Zaccheus J. Ahonle, PhD, CRC

The Veterans Health Administration (VHA) is the largest US integrated health system, providing care to veterans through VHA and non-VHA practitioners and facilities.^1,2 Providing high-quality, timely, and veteran-centric care remains a priority for the VHA. Legislative efforts have expanded opportunities for eligible veterans to receive care in the community purchased by VHA, known as community care (CC).¹ The Veterans Access, Choice, and Accountability Act of 2014 came in response to reports of long wait times and drive times for patients.^3-5 The MISSION Act of 2018 expanded access to CC by streamlining it and broadening eligibility criteria, especially for veterans in rural communities who often experience more barriers in accessing care than veterans living in urban communities.^1,6-10 Since the implementation of the Choice and MISSION Acts, > 2.7 million veterans have received care through community practitioners within the VHA CC network.¹¹

Background

Increased access to CC could benefit veterans living in rural communities by increasing care options and circumventing challenges to accessing VHA care (ie, geographic, transportation, and distance barriers, practitioner and specialist shortages, and hospital closures). ^5,9,10,12,13 However, health care system deficits in rural areas could also limit CC effectiveness for veterans living in those communities. ³ Other challenges posed by using CC include care coordination, information sharing, care continuity, delayed payments to CC practitioners, and mixed findings regarding CC quality.^5,8,13,14 VHA practitioners are specifically trained to meet the multifaceted needs unique to veterans’ health and subculture, training CC practitioners may not receive.^5,15

CC offers services for primary care and a broad range of specialties, including rehabilitation services such as physical therapy (PT).⁶ PT is used for the effective treatment of various conditions veterans experience and promote wellbeing and independence.¹⁶ US Department of Veterans Affairs (VA) databases reveal a high prevalence of veterans receiving PT services through CC; PT is one of the most frequently used CC outpatient specialty services by veterans living in rural communities.^14,17

Telerehabitltation Enterprisewide Initiative

VHA has greatly invested in delivering care virtually, especially for veterans living in rural communities.¹⁸ In 2017, the VHA Office of Rural Health funded the Telerehabilitation Enterprise-Wide Initiative (TR-EWI) in partnership with the Physical Medicine and Rehabilitation Services national program office to increase access to specialized rehabilitation services for veterans living in rural communities by leveraging telehealth technologies.^18-21 This alternative mode of health care delivery allows clinicians to overcome access barriers by delivering rehabilitation therapies directly to veterans' homes or nearby community-based outpatient clinics. TR-EWI was conceived as a hub-and-spoke model, where rehabilitation expertise at the hub was virtually delivered to spoke sites that did not have in-house expertise. In subsequent years, the TR-EWI also evolved to provide targeted telerehabilitation programs within rural-serving community-based outpatient clinics, including PT as a predominant service.^19,20

As TR-EWI progressed—and in conjunction with the uptake of telehealth across VHA during the COVID-19 pandemic—there has been increased focus on PT telerehabilitation, especially for the 4.6 million veterans in rural communities.^18,22,23 Because health care delivery system deficits in rural areas could limit the effective use of CC, many TR-EWI sites hope to reduce their CC referrals by providing telehealth PT services to veterans who might otherwise need to be referred to CC. This strategy aligns with VHA goals of providing high-quality and timely care. To better understand opportunities for programs like TR-EWI to provide rehabilitation services for veterans and reduce care sent to the community, research that examines CC referral trends for PT over time is warranted.

This study examines CC from a rehabilitation perspective with a focus on CC referral trends for PT, specifically for Veterans Integrated Service Networks (VISNs) where TREWI sites are located. The study’s objectives were to describe rehabilitation PT services being referred to CC and examine associated CC costs for PT services. Two research questions guided the study. First, what are the utilization trends for CC PT referrals from fiscal year (FY) 2019 to FY 2022? Secondly, what is the cost breakdown of CC for PT referrals from FY 2020 to FY 2022?

Methods

This study was conducted by a multidisciplinary team comprised of public health, disability, rehabilitation counseling, and PT professionals. It was deemed a quality improvement project under VA guidance and followed the SQUIRE guidelines for quality improvement reporting.^24,25 The study used the VA Common Operating Platform (Palantir) to obtain individual-level CC referral data from the HealthShare Referral Manager (HSRM) database and consult data from the Computerized Patient Record System. Palantir is used to store and integrate VA data derived from the VA Corporate Data Warehouse and VHA Support Service Center. Referrals are authorizations for care to be delivered by a CC practitioner.

TR-EWI is comprised of 7 sites: VISN 2, VISN 4, VISN 8, VISN 12, VISN 15, VISN 19, and VISN 22. Each site provides telerehabilitation services with an emphasis on reaching veterans living in rural communities. We joined the referrals and consults cubes in Palantir to extract PT referrals for FY 2019 to FY 2022 for the 7 VISNs with TR-EWI sites and obtain referral-specific information and demographic characteristics. ²⁶ Data were extracted in October 2022.

The VHA Community Care Referral Dashboard (CC Dashboard) provided nonindividual level CC cost data.²⁷ The CC Dashboard provides insights into the costs of CC services for VHA enrollees by category of care, standardized episode of care, and eligibility. Data are based on nationallevel HSRM referrals that are not suspended or linked to a canceled or discontinued consult. Data were aggregated by VISN. The dashboard only includes referrals dating back to FY 2020; therefore, PT data from FY 2020 through FY 2022 for VISNs with TR-EWI sites were collected. Data were extracted in December 2022.

This study examined CC referrals, station name, eligibility types, clinical diagnoses (International Classification of Diseases, Tenth Revision codes), and demographic information in the Palantir dataset. Six eligibility criteria can qualify a veteran to receive CC.²⁸ Within clinical diagnoses, the variable of interest was the provisional diagnosis. Patient demographics included age, gender, and rurality of residence, as determined by the Rural-Urban Commuting Area system.^29,30 Rural and highly rural categories were combined for analysis. For the CC cost dataset, this study examined CC referrals, referral cost, and eligibility type.

Analysis

For the first research question, we examined referral data from FY 2019 to FY 2022 using the Palantir dataset, performed descriptive statistical analysis for all variables, and analyzed data to identify trends. Descriptive statistics were completed using IBM SPSS Statistics for Windows Version 29.0.0.0.

A qualitative analysis of provisional diagnosis data revealed what is being referred to CC for PT. A preliminary overview of provisional diagnosis data was conducted to familiarize coders with the data. We developed a coding framework to categorize diagnoses based on anatomical location, body structure, and clinical areas of interest. Data were reviewed individually and grouped into categories within the coding framework before meeting as a team to achieve group consensus on categorization. We then totaled the frequency of occurrence for provisional diagnoses within each category. Qualitative analyses were completed using Microsoft Excel.

For the second research question, the study used the CC cost dataset to examine the cost breakdown of CC PT referrals from FY 2020 to FY 2022. We calculated the number and cost of PT referrals across eligibility groups for each FY and VISN. Data were analyzed using SPSS to identify cost trends.

Results

There were 344,406 referrals to CC for PT from FY 2019 to FY 2022 for the 7 VISNs analyzed (Table 1). Of these, 22.5% were from FY 2019, 19.1% from FY 2020, 28.2% from FY 2021, and 30.3% from FY 2022. VISN 8 and VISN 22 reported the most overall PT referrals, with VISN 8 comprising 22.2% and VISN 22 comprising 18.1% of all referrals. VISN 2 reported the least overall referrals (3.7%). VISN 4 and VISN 12 had decreases in referrals over time. VISN 2 and VISN 15 had decreases in referrals from FY 2019 to FY 2021 and slight increases from FY 2021 to FY 2022. VISN 19 and VISN 22 both saw slight increases from FY 2019 to FY 2020 and substantial increases from FY 2020 to FY 2022, with FY 2022 accounting for 40.0% and 42.3% of all referrals for VISN 19 and VISN 20, respectively (Figure 1).

For FY 2019 and FY 2020, VISN 8 had the highest percentage of referrals (26.7% and 23.2%, respectively), whereas VISN 22 was among the lowest (7.3% and 11.4%, respectively). However, for FY 2021 and FY 2022, VISN 22 reported the highest percentage of referrals (23.5% and 25.3%, respectively) compared to all other VISNs. VISN 2 consistently reported the lowest percentage of referrals across all years.

There were 56 stations analyzed across the 7 VISNs (Appendix 1). Nine stations each accounted for ≥ 3.0% of the total PT referrals and only 2 stations accounted for > 5.0% of referrals. Orlando, Florida (6.0%), Philadelphia, Pennsylvania (5.2%), Tampa, Florida (4.9%), Aurora, Colorado (4.9%), and Gainesville, Florida (4.4%) reported the top 5 highest referrals, with 3 being from VISN 8 (Orlando, Tampa, Gainesville). Stations with the lowest reported referrals were all in VISN 2 in New York: The Bronx, (0%), New York Harbor (0%), Hudson Valley (0.1%) and Finger Lakes (0.2%).

