Expansion of the Veterans Health Administration Network and Surgical Outcomes (2024)

Key Points

Question Did expanded access to health care across previously separate health care systems—the Veterans Health Administration (VHA) and non-VHA health care community—affect surgical outcomes?

Findings In this nonrandomized regression discontinuity study of 615 473 unique surgical procedures among 498 427 patients, expanded access to care was associated with a greater proportion of surgical procedures in a community setting, and this proportion varied by procedure type. However, no difference in postoperative emergency department visits, inpatient readmissions, or mortality was found between VHA-provided and VHA-paid procedures in the community setting.

Meaning Expanding access to health care outside of the VHA was associated with a shift in the location of surgical procedures among veterans but had no association with postoperative outcomes; these findings may assuage concerns of worsened patient outcomes resulting from care coordination issues when care is expanded outside of a single health care system.

Importance The US Department of Veterans Affairs (VA) Veterans Choice Program (VCP) expanded health care access to community settings outside the VA for eligible patients. Little is known about the effect of VCP on access to surgery and postoperative outcomes. Since its initiation, care coordination issues, which are often associated with adverse postoperative outcomes, have been reported. Research findings on the association of VCP and postoperative outcomes are limited to only a few select procedures and have been mixed, potentially due to bias from unmeasured confounding.

Objective To investigate the association of the VCP with access to surgery and postoperative outcomes using a nonrandomized controlled regression discontinuity design (RDD) to reduce the impact of unmeasured confounders.

Design, Setting, and Participants This was a nonrandomized RDD study of the Veterans Health Administration (VHA). Participants included veterans enrolled in the VHA who required surgery between October 1, 2014, and June 1, 2019.

Interventions The VCP, which expanded access to VA-paid community care for eligible veterans living 40 miles or more from their closest VA hospital.

Main Outcomes and Measures Postoperative emergency department visits, inpatient readmissions, and mortality at 30 and 90 days.

Results A total of 615 473 unique surgical procedures among 498 427 patients (mean [SD] age, 63.0 [12.9] years; 450 366 male [90.4%]) were identified. Overall, 94 783 procedures (15.4%) were paid by the VHA, and the proportion of VHA-paid procedures varied by procedure type. Patients who underwent VA-paid procedures were more likely to be women (9209 [12.7%] vs men, 38 771 [9.1%]), White race (VA paid, 54 544 [74.4%] vs VA provided, 310 077 [73.0%]), and younger than 65 years (VA paid, 36 054 [49.1%] vs 229 411 [46.0%] VA provided), with a significantly lower comorbidity burden (mean [SD], 1.8 [2.2] vs 2.6 [2.7]). The nonrandomized RDD revealed that VCP was associated with a slight increase of 0.03 in the proportion of VA-paid surgical procedures among eligible veterans (95% CI, 0.01-0.05; P = .01). However, there was no difference in postoperative mortality, readmissions, or emergency department visits.

Conclusions and Relevance Expanded access to health care in the VHA was associated with a shift in the performance of surgical procedures in the private sector but had no measurable association with surgical outcomes. These findings may assuage concerns of worsened patient outcomes resulting from care coordination issues when care is expanded outside of a single health care system, although it remains unclear whether these additional procedures were appropriate or improved patient outcomes.

The US Department of Veterans Affairs (VA) Veterans Choice Program (VCP) is a congressionally mandated program that is intended to improve access to care for VA patients.¹ Its introduction provides an ideal natural experiment to examine the outcomes of expanding health care access across separate health care systems. Under the VCP, the VA pays for community health care services (ie, care received outside of the VA) for veteran applicants who meet at least 1 of the following eligibility criteria: (1) driving distance of 40 miles or more to the closest VA with primary care capabilities, (2) appointment wait time greater than 30 days, or (3) excessive geographic, environmental, or medical comorbidity burden. In 2014 alone, 1.3 million veterans met these eligibility criteria to receive community care through the VCP, and the use of community care services has increased drastically since that time. In 2021, 20% of the VA’s budget was spent on VA-paid care outside of the VA, accounting for $17.6 billion.¹

Soon after the VCP was initiated, however, health care professionals began to voice concerns about care coordination and continuity issues between VA-provided and VA-paid community care through the VCP, citing difficulties communicating with specialists among other issues.² The issues raised are important because care coordination is a key factor in delivering high-quality care.³ Fragmentation of care across care transitions, such as seen between VA and community health care professionals, can lead to higher costs, increased readmissions, and greater mortality.^4-6 Unfortunately, care fragmentation concerns due to the introduction of VCP have been noted by both patients and clinicians in several qualitative studies.^7-9 Reports of longer wait times among veterans receiving care outside the VA further support concerns about communication issues between the 2 typically separate health care systems.^10,11 Specialty services, such as surgery, may be particularly affected, which could lead to increased costs, redundant or inadequate health care utilization, and greater mortality. However, little is known about the association of VCP with access to surgery and postoperative outcomes.

