459	0.0651
Female, n (%)	25 (60%)	65 (71%)	110,388 (74%)	0.115	NA	NA
Age, mean (SD)	48.9 (9.0)	46.6 (9.5)	49.5 (11.2)	0.062	NA	NA
Rurality, mean (SD)	0.67 (0.35)	0.50 (0.37)	0.30 (0.33)	<0.001	NA	NA
Turnover at 1 year	5 (11.9%)	1 (1.1%)	15,755 (10.5%)	<0.001	NA	NA

^a Pearson's correlation tests used for categorical characteristics (occupation, VHA tenure, and female). Regression tests used for continuous characteristics (age, rurality).

^b SMD: standardized mean difference. SMD values not available for characteristics not included in the propensity score model.

3.3. Outcomes Modeling

Unadjusted turnover was 11.9% among lower engagement participants, 1.1% among higher engagement participants, and 10.5% among non‐participants. In our adjusted model, turnover was 12.8% among lower engagement participants, <1.0% among higher engagement participants, and 11.3% among non‐RIFDI participants. In our pairwise comparison analysis, the probability of turnover was approximately 11 times lower for higher engagers compared with non‐RIFDI participants (point estimate: –0.113; 95% CI: –0.115, –0.111; p‐value < 0.001). We did not detect statistically significant differences comparing turnover probabilities between lower engagers versus non‐RIFDI participants (point estimate: 0.015; 95% CI: –0.134, 0.163; p‐value = 0.845) or higher engagers versus lower engagers (point estimate: –0.128; 95% CI: –0.277, 0.020; p‐value = 0.091).

4. Discussion

This quasi‐experimental study examined the relationship between level of engagement in VHA's RIFDI program and clinician turnover. Our findings contribute important evidence to support the strategic use of clinical educator development programs like RIFDI as a retention tool among the rural healthcare workforce, which may be particularly relevant given VHA's unique position as a major rural healthcare provider. We found that participation in RIFDI is associated with reductions in VHA clinician turnover, with highly engaged participants experiencing significantly lower turnover rates 1 year after completing the program compared with non‐participants. Our research provides empirical support for investing in structured professional development initiatives to address the persistent staffing challenges that threaten healthcare access for rural Veterans and communities more broadly.

Our findings align with existing literature demonstrating the staff retention benefits of professional development programs in healthcare settings. Previous research shows that employees who participate in career development initiatives are more likely to stay in their current jobs than those who do not [6, 20]. Our work extends this evidence base to a rural healthcare context, where research in this area is more sparse and clinical workforce shortages are particularly acute [1]. This research also contributes to the wider literature by quantifying how different levels of engagement in a professional development program affect clinician turnover. Our findings are consistent with a pattern in which greater engagement correlates with lower turnover—a pattern that aligns with theoretical models suggesting that meaningful professional development requires sustained engagement [21, 22]. However, to our knowledge, no prior work has quantitatively evaluated this relationship in the rural healthcare context. We found that higher levels of engagement in core RIFDI activities were associated with lower rates of turnover. Importantly, while turnover was significantly lower among high engagers relative to non‐participants, the comparison between higher and lower engagement groups did not reach statistical significance. This result could reflect the relatively small RIFDI sample and limited statistical power, rather than a true absence of effect. In other words, the observed pattern of results is consistent with a graduated relationship between engagement and retention, but the study was likely underpowered to confirm this statistically.

An alternative interpretation of our findings warrants consideration. Rather than reflecting a causal effect of program engagement on retention, the observed pattern may partly reflect pre‐existing differences between clinicians who engaged deeply with RIFDI and those who did not: an adherence‐type explanation in which higher engagers were already at lower risk of turnover before the program began. Factors such as intrinsic motivation, career commitment, or institutional stability may simultaneously predict both sustained program engagement and a lower propensity to leave. This phenomenon may translate into an association that is driven partly by selection into engagement rather than engagement itself. Nevertheless, this alternative explanation does not diminish the practical relevance of our findings. Whether the retention benefit reflects RIFDI's direct effect or the characteristics of those most able to engage fully, the results underscore the value of identifying and supporting clinicians who can participate in sustained professional development—and of addressing the structural barriers that may prevent others from doing so.