Rurality

Urban stations comprised 56.2% and rural stations comprised 39.8% of PT CC referrals, while 0.2% of referrals were from insular isle US territories: Guam, American Samoa, Northern Marianas, and the Virgin Islands. The sample had missing or unknown data for 3.8% of referrals. FY 2022 had the largest difference in rural and urban referrals. Additionally, there was an overall trend of more referrals over time for rural and urban, with a large increase in rural (+40.0%) and urban (+62.7%) referrals from FY 2020 to FY 2021 and a modest increase from FY 2021 to FY 2022 (+5.2% for rural and +9.1% for urban). There was a decrease in rural (-7.0%) and urban (-3.5%) referrals from FY 2019 to FY 2020 (Figure 2).

There were differences in referrals by rurality and VISN (Table 2). VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals, whereas VISN 4, VISN 8, and VISN 22 reported more urban than rural referrals. VISN 2 reported similar numbers for both, with slightly more urban than rural referrals. When reviewing trends over time for each FY, VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals and VISN 4, VISN 8, and VISN 22 had more urban than rural referrals. In FY 2019 and FY 2020, VISN 2 reported slightly more urban than rural referrals but almost the same number of referrals in FY 2021 and FY 2022 (Appendix 2).

Demographics

The mean (SD) age was 61.2 (15.8) years (range, 20-105). Most PT CC referrals were for veterans aged 70 to 79 years (26.9%), followed by 60 to 69 years (20.7%), and 50 to 59 years (16.4%) (Appendix 3). Trends were consistent across VISNs. There was less of a difference between rural and urban referral percentages as the population aged. Veterans aged < 49 years residing in more urban areas accounted for more referrals to CC compared to their rural counterparts. This difference was less apparent in the 70 to 79 years and 80 to 89 years age brackets.

Most PT CC referrals (81.2%) were male and 14.8% were female. About 3.6% of referral data were missing sex information, and there was a smaller difference between male veterans living in rural communities and male veterans living in urban communities compared with female veterans. A total of 42.9% of male veterans resided in rural areas compared to 56.8% in urban areas; 32.7% of female veterans resided in rural areas compared to 66.9% in urban areas (Appendix 3).

Other Criteria

Of the 334,406 referrals, 114,983 (34.4%) had eligibility data, mostly from FY 2021 and FY 2022 (Table 3). Available eligibility data were likely affected by the MISSION Act and new regulations for reporting CC eligibility. Distance (33.4%) was the most common eligibility criteria, followed by timeliness of care (28.8%), and best medical interest (19.8%); 40.4% were rural and 59.5% were urban. Distance (55.4%) was most common for rural veterans, while timeliness of care (39.7%) was most common for urban veterans. For both groups, the second most common eligibility reason was best medical interest (Appendix 4).

Bone, joint, or soft tissue disorders were common diagnoses, with 25.2% located in the lower back, 14.7% in the shoulder, and 12.8% in the knee (Appendix 5). Amputations of the upper and lower limbs, fractures, cancer-related diagnoses, integumentary system disorders, thoracic and abdominal injuries and disorders, and other medical and mental health conditions each accounted for < 1% of the total diagnoses.

Costs

At time of analysis, the CC Dashboard had cost data available for 200,204 CC PT referrals from FY 2020 to FY 2022. The difference in referral numbers for the 2 datasets is likely attributed to several factors: CC cost data is exclusively from the HSRM, whereas Palantir includes other data sources; how VA cleans data pulled into Palantir; how the CC Dashboard algorithm populates data; and variances based on timing of reporting and/or if referrals are eventually canceled.

The total cost of PT CC referrals from FY 2020 to FY 2022 in selected VISNs was about $220,615,399 (Appendix 6). Appendix 7 details the methodology for determining the average standardized episode- of-care cost by VISN and how referral costs are calculated. Data show a continuous increase in total estimated cost from $46.8 million in FY 2020 to $92.1 million in FY 2022. From FY 2020 to FY 2022, aggregate costs ranged from $6,758,053 in VISN 2 to $47,209,162 in VISN 8 (Figure 3). The total referral cost for PT was highest at VISN 4 in FY 2020 ($10,447,140) and highest at VISN 22 in FY 2021 ($18,835,657) and FY 2022 ($22,962,438) (Figure 4). For referral costs from FY 2020 to FY 2022, distance accounted for $75,561,948 (34.3%), timeliness of care accounted for $60,413,496 (27.3%), and best medical interest accounted for $46,291,390 (21.0%) (Table 4).

Overall costs were primarily driven by specific VISNs within each eligibility type (Appendix 8; Figure 5). VISN 19, VISN 22, and VISN 15 accounted for the highest referral costs for distance; VISN 22, VISN 8, and VISN 19 accounted for the secondhighest referral cost, timeliness of care; and VISN 4, VISN 8, and VISN 12 accounted for the third-highest referral cost, best medical interest (Figure 5). VISN 2, VISN 4, VISN 12, VISN 15, and VISN 22 had service unavailable as an eligibility type with 1 of the top 3 associated referral costs, which was higher in cost than timeliness of care for VISN 2, VISN 4, VISN 12, and VISN 15.

Discussion

This study examines the referral of rehabilitation PT services to CC, evaluates CC costs for PT services, and analyzes utilization and cost trends among veterans within the VHA. Utilization data demonstrated a decrease in referrals from FY 2019 to FY 2020 and increases in referrals from FY 2020 to FY 2022 for most variables of interest, with cost data exhibiting similar trends. Results highlight the need for further investigation to address variations in PT referrals and costs across VISNs and eligibility reasons for CC referral.

Results demonstrated a noteworthy increase in PT CC referrals over time. The largest increase occurred from FY 2020 to FY 2021, with a smaller increase from FY 2021 to FY 2022. During this period, total enrollee numbers decreased by 3.0% across the 7 VISNs included in this analysis and by 1.6% across all VISNs, a trend that illustrates an overall decrease in enrollees as CC use increased. Results align with the implementation of the MISSION Act of 2018, which further expanded veterans’ options to use CC.^1,6,7 Results also align with the onset of the COVID-19 pandemic, which disrupted care access for many veterans, placed a larger emphasis on the use of telehealth, and increased opportunities to stay within the VA for care by rapidly shifting to telehealth and leveraging telerehabilitation investments and initiatives (such as TR-EWI).^20,31

VISN 8, VISN 19, and VISN 22, accounted for more than half of PT referrals. These VISNs had higher enrollee counts compared to the other VISNs.³² VISN 8 consistently had high levels of referrals, whereas VISN 19 and VISN 22 saw dramatic increases in FY 2021 and FY 2022. In contrast, VISN 4 and VISN 12 gradually decreased referrals during the study. VISN 2 had the lowest referral numbers during the study period, and all stations with the lowest individual referral numbers were located within VISN 2. Of the VISNs included in this study, VISN 2 had the second lowest number of enrollees (324,042).³² Reasons for increases and decreases over time could not be determined based on data collected in this study.

There were more urban than rural PT CC referrals; however, both exhibited an increase in referrals over time. This is consistent with population trends showing that most VHA patients (62.6%) and veterans (75.9%) reside in urban areas, which could explain some of the trends in this study.³³ Some VISNs have larger urban catchment areas (eg, VISN 8 and VISN 22), and some have larger rural catchment areas (eg, VISN 15 and VISN 19), which could partially explain the rural-urban differences by VISN.³² Rural-urban referral trends might also reflect existing health care delivery system deficits in rural areas and known challenges associated with accessing health care for veterans living in rural communities.^8,9

This study found larger differences in rural and urban PT CC referrals for younger age groups, with more than twice as many urban referrals in veterans aged 20 to 29 years and aged 30 to 39 years, and roughly 1.8 times as many urban referrals in veterans aged 40 to 49 years. However, there were similar numbers of rural and urban referrals in those aged 70 to 79 years and aged 80 to 89 years. These trends are consistent with data showing veterans residing in rural communities are older than their urban counterparts.^23,34 Data suggest that older veteran populations might seek PT at higher rates than younger veteran populations. Moreover, data suggest there could be differences in PT-seeking rates for younger veteran populations who reside in rural vs urban areas. Additional research is needed to understand these trends.

Distance and timeliness of care were the predominant reasons for referral among eligibility groups, which is consistent with the MISSION Act goals.^1,6,7 The most common eligibility reason for rural referrals was distance; timeliness of care was most common for urban referrals. This finding is expected, as veterans living in rural communities are farther away from VHA facilities and have longer drive times, whereas veterans living in urban communities might live closer, yet experience longer wait times due to services and/or appointment availability. Best medical interest accounted for almost 20% of referrals, which does not provide detailed insights into why those veterans were referred to CC.

The top PT diagnoses referred to CC were related to bone, joint, or soft tissue disorders of the lower back, shoulder, and knee. This suggests that musculoskeletal-related issues are prevalent among veterans seeking PT care, which is consistent with research that found > 50% of veterans receiving VHA care have musculoskeletal disorders.³⁵ The probability of experiencing musculoskeletal problems increases with age, as does the need for PT services. Amputations and fractures accounted for < 1% of CC referrals, which is consistent with the historic provision of VHA clinical specialized care to conditions prevalent among veterans. It may also represent VHA efforts to internally provide care for complex conditions requiring more extensive interdisciplinary coordination.