Thus, the primary objective of this study was to describe the association of VCP with access to and outcomes of VA-paid care in a surgical population. It is well documented that veterans who use VA-paid care are healthier than those who receive care through the VA, complicating observational study designs.^12-14 Because of this difference, all previous studies of VCP carry the limitation of residual confounding. To minimize this limitation, we used a regression discontinuity design (RDD), a quasi-experimental study design that can reduce the bias introduced by unmeasured confounding.¹⁵ We hypothesized that the potential benefits of improved health care access provided through the VCP may be dampened by worsened postoperative outcomes due to increased care fragmentation.

We used a retrospective quasi-experimental study design to explore the health care utilization of veterans undergoing either VA-provided or VA-paid surgery (ie, community care) between October 1, 2014, to June 1, 2019, when VCP ended. The study period was chosen to align with the time that VCP was active. Patients were excluded from the study if they did not undergo an operative procedure as defined by the Centers for Disease Control and Prevention National Healthcare Safety Network (NHSN). This study was reviewed and approved by the Stanford institutional review board with a waiver of written informed consent owing to the use of deidentified patient data. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines.

Study data were obtained from 3 sources: (1) the VA Patient Treatment File (PTF, VA provided), (2) the VA Fee Basis data (VA paid), and (3) the VA Program Integrity Tool (PIT, VA paid). The VA PTF is the cleaned version of the VA electronic health record available for research and operations use. It contains all instances of VA-provided care. There are 2 sources for non-VA and VA-paid care. The first is Fee Basis data, which contain fee data for VA-purchased care. It is the primary source of non-VA claims data for episodes of care before October 1, 2015. In 2013, the VA introduced the PIT. The PIT data contain Office of Community Care claims data. Starting October 1, 2015, PIT became the primary source for processing VCP claims.

The NHSN definition of operative procedures was used to identify and categorize surgical procedures coded in these 3 data sources.¹⁶ This algorithm categorizes surgical procedures into 40 groups that are used for nationwide surveillance of hospital-acquired infections. Given the wide variety of surgical procedures, we used the NHSN definition to identify a grouping of commonly performed procedures that are consistently viewed as operative procedures.

The intervention of interest was eligibility for the VCP based on the VCP’s distance criteria. Patients and health care professionals were not blinded to their treatment status. Patients were eligible for VCP if they lived 40 or more miles from the closest VA facility with primary care capabilities at the time of surgery. Geographical data in the VA are maintained by the VA Geospatial Service Support Center, formally known as the Planning Systems Support Group (PSSG). The PSSG calculates this distance as the straight-line distance, in miles, from the patient’s zip code to the closest VA facility with primary care capabilities.

Postdischarge outcomes included 30-day emergency department (ED) visits, 30- and 90-day all-cause readmission, and 30- and 90-day mortality. These outcomes have been shown previously to be associated with care fragmentation in surgical populations.^4,17-19 Exploring surgery-related health care utilization was not possible due to the lack of surgery-specific data in the VA-paid data resources.

Postoperative mortality was obtained from the VA Vital Status file. The VA queries multiple mortality sources, including the Veterans Benefits Administration and Centers for Medicaid and Medicare, making it the most comprehensive resource for mortality assessment. All-cause readmissions were identified as any inpatient admission documented in any of the 3 data sources. ED visits were identified in the VA Patient Treatment File by clinic stop code, a unique VA identifier for the type of VA outpatient visit (clinic stop = 130 or 131). ED visits were also obtained from VA-purchased care; fee basis data were queried for Current Procedure Terminology (CPT) codes consistent with an ED-related visit (99281-99285), and PIT data were queried for claims that were coded with a place of service equal to “emergency department.”