Given these findings, professional development programs like RIFDI should focus future efforts on identifying and addressing factors that shape levels of program engagement. Importantly, engagement should not be understood solely as a program design variable to be optimized—it is also shaped by structural and contextual conditions within participating facilities, including staffing pressures, leadership support, and whether clinical release time is genuinely protected in practice. Future research should examine the conditions that enable or constrain engagement, particularly in small and often under‐resourced rural settings. Prior research exploring how professional development influences employee retention offers relevant insights. Existing healthcare literature identifies job satisfaction and organizational commitment as key factors mediating the relationship between professional development and intention to leave employment [6, 23, 24]. Sense of self‐efficacy and moral distress are also recognized as important mechanisms [25, 26, 27], particularly regarding clinicians’ perceptions of their ability to make improvements in their work environment [26]. By understanding and addressing these factors, VHA and other health systems can design more targeted strategies for their clinical educator programs that enhance participant engagement, potentially leading to reduced clinician turnover in resource‐constrained rural healthcare delivery systems.

The reduction in turnover observed among highly engaged RIFDI participants has important implications for rural healthcare delivery and workforce planning. Rural America faces critical healthcare workforce shortages, as healthcare facilities in these geographic areas often struggle to attract and retain staff [28]. While much attention has been focused on recruiting new providers to rural areas, programs like RIFDI suggest the potential value of preserving and investing in the existing rural healthcare workforce. Retaining experienced clinicians who are already familiar with rural practice challenges, community needs, and local healthcare delivery systems may be more cost‐effective and immediately impactful than continuous recruitment efforts. Furthermore, by enhancing the teaching and mentoring capabilities of current rural clinicians through professional development, programs like RIFDI can create a more supportive learning environment that not only retains existing staff but also improves the training experience for new providers, potentially increasing the likelihood that trainees will choose to remain in rural practice settings after completion of their programs. However, as our study findings demonstrate, simply offering such programs is insufficient; their efficacy depends on ensuring that participants actively and consistently engage in core program activities.

Further, while it is possible that retaining experienced clinicians may offer cost advantages relative to continuous recruitment efforts, this has not been formally evaluated in the context of programs like RIFDI. Future research should consider the direct and indirect costs of program implementation—including administrative overhead, protected clinical release time, and potential variation in costs across cohorts and delivery formats—alongside retention outcomes, to support rigorous cost‐effectiveness analyses that can inform decisions about adopting and scaling such programs.

5. Limitations

Several limitations warrant consideration when interpreting our study results. First, despite using statistical techniques to account for many observable confounders, there is still potential for bias. It is possible unobserved factors that influence a clinician's decision to both participate in RIFDI and engage deeply with the program also predict the likelihood of turnover (i.e., confounding by indication) that are not accounted for in our IPTW approach. While post‐weighting SMDs were all below 0.01, indicating adequate balance on observable dimensions, future research with larger cohorts accommodating a wider set of covariates is needed to strength causal inference. Second, our RIFDI sub‐group classification was based on program activity completion rates, which may not fully capture the quality or depth of participant engagement. However, our analytic dataset provides the most robust measure of RIFDI engagement currently available. Third, we defined turnover as leaving employment within 1 year of program completion, which may not capture all relevant retention outcomes or longer‐term effects. Follow‐up studies examining longer‐term or continuous measures of turnover will be needed to address this gap as more clinicians complete RIFDI. Fourth, our turnover measure reflects separation from the VHA system overall and does not fully capture rural workforce retention. That said, because RIFDI eligibility requires facilities to serve a patient population of at least 40% rural Veterans, the program is by design concentrated in rural‐serving sites, and system‐wide separation from these facilities does reflect, at least partially, retention within the rural healthcare workforce. In addition, sample size constrains facility‐level or rural‐specific retention analysis in the current study; future work with larger cohorts should examine this distinction. Lastly, our study was conducted in the VHA health system, which has unique institutional characteristics that may limit generalization to other practice settings.

6. Conclusion

This study provides quantitative evidence that structured educator development programs can significantly impact clinician retention in rural VHA healthcare settings, with effects moderated by level of participant engagement. Our findings suggest that well‐designed professional development initiatives may provide an important tool for addressing rural healthcare workforce challenges. Health systems considering retention strategies should prioritize programs that promote sustained engagement and meaningful professional development opportunities for their clinical staff.

Funding

This project was funded by the VA Office of Rural Health Enterprise Wide Initiative.

Conflicts of Interest

The authors declare no conflict of interest. The conclusions are those of the author and do not necessarily represent the views of the VA.

Appendix A. Latent class analysis model fit statistics

A.1.

Classes	AIC	BIC
2	231.583	323.097
3	308.656	365.853
4	327.966	490.960

AIC: Akaike Information Criterion; BIC: Bayesian Information Criterion.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.