The total cost of referrals over time was about $221 million. VISN 8 accounted for the highest overall cost; VISN 2 had the lowest, mirroring referral utilization trends and aligning with VISN enrollee numbers. VISN 19 and VISN 22 reported large cost increases from FY 2020 to FY 2021. Total referral costs increased by $34.9 million from FY 2020 to FY 2021, which may be due to health care inflation (2.9% during FY 2019 to FY 2022), increased awareness of CC services, or increased VHA wait times.³⁶ Additionally, there were limitations in care provided across health care systems during the COVID-19 pandemic, including the VA.⁵ The increase from FY 2020 to FY 2021 may reflect a rebound from restrictions in appointments across VA, CC, and the private sector.

While the increase in total referral cost may be partly attributed to inflation, the cost effectiveness and efficiency of referring veterans to CC vs keeping veterans within VHA care is an ongoing debate.⁵ Examining and addressing cost drivers within the top eligibility types and their respective VISNs is necessary to determine resource allocation and improve quality of care. This study found that best medical interest and unavailable services accounted for 33.4% of the total cost of CC referrals, highlighting the need for policies that strengthen in-house competencies and recruit personnel to provide PT services currently unavailable within the VA.

Future Directions

The VHA should explore opportunities for in-house care, especially for services appropriate for telehealth.^18,20,37 Data indicated a smaller cost increase from FY 2021 to FY 2022 compared to the relatively large increase from FY 2020 to FY 2021. The increased telehealth usage across VHA by TR-EWI and non—TR-EWI sites within selected VISNs may have contributed to limiting the increase in CC costs. Future studies should investigate contextual factors of increased telehealth usage, which would offer guidance for implementation to optimize the integration of telehealth with PT rehabilitation provided in-house. Additionally, future studies can examine potential limitations experienced during PT telehealth visits, such as the inability to conduct hands-on assessments, challenges in viewing the quality of patient movement, ensuring patient safety in the remote environment, and the lack of PT equipment in homes for telehealth visits, and how these challenges are being addressed.^38,39 Research is also needed to understand tradeoffs of CC vs VHA care and the potential and cost benefits of keeping veterans within VHA using programs like TR-EWI.⁵ Veterans living in rural communities may especially benefit from this as expanding telehealth options can provide access to PT care that may not be readily available, enabling them to stay connected and engaged in their care.^18,40

Future studies could examine contributory factors to rising costs, such as demographic shifts, changes in PT service utilization, and policy. Researchers might also consider qualitative studies with clinicians and veterans within each VISN, which may provide insights into how local factors impact PT referral to the community.

Limitations

Due to its descriptive nature, this study can only speculate about factors influencing trends. Limitations include the inability to link the Palantir and CC Dashboard datasets for cost comparisons and potential data change over time on Palantir due to platform updates. The focus on VISNs with TREWI sites limited generalizability and this study did not compare CC PT vs VHA PT. Finally, there may have been cost drivers not identified in this study.

Conclusions

This descriptive study provides insights into the utilization and cost of PT CC referrals for selected VISNs. Cost trends underscore the financial commitment to providing PT services to veterans. Understanding what factors are driving this cost is necessary for VHA to optimally provide and manage the rehabilitation resources needed to serve veterans through traditional in-person care, telehealth, and CC options while ensuring timely, highquality care.

The Veterans Health Administration (VHA) is the largest US integrated health system, providing care to veterans through VHA and non-VHA practitioners and facilities.^1,2 Providing high-quality, timely, and veteran-centric care remains a priority for the VHA. Legislative efforts have expanded opportunities for eligible veterans to receive care in the community purchased by VHA, known as community care (CC).¹ The Veterans Access, Choice, and Accountability Act of 2014 came in response to reports of long wait times and drive times for patients.^3-5 The MISSION Act of 2018 expanded access to CC by streamlining it and broadening eligibility criteria, especially for veterans in rural communities who often experience more barriers in accessing care than veterans living in urban communities.^1,6-10 Since the implementation of the Choice and MISSION Acts, > 2.7 million veterans have received care through community practitioners within the VHA CC network.¹¹

Background

Increased access to CC could benefit veterans living in rural communities by increasing care options and circumventing challenges to accessing VHA care (ie, geographic, transportation, and distance barriers, practitioner and specialist shortages, and hospital closures). ^5,9,10,12,13 However, health care system deficits in rural areas could also limit CC effectiveness for veterans living in those communities. ³ Other challenges posed by using CC include care coordination, information sharing, care continuity, delayed payments to CC practitioners, and mixed findings regarding CC quality.^5,8,13,14 VHA practitioners are specifically trained to meet the multifaceted needs unique to veterans’ health and subculture, training CC practitioners may not receive.^5,15

CC offers services for primary care and a broad range of specialties, including rehabilitation services such as physical therapy (PT).⁶ PT is used for the effective treatment of various conditions veterans experience and promote wellbeing and independence.¹⁶ US Department of Veterans Affairs (VA) databases reveal a high prevalence of veterans receiving PT services through CC; PT is one of the most frequently used CC outpatient specialty services by veterans living in rural communities.^14,17

Telerehabitltation Enterprisewide Initiative

VHA has greatly invested in delivering care virtually, especially for veterans living in rural communities.¹⁸ In 2017, the VHA Office of Rural Health funded the Telerehabilitation Enterprise-Wide Initiative (TR-EWI) in partnership with the Physical Medicine and Rehabilitation Services national program office to increase access to specialized rehabilitation services for veterans living in rural communities by leveraging telehealth technologies.^18-21 This alternative mode of health care delivery allows clinicians to overcome access barriers by delivering rehabilitation therapies directly to veterans' homes or nearby community-based outpatient clinics. TR-EWI was conceived as a hub-and-spoke model, where rehabilitation expertise at the hub was virtually delivered to spoke sites that did not have in-house expertise. In subsequent years, the TR-EWI also evolved to provide targeted telerehabilitation programs within rural-serving community-based outpatient clinics, including PT as a predominant service.^19,20

As TR-EWI progressed—and in conjunction with the uptake of telehealth across VHA during the COVID-19 pandemic—there has been increased focus on PT telerehabilitation, especially for the 4.6 million veterans in rural communities.^18,22,23 Because health care delivery system deficits in rural areas could limit the effective use of CC, many TR-EWI sites hope to reduce their CC referrals by providing telehealth PT services to veterans who might otherwise need to be referred to CC. This strategy aligns with VHA goals of providing high-quality and timely care. To better understand opportunities for programs like TR-EWI to provide rehabilitation services for veterans and reduce care sent to the community, research that examines CC referral trends for PT over time is warranted.

This study examines CC from a rehabilitation perspective with a focus on CC referral trends for PT, specifically for Veterans Integrated Service Networks (VISNs) where TREWI sites are located. The study’s objectives were to describe rehabilitation PT services being referred to CC and examine associated CC costs for PT services. Two research questions guided the study. First, what are the utilization trends for CC PT referrals from fiscal year (FY) 2019 to FY 2022? Secondly, what is the cost breakdown of CC for PT referrals from FY 2020 to FY 2022?

Methods

This study was conducted by a multidisciplinary team comprised of public health, disability, rehabilitation counseling, and PT professionals. It was deemed a quality improvement project under VA guidance and followed the SQUIRE guidelines for quality improvement reporting.^24,25 The study used the VA Common Operating Platform (Palantir) to obtain individual-level CC referral data from the HealthShare Referral Manager (HSRM) database and consult data from the Computerized Patient Record System. Palantir is used to store and integrate VA data derived from the VA Corporate Data Warehouse and VHA Support Service Center. Referrals are authorizations for care to be delivered by a CC practitioner.

TR-EWI is comprised of 7 sites: VISN 2, VISN 4, VISN 8, VISN 12, VISN 15, VISN 19, and VISN 22. Each site provides telerehabilitation services with an emphasis on reaching veterans living in rural communities. We joined the referrals and consults cubes in Palantir to extract PT referrals for FY 2019 to FY 2022 for the 7 VISNs with TR-EWI sites and obtain referral-specific information and demographic characteristics. ²⁶ Data were extracted in October 2022.

The VHA Community Care Referral Dashboard (CC Dashboard) provided nonindividual level CC cost data.²⁷ The CC Dashboard provides insights into the costs of CC services for VHA enrollees by category of care, standardized episode of care, and eligibility. Data are based on nationallevel HSRM referrals that are not suspended or linked to a canceled or discontinued consult. Data were aggregated by VISN. The dashboard only includes referrals dating back to FY 2020; therefore, PT data from FY 2020 through FY 2022 for VISNs with TR-EWI sites were collected. Data were extracted in December 2022.