The International Classification of Diseases (ICD), Ninth and Tenth Revision and CPT codes were queried from all data sources to identify the date and type of surgical procedure.¹⁶ Procedures were defined as VA provided if they were obtained from the VA Patient Treatment File or VA paid if they were identified in the VA Fee Basis data or the VA PIT data. Patient demographics (age, sex, race) were obtained from the VA Patient Treatment File. Patients from the following races were included: Black, White, other (ie, patients identifying as Alaska Native, American Indian, Asian, Native Hawaiian, Other Pacific Islander, or other), and missing. As a measure of comorbidity burden, the Charlson Comorbidity Index (CCI) was calculated by querying ICD diagnosis codes recorded in the VA Patient Treatment File in the 1 year before the date of surgery.²⁰ VA Priority Grouping at the time of surgery was also obtained from the VA Patient Treatment File. The VA Priority Group is an 8-category grouping calculated by the Veterans Benefits Administration to determine VA benefit eligibility. Factors in the determination of VA Priority Group include the percentage of service-connected disability, receipt of other benefits such as Medicaid, service period, and income level.²¹

The unit of analysis was the surgical procedure. All variables’ distributional characteristics and missingness were explored before beginning bivariate analyses. The characteristics of patients undergoing VA-paid compared with VA-provided surgery were examined using χ² tests and t tests. We then examined the assumptions of RDD. RDD is a quasi-experimental design that examines the outcome of an intervention, such as VCP, across an arbitrary eligibility threshold (ie, one that is difficult for individuals to manipulate).¹⁵ In this case, the VCP defines a threshold of 40 miles to the nearest VA facility with primary care capabilities as an eligibility criterion. Thus, we assumed that patients living 39 miles from a VA facility were like patients living 40 miles from a VA facility for all factors except for VCP eligibility. Although the number of procedures decreased as miles from a VA facility increased, we noted no differences in patient characteristics by mile among patients living 34 to 44 miles from a facility. Thus, the magnitude of the discontinuity at the threshold (ie, 40 miles) can be used to estimate the causal effect of the intervention (ie, VCP). We used the RDD to examine the association of VCP eligibility with VA-paid surgery use, postdischarge ED visits, inpatient admissions (all-cause readmissions), and postoperative mortality.

Given that there are other eligibility criteria for the VCP, we conducted a fuzzy RDD. First, we plotted the probability of each outcome by distance to a VA facility for the subset of patients who lived 20 to 60 miles from a VA facility. We fitted separate regression lines for estimates less than 40 miles and 40 miles or more and visually examined the plot for discontinuity at mile 40. Next, we used local polynomial regression to determine the mean square error optimal bandwidth (ie, miles) to estimate an intervention effect. In RDD, not all available data are used to determine the final effect estimate. The optimal bandwidth is based on a tradeoff between bias and variance. For our study, the optimal bandwidth was determined to be 5 miles meaning that the final RDD estimate is based on individuals living 35 to 39 miles and 40 to 44 miles. We used robust bias-corrected CIs to determine the statistical significance of the effect. An α level of .05 was considered to be statistically significant, and all P values were 2-sided. Only patients living 20 to 60 miles from a VA facility were included in the RDD analyses. We did not find that other bandwidths affected our results. These steps were repeated for all outcomes. All analyses were conducted in SAS Enterprise Guide, version 8.1 (SAS Institute) and R (R Project for Statistical Computing) using the rdrobust²² and ggplot2²³ packages.