This study examined CC referrals, station name, eligibility types, clinical diagnoses (International Classification of Diseases, Tenth Revision codes), and demographic information in the Palantir dataset. Six eligibility criteria can qualify a veteran to receive CC.²⁸ Within clinical diagnoses, the variable of interest was the provisional diagnosis. Patient demographics included age, gender, and rurality of residence, as determined by the Rural-Urban Commuting Area system.^29,30 Rural and highly rural categories were combined for analysis. For the CC cost dataset, this study examined CC referrals, referral cost, and eligibility type.

Analysis

For the first research question, we examined referral data from FY 2019 to FY 2022 using the Palantir dataset, performed descriptive statistical analysis for all variables, and analyzed data to identify trends. Descriptive statistics were completed using IBM SPSS Statistics for Windows Version 29.0.0.0.

A qualitative analysis of provisional diagnosis data revealed what is being referred to CC for PT. A preliminary overview of provisional diagnosis data was conducted to familiarize coders with the data. We developed a coding framework to categorize diagnoses based on anatomical location, body structure, and clinical areas of interest. Data were reviewed individually and grouped into categories within the coding framework before meeting as a team to achieve group consensus on categorization. We then totaled the frequency of occurrence for provisional diagnoses within each category. Qualitative analyses were completed using Microsoft Excel.

For the second research question, the study used the CC cost dataset to examine the cost breakdown of CC PT referrals from FY 2020 to FY 2022. We calculated the number and cost of PT referrals across eligibility groups for each FY and VISN. Data were analyzed using SPSS to identify cost trends.

Results

There were 344,406 referrals to CC for PT from FY 2019 to FY 2022 for the 7 VISNs analyzed (Table 1). Of these, 22.5% were from FY 2019, 19.1% from FY 2020, 28.2% from FY 2021, and 30.3% from FY 2022. VISN 8 and VISN 22 reported the most overall PT referrals, with VISN 8 comprising 22.2% and VISN 22 comprising 18.1% of all referrals. VISN 2 reported the least overall referrals (3.7%). VISN 4 and VISN 12 had decreases in referrals over time. VISN 2 and VISN 15 had decreases in referrals from FY 2019 to FY 2021 and slight increases from FY 2021 to FY 2022. VISN 19 and VISN 22 both saw slight increases from FY 2019 to FY 2020 and substantial increases from FY 2020 to FY 2022, with FY 2022 accounting for 40.0% and 42.3% of all referrals for VISN 19 and VISN 20, respectively (Figure 1).

For FY 2019 and FY 2020, VISN 8 had the highest percentage of referrals (26.7% and 23.2%, respectively), whereas VISN 22 was among the lowest (7.3% and 11.4%, respectively). However, for FY 2021 and FY 2022, VISN 22 reported the highest percentage of referrals (23.5% and 25.3%, respectively) compared to all other VISNs. VISN 2 consistently reported the lowest percentage of referrals across all years.

There were 56 stations analyzed across the 7 VISNs (Appendix 1). Nine stations each accounted for ≥ 3.0% of the total PT referrals and only 2 stations accounted for > 5.0% of referrals. Orlando, Florida (6.0%), Philadelphia, Pennsylvania (5.2%), Tampa, Florida (4.9%), Aurora, Colorado (4.9%), and Gainesville, Florida (4.4%) reported the top 5 highest referrals, with 3 being from VISN 8 (Orlando, Tampa, Gainesville). Stations with the lowest reported referrals were all in VISN 2 in New York: The Bronx, (0%), New York Harbor (0%), Hudson Valley (0.1%) and Finger Lakes (0.2%).

Rurality

Urban stations comprised 56.2% and rural stations comprised 39.8% of PT CC referrals, while 0.2% of referrals were from insular isle US territories: Guam, American Samoa, Northern Marianas, and the Virgin Islands. The sample had missing or unknown data for 3.8% of referrals. FY 2022 had the largest difference in rural and urban referrals. Additionally, there was an overall trend of more referrals over time for rural and urban, with a large increase in rural (+40.0%) and urban (+62.7%) referrals from FY 2020 to FY 2021 and a modest increase from FY 2021 to FY 2022 (+5.2% for rural and +9.1% for urban). There was a decrease in rural (-7.0%) and urban (-3.5%) referrals from FY 2019 to FY 2020 (Figure 2).

There were differences in referrals by rurality and VISN (Table 2). VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals, whereas VISN 4, VISN 8, and VISN 22 reported more urban than rural referrals. VISN 2 reported similar numbers for both, with slightly more urban than rural referrals. When reviewing trends over time for each FY, VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals and VISN 4, VISN 8, and VISN 22 had more urban than rural referrals. In FY 2019 and FY 2020, VISN 2 reported slightly more urban than rural referrals but almost the same number of referrals in FY 2021 and FY 2022 (Appendix 2).

Demographics

The mean (SD) age was 61.2 (15.8) years (range, 20-105). Most PT CC referrals were for veterans aged 70 to 79 years (26.9%), followed by 60 to 69 years (20.7%), and 50 to 59 years (16.4%) (Appendix 3). Trends were consistent across VISNs. There was less of a difference between rural and urban referral percentages as the population aged. Veterans aged < 49 years residing in more urban areas accounted for more referrals to CC compared to their rural counterparts. This difference was less apparent in the 70 to 79 years and 80 to 89 years age brackets.

Most PT CC referrals (81.2%) were male and 14.8% were female. About 3.6% of referral data were missing sex information, and there was a smaller difference between male veterans living in rural communities and male veterans living in urban communities compared with female veterans. A total of 42.9% of male veterans resided in rural areas compared to 56.8% in urban areas; 32.7% of female veterans resided in rural areas compared to 66.9% in urban areas (Appendix 3).

Other Criteria

Of the 334,406 referrals, 114,983 (34.4%) had eligibility data, mostly from FY 2021 and FY 2022 (Table 3). Available eligibility data were likely affected by the MISSION Act and new regulations for reporting CC eligibility. Distance (33.4%) was the most common eligibility criteria, followed by timeliness of care (28.8%), and best medical interest (19.8%); 40.4% were rural and 59.5% were urban. Distance (55.4%) was most common for rural veterans, while timeliness of care (39.7%) was most common for urban veterans. For both groups, the second most common eligibility reason was best medical interest (Appendix 4).

Bone, joint, or soft tissue disorders were common diagnoses, with 25.2% located in the lower back, 14.7% in the shoulder, and 12.8% in the knee (Appendix 5). Amputations of the upper and lower limbs, fractures, cancer-related diagnoses, integumentary system disorders, thoracic and abdominal injuries and disorders, and other medical and mental health conditions each accounted for < 1% of the total diagnoses.

Costs

At time of analysis, the CC Dashboard had cost data available for 200,204 CC PT referrals from FY 2020 to FY 2022. The difference in referral numbers for the 2 datasets is likely attributed to several factors: CC cost data is exclusively from the HSRM, whereas Palantir includes other data sources; how VA cleans data pulled into Palantir; how the CC Dashboard algorithm populates data; and variances based on timing of reporting and/or if referrals are eventually canceled.

The total cost of PT CC referrals from FY 2020 to FY 2022 in selected VISNs was about $220,615,399 (Appendix 6). Appendix 7 details the methodology for determining the average standardized episode- of-care cost by VISN and how referral costs are calculated. Data show a continuous increase in total estimated cost from $46.8 million in FY 2020 to $92.1 million in FY 2022. From FY 2020 to FY 2022, aggregate costs ranged from $6,758,053 in VISN 2 to $47,209,162 in VISN 8 (Figure 3). The total referral cost for PT was highest at VISN 4 in FY 2020 ($10,447,140) and highest at VISN 22 in FY 2021 ($18,835,657) and FY 2022 ($22,962,438) (Figure 4). For referral costs from FY 2020 to FY 2022, distance accounted for $75,561,948 (34.3%), timeliness of care accounted for $60,413,496 (27.3%), and best medical interest accounted for $46,291,390 (21.0%) (Table 4).

Overall costs were primarily driven by specific VISNs within each eligibility type (Appendix 8; Figure 5). VISN 19, VISN 22, and VISN 15 accounted for the highest referral costs for distance; VISN 22, VISN 8, and VISN 19 accounted for the secondhighest referral cost, timeliness of care; and VISN 4, VISN 8, and VISN 12 accounted for the third-highest referral cost, best medical interest (Figure 5). VISN 2, VISN 4, VISN 12, VISN 15, and VISN 22 had service unavailable as an eligibility type with 1 of the top 3 associated referral costs, which was higher in cost than timeliness of care for VISN 2, VISN 4, VISN 12, and VISN 15.

Discussion

This study examines the referral of rehabilitation PT services to CC, evaluates CC costs for PT services, and analyzes utilization and cost trends among veterans within the VHA. Utilization data demonstrated a decrease in referrals from FY 2019 to FY 2020 and increases in referrals from FY 2020 to FY 2022 for most variables of interest, with cost data exhibiting similar trends. Results highlight the need for further investigation to address variations in PT referrals and costs across VISNs and eligibility reasons for CC referral.

Results demonstrated a noteworthy increase in PT CC referrals over time. The largest increase occurred from FY 2020 to FY 2021, with a smaller increase from FY 2021 to FY 2022. During this period, total enrollee numbers decreased by 3.0% across the 7 VISNs included in this analysis and by 1.6% across all VISNs, a trend that illustrates an overall decrease in enrollees as CC use increased. Results align with the implementation of the MISSION Act of 2018, which further expanded veterans’ options to use CC.^1,6,7 Results also align with the onset of the COVID-19 pandemic, which disrupted care access for many veterans, placed a larger emphasis on the use of telehealth, and increased opportunities to stay within the VA for care by rapidly shifting to telehealth and leveraging telerehabilitation investments and initiatives (such as TR-EWI).^20,31

VISN 8, VISN 19, and VISN 22, accounted for more than half of PT referrals. These VISNs had higher enrollee counts compared to the other VISNs.³² VISN 8 consistently had high levels of referrals, whereas VISN 19 and VISN 22 saw dramatic increases in FY 2021 and FY 2022. In contrast, VISN 4 and VISN 12 gradually decreased referrals during the study. VISN 2 had the lowest referral numbers during the study period, and all stations with the lowest individual referral numbers were located within VISN 2. Of the VISNs included in this study, VISN 2 had the second lowest number of enrollees (324,042).³² Reasons for increases and decreases over time could not be determined based on data collected in this study.

There were more urban than rural PT CC referrals; however, both exhibited an increase in referrals over time. This is consistent with population trends showing that most VHA patients (62.6%) and veterans (75.9%) reside in urban areas, which could explain some of the trends in this study.³³ Some VISNs have larger urban catchment areas (eg, VISN 8 and VISN 22), and some have larger rural catchment areas (eg, VISN 15 and VISN 19), which could partially explain the rural-urban differences by VISN.³² Rural-urban referral trends might also reflect existing health care delivery system deficits in rural areas and known challenges associated with accessing health care for veterans living in rural communities.^8,9

This study found larger differences in rural and urban PT CC referrals for younger age groups, with more than twice as many urban referrals in veterans aged 20 to 29 years and aged 30 to 39 years, and roughly 1.8 times as many urban referrals in veterans aged 40 to 49 years. However, there were similar numbers of rural and urban referrals in those aged 70 to 79 years and aged 80 to 89 years. These trends are consistent with data showing veterans residing in rural communities are older than their urban counterparts.^23,34 Data suggest that older veteran populations might seek PT at higher rates than younger veteran populations. Moreover, data suggest there could be differences in PT-seeking rates for younger veteran populations who reside in rural vs urban areas. Additional research is needed to understand these trends.

Distance and timeliness of care were the predominant reasons for referral among eligibility groups, which is consistent with the MISSION Act goals.^1,6,7 The most common eligibility reason for rural referrals was distance; timeliness of care was most common for urban referrals. This finding is expected, as veterans living in rural communities are farther away from VHA facilities and have longer drive times, whereas veterans living in urban communities might live closer, yet experience longer wait times due to services and/or appointment availability. Best medical interest accounted for almost 20% of referrals, which does not provide detailed insights into why those veterans were referred to CC.

The top PT diagnoses referred to CC were related to bone, joint, or soft tissue disorders of the lower back, shoulder, and knee. This suggests that musculoskeletal-related issues are prevalent among veterans seeking PT care, which is consistent with research that found > 50% of veterans receiving VHA care have musculoskeletal disorders.³⁵ The probability of experiencing musculoskeletal problems increases with age, as does the need for PT services. Amputations and fractures accounted for < 1% of CC referrals, which is consistent with the historic provision of VHA clinical specialized care to conditions prevalent among veterans. It may also represent VHA efforts to internally provide care for complex conditions requiring more extensive interdisciplinary coordination.

The total cost of referrals over time was about $221 million. VISN 8 accounted for the highest overall cost; VISN 2 had the lowest, mirroring referral utilization trends and aligning with VISN enrollee numbers. VISN 19 and VISN 22 reported large cost increases from FY 2020 to FY 2021. Total referral costs increased by $34.9 million from FY 2020 to FY 2021, which may be due to health care inflation (2.9% during FY 2019 to FY 2022), increased awareness of CC services, or increased VHA wait times.³⁶ Additionally, there were limitations in care provided across health care systems during the COVID-19 pandemic, including the VA.⁵ The increase from FY 2020 to FY 2021 may reflect a rebound from restrictions in appointments across VA, CC, and the private sector.

While the increase in total referral cost may be partly attributed to inflation, the cost effectiveness and efficiency of referring veterans to CC vs keeping veterans within VHA care is an ongoing debate.⁵ Examining and addressing cost drivers within the top eligibility types and their respective VISNs is necessary to determine resource allocation and improve quality of care. This study found that best medical interest and unavailable services accounted for 33.4% of the total cost of CC referrals, highlighting the need for policies that strengthen in-house competencies and recruit personnel to provide PT services currently unavailable within the VA.

Future Directions

The VHA should explore opportunities for in-house care, especially for services appropriate for telehealth.^18,20,37 Data indicated a smaller cost increase from FY 2021 to FY 2022 compared to the relatively large increase from FY 2020 to FY 2021. The increased telehealth usage across VHA by TR-EWI and non—TR-EWI sites within selected VISNs may have contributed to limiting the increase in CC costs. Future studies should investigate contextual factors of increased telehealth usage, which would offer guidance for implementation to optimize the integration of telehealth with PT rehabilitation provided in-house. Additionally, future studies can examine potential limitations experienced during PT telehealth visits, such as the inability to conduct hands-on assessments, challenges in viewing the quality of patient movement, ensuring patient safety in the remote environment, and the lack of PT equipment in homes for telehealth visits, and how these challenges are being addressed.^38,39 Research is also needed to understand tradeoffs of CC vs VHA care and the potential and cost benefits of keeping veterans within VHA using programs like TR-EWI.⁵ Veterans living in rural communities may especially benefit from this as expanding telehealth options can provide access to PT care that may not be readily available, enabling them to stay connected and engaged in their care.^18,40

Future studies could examine contributory factors to rising costs, such as demographic shifts, changes in PT service utilization, and policy. Researchers might also consider qualitative studies with clinicians and veterans within each VISN, which may provide insights into how local factors impact PT referral to the community.

Limitations

Due to its descriptive nature, this study can only speculate about factors influencing trends. Limitations include the inability to link the Palantir and CC Dashboard datasets for cost comparisons and potential data change over time on Palantir due to platform updates. The focus on VISNs with TREWI sites limited generalizability and this study did not compare CC PT vs VHA PT. Finally, there may have been cost drivers not identified in this study.

Conclusions

This descriptive study provides insights into the utilization and cost of PT CC referrals for selected VISNs. Cost trends underscore the financial commitment to providing PT services to veterans. Understanding what factors are driving this cost is necessary for VHA to optimally provide and manage the rehabilitation resources needed to serve veterans through traditional in-person care, telehealth, and CC options while ensuring timely, highquality care.

The Veterans Health Administration (VHA) is the largest US integrated health system, providing care to veterans through VHA and non-VHA practitioners and facilities.^1,2 Providing high-quality, timely, and veteran-centric care remains a priority for the VHA. Legislative efforts have expanded opportunities for eligible veterans to receive care in the community purchased by VHA, known as community care (CC).¹ The Veterans Access, Choice, and Accountability Act of 2014 came in response to reports of long wait times and drive times for patients.^3-5 The MISSION Act of 2018 expanded access to CC by streamlining it and broadening eligibility criteria, especially for veterans in rural communities who often experience more barriers in accessing care than veterans living in urban communities.^1,6-10 Since the implementation of the Choice and MISSION Acts, > 2.7 million veterans have received care through community practitioners within the VHA CC network.¹¹

Background

Increased access to CC could benefit veterans living in rural communities by increasing care options and circumventing challenges to accessing VHA care (ie, geographic, transportation, and distance barriers, practitioner and specialist shortages, and hospital closures). ^5,9,10,12,13 However, health care system deficits in rural areas could also limit CC effectiveness for veterans living in those communities. ³ Other challenges posed by using CC include care coordination, information sharing, care continuity, delayed payments to CC practitioners, and mixed findings regarding CC quality.^5,8,13,14 VHA practitioners are specifically trained to meet the multifaceted needs unique to veterans’ health and subculture, training CC practitioners may not receive.^5,15

CC offers services for primary care and a broad range of specialties, including rehabilitation services such as physical therapy (PT).⁶ PT is used for the effective treatment of various conditions veterans experience and promote wellbeing and independence.¹⁶ US Department of Veterans Affairs (VA) databases reveal a high prevalence of veterans receiving PT services through CC; PT is one of the most frequently used CC outpatient specialty services by veterans living in rural communities.^14,17

Telerehabitltation Enterprisewide Initiative

VHA has greatly invested in delivering care virtually, especially for veterans living in rural communities.¹⁸ In 2017, the VHA Office of Rural Health funded the Telerehabilitation Enterprise-Wide Initiative (TR-EWI) in partnership with the Physical Medicine and Rehabilitation Services national program office to increase access to specialized rehabilitation services for veterans living in rural communities by leveraging telehealth technologies.^18-21 This alternative mode of health care delivery allows clinicians to overcome access barriers by delivering rehabilitation therapies directly to veterans' homes or nearby community-based outpatient clinics. TR-EWI was conceived as a hub-and-spoke model, where rehabilitation expertise at the hub was virtually delivered to spoke sites that did not have in-house expertise. In subsequent years, the TR-EWI also evolved to provide targeted telerehabilitation programs within rural-serving community-based outpatient clinics, including PT as a predominant service.^19,20

As TR-EWI progressed—and in conjunction with the uptake of telehealth across VHA during the COVID-19 pandemic—there has been increased focus on PT telerehabilitation, especially for the 4.6 million veterans in rural communities.^18,22,23 Because health care delivery system deficits in rural areas could limit the effective use of CC, many TR-EWI sites hope to reduce their CC referrals by providing telehealth PT services to veterans who might otherwise need to be referred to CC. This strategy aligns with VHA goals of providing high-quality and timely care. To better understand opportunities for programs like TR-EWI to provide rehabilitation services for veterans and reduce care sent to the community, research that examines CC referral trends for PT over time is warranted.