We identified 615 473 unique surgical procedures among 498 427 patients (mean [SD] age, 63.0 [12.9] years; 450 366 male [90.4%]; 48 061 female [9.6%]). Patients from the following races were included: 80 660 Black (16.2%), 364 621 White (73.2%), 9852 other (2.0%), and 43 294 missing (8.7%). Approximately 9% of patients were missing an assessment of race, but all other variables were nearly complete (>99%) (Table 1). The most common NHSN surgical procedure types were herniorrhaphy (76 005 [12.3%]), knee prosthesis (74 734 [12.1%]), and pacemaker surgery (48 801 [7.9%]) (Table 2). Overall, 94 881 procedures (15.4%) were VA paid, and the proportion of VA-paid procedures varied by procedure type (Table 2). Procedures involving spinal fusion or knee prosthesis had higher proportions of VA-paid care (13 332 [33.8%] and 20 992 [28.1%], respectively). In comparison, procedures involving appendix surgery or limb amputation had lower proportions of VA-provided care (451 [4.0%] and 1841 [6.2%], respectively) (Table 2). Patients who underwent VA-paid procedures were more likely to be female (9209 [12.7%] vs men, 38 771 [9.1%]), White race (VA paid, 54 544 [74.3%] vs VA provided, 310 077 [73.0%]), and younger than 65 years (VA paid, 36 054 [49.1%] vs 195 572 [46.0%] VA provided). Patients who underwent VA-paid procedures also had a significantly lower comorbidity burden (mean [SD] CCI, 1.8 [2.2] vs 2.6 [2.7]) and were more likely to be VA priority group 1 or 2, which receives the most VA benefits (VA paid, 38 164 [52.0%] vs 188 091 [44.3%]) (Table 1).

Postoperatively, 109 186 patients (21.9%) experienced readmission within 90 days, with more readmissions after VA-paid surgery (VA paid, 17 544 [23.9%] vs VA provided, 91 642 [21.6%]); 97 737 patients (19.6%) experienced an ED visit within 30 days also with fewer ED visits after VA-paid surgery (9052 [12.3%] vs 88 685 [20.9%]). Overall, 13 422 patients (2.7%) experienced mortality within 90 days of surgery. Patients who underwent VA-paid surgery had significantly lower 90-day mortality than those with VA-provided surgery (VA paid, 902 [1.2%] vs VA provided 12 520 [3.0%]) (Table 1). Readmission, ED visits, and mortality varied significantly across procedure types (Table 2).

The distribution of miles to the closest VA hospital with primary care capabilities was highly skewed (Figure 1). The RDD analyses identified a significant increase in the proportion of VA-paid surgery among patients who lived above the 40-mile threshold (0.03; 95% CI, 0.01-0.05; P = .01) (Figure 2) compared with patients just below the threshold. However, no association with ED visits, readmissions, or mortality was observed (Figure 3). The average treatment association of VCP eligibility with the proportion of VA-paid procedures was an increase of 3 percentage points (0.03; 95% CI, 0.01-0.05; P = .01).

The objective of this nonrandomized regression discontinuity study was to investigate the association of expanded availability of community care through the VCP with access to surgical care and postoperative outcomes. Consistent with prior research, we found that patients who underwent VA-paid surgery (ie, community care) were very different from patients who underwent VA-provided surgery.^12-14 Patients who underwent VA-paid procedures were significantly more likely to be female, younger, and White, with a significantly lower comorbidity burden. Without any adjustments for these differences, patients undergoing VA-paid surgery had significantly higher readmission and ED rates but lower mortality rates. With RDD, we found a significant increase in the number of VA-paid procedures among patients who were eligible for VCP, but there was no association with postoperative ED visits, readmissions, or mortality. Our results suggest that, although the VCP may have influenced the allocation of surgical procedures performed in the VA by shifting more surgical procedures to a community setting, it does not appear to have resulted in adverse postoperative outcomes associated with care fragmentation.

To our knowledge, very few studies have examined the association of the VCP with surgical procedures, and none have used a quasi-experimental design to examine causality.^24,25 The few extant studies have focused on a small number of procedures and have shown mixed results. An observational cohort study of 90-day complication rates in outpatient cataract procedures found no difference in outcomes.²⁴ However, a separate cohort study examining elective total knee arthroplasties found fewer readmissions among VA-provided procedures.²⁵ The differences between these cohorts has limited comparable patient samples for observational studies. Inconsistent findings among the few studies that have examined differences in a surgical population could be due to a high potential for residual confounding.

Our findings have interesting implications for VA care. First, the location of surgical care was shifted by the VCP. Despite reports of care coordination issues with VA-paid care,^7-9 this shift to more surgical procedures performed outside of the VA was not associated with adverse short-term postoperative outcomes (30 and 90 days). Second, the findings are most relevant in situations where the VA can provide the surgical procedure or buy it in the community. There are some procedures, such as transplant, lithotripsy, gastric bypass, or transcatheter aortic valve replacement, where the VA provides these procedures in select locations only depending on local capabilities. Our results are less applicable to these select procedures, and VA should continue to make these decisions on a case-by-case basis.