This study examines CC from a rehabilitation perspective with a focus on CC referral trends for PT, specifically for Veterans Integrated Service Networks (VISNs) where TREWI sites are located. The study’s objectives were to describe rehabilitation PT services being referred to CC and examine associated CC costs for PT services. Two research questions guided the study. First, what are the utilization trends for CC PT referrals from fiscal year (FY) 2019 to FY 2022? Secondly, what is the cost breakdown of CC for PT referrals from FY 2020 to FY 2022?

Methods

This study was conducted by a multidisciplinary team comprised of public health, disability, rehabilitation counseling, and PT professionals. It was deemed a quality improvement project under VA guidance and followed the SQUIRE guidelines for quality improvement reporting.^24,25 The study used the VA Common Operating Platform (Palantir) to obtain individual-level CC referral data from the HealthShare Referral Manager (HSRM) database and consult data from the Computerized Patient Record System. Palantir is used to store and integrate VA data derived from the VA Corporate Data Warehouse and VHA Support Service Center. Referrals are authorizations for care to be delivered by a CC practitioner.

TR-EWI is comprised of 7 sites: VISN 2, VISN 4, VISN 8, VISN 12, VISN 15, VISN 19, and VISN 22. Each site provides telerehabilitation services with an emphasis on reaching veterans living in rural communities. We joined the referrals and consults cubes in Palantir to extract PT referrals for FY 2019 to FY 2022 for the 7 VISNs with TR-EWI sites and obtain referral-specific information and demographic characteristics. ²⁶ Data were extracted in October 2022.

The VHA Community Care Referral Dashboard (CC Dashboard) provided nonindividual level CC cost data.²⁷ The CC Dashboard provides insights into the costs of CC services for VHA enrollees by category of care, standardized episode of care, and eligibility. Data are based on nationallevel HSRM referrals that are not suspended or linked to a canceled or discontinued consult. Data were aggregated by VISN. The dashboard only includes referrals dating back to FY 2020; therefore, PT data from FY 2020 through FY 2022 for VISNs with TR-EWI sites were collected. Data were extracted in December 2022.

This study examined CC referrals, station name, eligibility types, clinical diagnoses (International Classification of Diseases, Tenth Revision codes), and demographic information in the Palantir dataset. Six eligibility criteria can qualify a veteran to receive CC.²⁸ Within clinical diagnoses, the variable of interest was the provisional diagnosis. Patient demographics included age, gender, and rurality of residence, as determined by the Rural-Urban Commuting Area system.^29,30 Rural and highly rural categories were combined for analysis. For the CC cost dataset, this study examined CC referrals, referral cost, and eligibility type.

Analysis

For the first research question, we examined referral data from FY 2019 to FY 2022 using the Palantir dataset, performed descriptive statistical analysis for all variables, and analyzed data to identify trends. Descriptive statistics were completed using IBM SPSS Statistics for Windows Version 29.0.0.0.

A qualitative analysis of provisional diagnosis data revealed what is being referred to CC for PT. A preliminary overview of provisional diagnosis data was conducted to familiarize coders with the data. We developed a coding framework to categorize diagnoses based on anatomical location, body structure, and clinical areas of interest. Data were reviewed individually and grouped into categories within the coding framework before meeting as a team to achieve group consensus on categorization. We then totaled the frequency of occurrence for provisional diagnoses within each category. Qualitative analyses were completed using Microsoft Excel.

For the second research question, the study used the CC cost dataset to examine the cost breakdown of CC PT referrals from FY 2020 to FY 2022. We calculated the number and cost of PT referrals across eligibility groups for each FY and VISN. Data were analyzed using SPSS to identify cost trends.

Results

There were 344,406 referrals to CC for PT from FY 2019 to FY 2022 for the 7 VISNs analyzed (Table 1). Of these, 22.5% were from FY 2019, 19.1% from FY 2020, 28.2% from FY 2021, and 30.3% from FY 2022. VISN 8 and VISN 22 reported the most overall PT referrals, with VISN 8 comprising 22.2% and VISN 22 comprising 18.1% of all referrals. VISN 2 reported the least overall referrals (3.7%). VISN 4 and VISN 12 had decreases in referrals over time. VISN 2 and VISN 15 had decreases in referrals from FY 2019 to FY 2021 and slight increases from FY 2021 to FY 2022. VISN 19 and VISN 22 both saw slight increases from FY 2019 to FY 2020 and substantial increases from FY 2020 to FY 2022, with FY 2022 accounting for 40.0% and 42.3% of all referrals for VISN 19 and VISN 20, respectively (Figure 1).

For FY 2019 and FY 2020, VISN 8 had the highest percentage of referrals (26.7% and 23.2%, respectively), whereas VISN 22 was among the lowest (7.3% and 11.4%, respectively). However, for FY 2021 and FY 2022, VISN 22 reported the highest percentage of referrals (23.5% and 25.3%, respectively) compared to all other VISNs. VISN 2 consistently reported the lowest percentage of referrals across all years.

There were 56 stations analyzed across the 7 VISNs (Appendix 1). Nine stations each accounted for ≥ 3.0% of the total PT referrals and only 2 stations accounted for > 5.0% of referrals. Orlando, Florida (6.0%), Philadelphia, Pennsylvania (5.2%), Tampa, Florida (4.9%), Aurora, Colorado (4.9%), and Gainesville, Florida (4.4%) reported the top 5 highest referrals, with 3 being from VISN 8 (Orlando, Tampa, Gainesville). Stations with the lowest reported referrals were all in VISN 2 in New York: The Bronx, (0%), New York Harbor (0%), Hudson Valley (0.1%) and Finger Lakes (0.2%).

Rurality

Urban stations comprised 56.2% and rural stations comprised 39.8% of PT CC referrals, while 0.2% of referrals were from insular isle US territories: Guam, American Samoa, Northern Marianas, and the Virgin Islands. The sample had missing or unknown data for 3.8% of referrals. FY 2022 had the largest difference in rural and urban referrals. Additionally, there was an overall trend of more referrals over time for rural and urban, with a large increase in rural (+40.0%) and urban (+62.7%) referrals from FY 2020 to FY 2021 and a modest increase from FY 2021 to FY 2022 (+5.2% for rural and +9.1% for urban). There was a decrease in rural (-7.0%) and urban (-3.5%) referrals from FY 2019 to FY 2020 (Figure 2).

There were differences in referrals by rurality and VISN (Table 2). VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals, whereas VISN 4, VISN 8, and VISN 22 reported more urban than rural referrals. VISN 2 reported similar numbers for both, with slightly more urban than rural referrals. When reviewing trends over time for each FY, VISN 12, VISN 15, and VISN 19 reported more rural than urban referrals and VISN 4, VISN 8, and VISN 22 had more urban than rural referrals. In FY 2019 and FY 2020, VISN 2 reported slightly more urban than rural referrals but almost the same number of referrals in FY 2021 and FY 2022 (Appendix 2).

Demographics

The mean (SD) age was 61.2 (15.8) years (range, 20-105). Most PT CC referrals were for veterans aged 70 to 79 years (26.9%), followed by 60 to 69 years (20.7%), and 50 to 59 years (16.4%) (Appendix 3). Trends were consistent across VISNs. There was less of a difference between rural and urban referral percentages as the population aged. Veterans aged < 49 years residing in more urban areas accounted for more referrals to CC compared to their rural counterparts. This difference was less apparent in the 70 to 79 years and 80 to 89 years age brackets.

Most PT CC referrals (81.2%) were male and 14.8% were female. About 3.6% of referral data were missing sex information, and there was a smaller difference between male veterans living in rural communities and male veterans living in urban communities compared with female veterans. A total of 42.9% of male veterans resided in rural areas compared to 56.8% in urban areas; 32.7% of female veterans resided in rural areas compared to 66.9% in urban areas (Appendix 3).