A significant strength of our study was the use of RDD, a quasi-experimental study design. Although RDD is still an observational design in which treatment groups are not manipulated, it provides higher-quality evidence than the more commonly used cohort and case-control study designs.¹⁵ RDD leverages the natural experiment that is created when an intervention, such as the VCP, is applied to populations based on a well-defined, arbitrary eligibility threshold. In the case of our analyses, we assumed that patients living 34 to 39 miles from a VA facility were not different from patients living 40 to 45 miles from a VA facility (eFigure in the Supplement). The only difference is that patients in the latter group are eligible to receive care outside of the VA. The RDD includes several assumptions that must be acknowledged. First, it is assumed that there is discontinuity in the probability of receiving the intervention at the threshold. In this case, although there are other factors that lead to VCP eligibility, there is certainly a subset of individuals that will be eligible for VCP on distance alone. Second, we must assume that an individual’s distance to a VA facility cannot be manipulated in order to gain or lose access to the intervention. Third, we must assume that the groups around the threshold are similar. Or in other words, there is no difference in patients who live 39 vs 40 miles from a VA facility that would affect the outcome of interest but their eligibility for VCP. Lastly, the outcome is continuous across the threshold.

As with all studies, this study is not void of limitations. First and foremost, this study used a veteran population that is mainly older and male. All interpretations of these results should be made with this limitation in mind. Second, due to heterogeneity in coding across data sources, we were unable to look at surgery-specific outcomes and thus selected 3 broad outcomes of readmissions, ED visits, and mortality. Research has suggested that approximately 70% of readmissions within 30 days of an operation are surgery related,²⁶ and we have no reason to expect differences in the proportion of surgery-related readmissions by the location of the surgery. In addition, there is wide variation in the definition of surgery for studies examining all surgical procedures. We opted to use the NHSN definition of an operative procedure, which excludes lower complexity procedures such as cataract surgery. Lastly, although this was a quasi-experimental study, it is still an observational study design. Although the 3 data sources included in this analysis likely represent most of the health care utilization in this population, we still cannot account for other sources of health care, such as private insurance.

To conclude, in this nonrandomized regression discontinuity study, the VCP was associated with an increase in the proportion of VA-paid surgical procedures occurring outside of the VA, but we found no difference in postoperative mortality, readmissions, or ED visits. In 2018, the VA continued to expand access to community care with the MISSION Act. Further, in October 2021, the VA announced the goal of centralizing the financial aspects of VA-paid care.²⁷ Our study results support these transitions and a continued path toward a seamless model of health care within the VA while also emphasizing the importance of access to community care in the veteran population and assuaging concerns of worsened outcomes due to care fragmentation.

Accepted for Publication: July 16, 2022.

Published Online: October 12, 2022. doi:10.1001/jamasurg.2022.4978

Correction: This article was corrected on December 14, 2022, to add a missing middle initial to an author name in the byline.

Corresponding Author: Laura A. Graham, PhD, MPH, Health Economics Resource Center, Veterans Affairs Palo Alto Health Care System, 795 Willow Rd, Bldg 324, Room D146, Menlo Park, CA 94105 (laura.graham@va.gov).

Author Contributions: Dr Graham 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: Graham, Rose, Aouad, Wagner.

Acquisition, analysis, or interpretation of data: Graham, Schoemaker, Rose, Morris, Aouad.

Drafting of the manuscript: Graham.

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

Statistical analysis: Graham, Rose, Aouad, Wagner.

Obtained funding: Rose, Aouad, Wagner.

Administrative, technical, or material support: Graham, Schoemaker, Rose, Morris, Wagner.

Supervision: Graham, Rose, Aouad, Wagner.

Conflict of Interest Disclosures: Drs Graham, Rose, and Wagner reported receiving grants from the US Department of Veterans Affairs during the conduct of the study. Dr Wagner reported receiving grants from the National Institutes of Health and Robert Wood Johnson Foundation, and having an Agency for Healthcare Research and Quality, InterAgency Agreement, outside the submitted work. No other disclosures were reported.

Funding/Support: This work was supported by Health Services Research and Development grant I01HX003106-01 from the US Department of Veterans Affairs, Veterans Health Administration, Office of Research and Development.

Role of the Funder/Sponsor: The funder 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.

Disclaimer: The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the US Department of Veterans Affairs or the US government.