Other Criteria

Of the 334,406 referrals, 114,983 (34.4%) had eligibility data, mostly from FY 2021 and FY 2022 (Table 3). Available eligibility data were likely affected by the MISSION Act and new regulations for reporting CC eligibility. Distance (33.4%) was the most common eligibility criteria, followed by timeliness of care (28.8%), and best medical interest (19.8%); 40.4% were rural and 59.5% were urban. Distance (55.4%) was most common for rural veterans, while timeliness of care (39.7%) was most common for urban veterans. For both groups, the second most common eligibility reason was best medical interest (Appendix 4).

Bone, joint, or soft tissue disorders were common diagnoses, with 25.2% located in the lower back, 14.7% in the shoulder, and 12.8% in the knee (Appendix 5). Amputations of the upper and lower limbs, fractures, cancer-related diagnoses, integumentary system disorders, thoracic and abdominal injuries and disorders, and other medical and mental health conditions each accounted for < 1% of the total diagnoses.

Costs

At time of analysis, the CC Dashboard had cost data available for 200,204 CC PT referrals from FY 2020 to FY 2022. The difference in referral numbers for the 2 datasets is likely attributed to several factors: CC cost data is exclusively from the HSRM, whereas Palantir includes other data sources; how VA cleans data pulled into Palantir; how the CC Dashboard algorithm populates data; and variances based on timing of reporting and/or if referrals are eventually canceled.

The total cost of PT CC referrals from FY 2020 to FY 2022 in selected VISNs was about $220,615,399 (Appendix 6). Appendix 7 details the methodology for determining the average standardized episode- of-care cost by VISN and how referral costs are calculated. Data show a continuous increase in total estimated cost from $46.8 million in FY 2020 to $92.1 million in FY 2022. From FY 2020 to FY 2022, aggregate costs ranged from $6,758,053 in VISN 2 to $47,209,162 in VISN 8 (Figure 3). The total referral cost for PT was highest at VISN 4 in FY 2020 ($10,447,140) and highest at VISN 22 in FY 2021 ($18,835,657) and FY 2022 ($22,962,438) (Figure 4). For referral costs from FY 2020 to FY 2022, distance accounted for $75,561,948 (34.3%), timeliness of care accounted for $60,413,496 (27.3%), and best medical interest accounted for $46,291,390 (21.0%) (Table 4).

Overall costs were primarily driven by specific VISNs within each eligibility type (Appendix 8; Figure 5). VISN 19, VISN 22, and VISN 15 accounted for the highest referral costs for distance; VISN 22, VISN 8, and VISN 19 accounted for the secondhighest referral cost, timeliness of care; and VISN 4, VISN 8, and VISN 12 accounted for the third-highest referral cost, best medical interest (Figure 5). VISN 2, VISN 4, VISN 12, VISN 15, and VISN 22 had service unavailable as an eligibility type with 1 of the top 3 associated referral costs, which was higher in cost than timeliness of care for VISN 2, VISN 4, VISN 12, and VISN 15.

Discussion

This study examines the referral of rehabilitation PT services to CC, evaluates CC costs for PT services, and analyzes utilization and cost trends among veterans within the VHA. Utilization data demonstrated a decrease in referrals from FY 2019 to FY 2020 and increases in referrals from FY 2020 to FY 2022 for most variables of interest, with cost data exhibiting similar trends. Results highlight the need for further investigation to address variations in PT referrals and costs across VISNs and eligibility reasons for CC referral.

Results demonstrated a noteworthy increase in PT CC referrals over time. The largest increase occurred from FY 2020 to FY 2021, with a smaller increase from FY 2021 to FY 2022. During this period, total enrollee numbers decreased by 3.0% across the 7 VISNs included in this analysis and by 1.6% across all VISNs, a trend that illustrates an overall decrease in enrollees as CC use increased. Results align with the implementation of the MISSION Act of 2018, which further expanded veterans’ options to use CC.^1,6,7 Results also align with the onset of the COVID-19 pandemic, which disrupted care access for many veterans, placed a larger emphasis on the use of telehealth, and increased opportunities to stay within the VA for care by rapidly shifting to telehealth and leveraging telerehabilitation investments and initiatives (such as TR-EWI).^20,31

VISN 8, VISN 19, and VISN 22, accounted for more than half of PT referrals. These VISNs had higher enrollee counts compared to the other VISNs.³² VISN 8 consistently had high levels of referrals, whereas VISN 19 and VISN 22 saw dramatic increases in FY 2021 and FY 2022. In contrast, VISN 4 and VISN 12 gradually decreased referrals during the study. VISN 2 had the lowest referral numbers during the study period, and all stations with the lowest individual referral numbers were located within VISN 2. Of the VISNs included in this study, VISN 2 had the second lowest number of enrollees (324,042).³² Reasons for increases and decreases over time could not be determined based on data collected in this study.

There were more urban than rural PT CC referrals; however, both exhibited an increase in referrals over time. This is consistent with population trends showing that most VHA patients (62.6%) and veterans (75.9%) reside in urban areas, which could explain some of the trends in this study.³³ Some VISNs have larger urban catchment areas (eg, VISN 8 and VISN 22), and some have larger rural catchment areas (eg, VISN 15 and VISN 19), which could partially explain the rural-urban differences by VISN.³² Rural-urban referral trends might also reflect existing health care delivery system deficits in rural areas and known challenges associated with accessing health care for veterans living in rural communities.^8,9

This study found larger differences in rural and urban PT CC referrals for younger age groups, with more than twice as many urban referrals in veterans aged 20 to 29 years and aged 30 to 39 years, and roughly 1.8 times as many urban referrals in veterans aged 40 to 49 years. However, there were similar numbers of rural and urban referrals in those aged 70 to 79 years and aged 80 to 89 years. These trends are consistent with data showing veterans residing in rural communities are older than their urban counterparts.^23,34 Data suggest that older veteran populations might seek PT at higher rates than younger veteran populations. Moreover, data suggest there could be differences in PT-seeking rates for younger veteran populations who reside in rural vs urban areas. Additional research is needed to understand these trends.

Distance and timeliness of care were the predominant reasons for referral among eligibility groups, which is consistent with the MISSION Act goals.^1,6,7 The most common eligibility reason for rural referrals was distance; timeliness of care was most common for urban referrals. This finding is expected, as veterans living in rural communities are farther away from VHA facilities and have longer drive times, whereas veterans living in urban communities might live closer, yet experience longer wait times due to services and/or appointment availability. Best medical interest accounted for almost 20% of referrals, which does not provide detailed insights into why those veterans were referred to CC.

The top PT diagnoses referred to CC were related to bone, joint, or soft tissue disorders of the lower back, shoulder, and knee. This suggests that musculoskeletal-related issues are prevalent among veterans seeking PT care, which is consistent with research that found > 50% of veterans receiving VHA care have musculoskeletal disorders.³⁵ The probability of experiencing musculoskeletal problems increases with age, as does the need for PT services. Amputations and fractures accounted for < 1% of CC referrals, which is consistent with the historic provision of VHA clinical specialized care to conditions prevalent among veterans. It may also represent VHA efforts to internally provide care for complex conditions requiring more extensive interdisciplinary coordination.

The total cost of referrals over time was about $221 million. VISN 8 accounted for the highest overall cost; VISN 2 had the lowest, mirroring referral utilization trends and aligning with VISN enrollee numbers. VISN 19 and VISN 22 reported large cost increases from FY 2020 to FY 2021. Total referral costs increased by $34.9 million from FY 2020 to FY 2021, which may be due to health care inflation (2.9% during FY 2019 to FY 2022), increased awareness of CC services, or increased VHA wait times.³⁶ Additionally, there were limitations in care provided across health care systems during the COVID-19 pandemic, including the VA.⁵ The increase from FY 2020 to FY 2021 may reflect a rebound from restrictions in appointments across VA, CC, and the private sector.

While the increase in total referral cost may be partly attributed to inflation, the cost effectiveness and efficiency of referring veterans to CC vs keeping veterans within VHA care is an ongoing debate.⁵ Examining and addressing cost drivers within the top eligibility types and their respective VISNs is necessary to determine resource allocation and improve quality of care. This study found that best medical interest and unavailable services accounted for 33.4% of the total cost of CC referrals, highlighting the need for policies that strengthen in-house competencies and recruit personnel to provide PT services currently unavailable within the VA.

Future Directions

The VHA should explore opportunities for in-house care, especially for services appropriate for telehealth.^18,20,37 Data indicated a smaller cost increase from FY 2021 to FY 2022 compared to the relatively large increase from FY 2020 to FY 2021. The increased telehealth usage across VHA by TR-EWI and non—TR-EWI sites within selected VISNs may have contributed to limiting the increase in CC costs. Future studies should investigate contextual factors of increased telehealth usage, which would offer guidance for implementation to optimize the integration of telehealth with PT rehabilitation provided in-house. Additionally, future studies can examine potential limitations experienced during PT telehealth visits, such as the inability to conduct hands-on assessments, challenges in viewing the quality of patient movement, ensuring patient safety in the remote environment, and the lack of PT equipment in homes for telehealth visits, and how these challenges are being addressed.^38,39 Research is also needed to understand tradeoffs of CC vs VHA care and the potential and cost benefits of keeping veterans within VHA using programs like TR-EWI.⁵ Veterans living in rural communities may especially benefit from this as expanding telehealth options can provide access to PT care that may not be readily available, enabling them to stay connected and engaged in their care.^18,40

Future studies could examine contributory factors to rising costs, such as demographic shifts, changes in PT service utilization, and policy. Researchers might also consider qualitative studies with clinicians and veterans within each VISN, which may provide insights into how local factors impact PT referral to the community.

Limitations

Due to its descriptive nature, this study can only speculate about factors influencing trends. Limitations include the inability to link the Palantir and CC Dashboard datasets for cost comparisons and potential data change over time on Palantir due to platform updates. The focus on VISNs with TREWI sites limited generalizability and this study did not compare CC PT vs VHA PT. Finally, there may have been cost drivers not identified in this study.

Conclusions

This descriptive study provides insights into the utilization and cost of PT CC referrals for selected VISNs. Cost trends underscore the financial commitment to providing PT services to veterans. Understanding what factors are driving this cost is necessary for VHA to optimally provide and manage the rehabilitation resources needed to serve veterans through traditional in-person care, telehealth, and CC options while ensuring timely, highquality care.

This article is part of a series of articles on the Home Improvements and Structural Alterations program (HISA), a home modification (HM) program within the Veterans Health Administration (VHA). HISA is a benefit awarded to veterans with disabilities (VWDs) and is instrumental in affording physical accessibility and structural alterations to veterans’ homes.¹ The overarching goals of this project are to describe and understand HISA use by VWDs. Previous work has shown geographical variation in the number of HISA prescriptions across counties in the US (Figure 1).¹ The current work seeks to describe and predict the county-level rates of HISA use. Information about what predicts HISA utilization at the county level is important because it enhances understanding of program utilization at a national level. The long-term goal of the series is to provide knowledge about HM services within VHA to improve community-based independent living of VWDs by increasing awareness and utilization of HM services. 

Background

A health care professional (HCP) approves a HM support award by evaluating the practicality of the support to improve the built environment of a given veteran’s disability.^1,2 Previously we detailed some of the preliminary research into the HISA program, including HISA user demographic and clinical characteristics, types of HMs received, user suggestions for improvement, and geospatial analysis of HISA prescriptions concentration.^1-4

The geospatial analyses of HISA prescriptions revealed clusters of high numbers of HISA users (hot spots) and low numbers of HISA users (cold spots), indicating that HISA is either not prescribed or uniformly used across the US. The previous research prompted investigation into county-level variables that may impact HISA utilization rates. This inquiry focuses on county characteristics associated with HISA use rates, such as measures of clinical care and quality of care (eg, access to health services variables, lack of insurance, preventable hospital stays), physical environment, and sociodemographic characteristics. Clinical care and quality of care measures promote the interaction with HCPs. Moreover, access to health care is an important indicator of health outcomes.^5,6 An individual’s capacity to access health services, such as a HM program, greatly impacts well-being, safety, independence, and health.^2,4 Well-being, safety, independence, and health become compromised if individuals cannot access care, if needed care is lacking in their area, if HCPs are not available, or are unwilling to provide care due to lack of insurance coverage.^7-12 In locations where health care services are minimal due to lack of specialists or health care facilities, the quality of (or access to) care may be compromised, resulting in preventable conditions becoming problematic.^13,14 These conditions may result in unnecessary hospitalizations for conditions that could have been treated during routine care. Financial barriers to care particularly among low-income people and the uninsured have proven detrimental to health.^15,16 On the other hand, preventable hospital stays are a quality of care measure (ie, a proxy for poor quality of care). HISA operates within a health care system; thus, it is imperative to include these measures impacting health. 

In this study, we sought to identify county-level predictors—in particular, county-level proxies for access to care—that may be associated with county-level HISA use. We define HISA utilization rate as the percentage of a county’s VHA patients who have received a HISA award.

Methods

This study used data from the National Prosthetics Patient Database (NPPD), US Department of Veterans Affairs (VA) medical database inpatient and outpatient datasets, VHA Support Service Center (VSSC) data cubes, and the County Health Rankings database (CHRD). First, the study cohort was identified from NPPD users who have obtained a HISA award from fiscal years (FY) 2015 to 2018. Analysis started with FY 2015 following new regulations (38 CFR § 17) governing the operations of the HISA program.² The study cohort was matched with records from NPPD and VA inpatient and outpatient datasets to obtain information about the veterans’ demographic characteristics and their HM characteristics and costs. The number of VHA end-of-year (EOY) patients per county was extracted from the VSSC Current Enrollment Cube, which was used in calculation of the county-level HISA utilization rate.¹⁷ Finally, zip code–based locational data were used to calculate approximate drive time and distance from the HISA user’s approximate location to the facility where they received their HM prescription. Drive times and drive distances were calculated with Esri ArcGIS Pro (v2.6.3) by placing zip code centroid and VHA facilities on a nationwide road network that contains both road speeds and distances.

Calculations

Patient-level data were aggregated up to county-level variables by calculating the sum, mean, or percent per county. HISA user sample characteristics, including sex, race, rurality (urban, rural), marital status, and Class 1 vs Class 2 disability-related eligibility groups, were aggregated to the county level by calculating percentages of HISA users of the given characteristics out of total HISA users in the county. Disability-related eligibility groups (Class 1 vs Class 2 HISA users) determines the maximum lifetime award dollar amount. Specifically, those with service-connected disabilities or those with a ≥ 50% disability rating (regardless of whether or not their disability is service connected) are classified as Class 1 HISA users and are eligible to receive a maximum lifetime award of $6800. Those with a recorded disability that is not connected to their military service, and who have a disability rating of < 50% are classified as Class 2 HISA users and are eligible to receive a lifetime maximum award of $2000.

The county-level number of HISA users was used as the numerator for calculation of county-level HISA utilization rate. Counties with zero HISA users were excluded. The number of EOY VHA patients per county in FY 2018 was divided by 1000 and used as the denominator in the calculation of county-level HISA utilization rate. Thus, the outcome variable is HISA utilization rate per 1000 VHA patients in FY 2018 (HISA utilization rate).

County-Level Variables

County-level variables were downloaded from the 2020 CHRD.^5,6 An explanation of the CHRD model and the factors used in this study are shown in the eAppendix (available at doi: 10.12788/fp.0279).⁶ County-level aggregated HISA user data and the CHRD data were matched using county Federal Information Processing Standards codes. Access to care measures collected from CHRD included percentages uninsured and ratios of population to primary care physicians, dentists, mental health professionals, and other primary care professionals. Other CHRD measures included those for quality of care (rate of preventable hospital stay) and housing quality (percent of households with high housing costs, percent of households with overcrowding, percent of households with lack of kitchen or plumbing, percent of households with severe housing cost burden, percent of homeownership). Of secondary interest was county population rurality, as previous research findings showed the annual average of HISA users who are from rural areas ranged from 30 to 35%.

Analysis Methods

SAS (v9.4), R (v4.0.2), and RStudio (v1.3.1093) were used for data preparation and analysis.¹⁸ Multiple regression analysis was used to predict county-level utilization rate from county-level variables. Sociodemographic characteristics from CHRD and HISA data were included as important control predictors in the regression model, though our focus for this paper are the access to care and housing quality predictors.

Model diagnostics (examination of model residuals, Breusch-Godfrey test, Breusch-Pagan test) revealed significant heteroskedasticity of the model; thus, robust standard errors and associated P values were computed using the R estimatr package (v0.30.2).¹⁹ Some predictor variables of interest (eg, ratio of mental health professionals) were removed during the model building process either due to problems of multicollinearity or excessive missingness that would have resulted in listwise deletion.

Results

County-level HISA utilization rate per 1000 EOY VHA patients ranged from 0.09 to 59.7%, with a 6.6% mean and 5% median (Figure 2). The data were highly positively skewed. The final model included 33 total predictor variables (Table 1). The final regression model was a significantly better predictor of county-level HISA utilization rate than a null model (F[33-2184], 10.18; P < .001). The adjusted model R² showed that the overall model accounted for approximately 23% of variance in county-level HISA utilization rate (Table 2).

Among the primary variables of interest, percent uninsured adults and rate of preventable hospital stays emerged as significant predictors of county-level HISA utilization rate. Specifically, county percentage of uninsured adults was negatively related to county-level HISA utilization rate (b = -8.99, P = .005), indicating that the higher the proportion of uninsured adults—with all other predictors held constant—the lower the HISA utilization rate. Percent uninsured adults ranged from 2.7 to 42.4% across counties, with a mean (SD) of 12.7% (5.8%) and 11.4% median.

Discussion

This study examined whether county-level characteristics, specifically variables for access to care, quality of care, and housing quality, were predictive of a county’s HISA utilization rate. Given that this series of work on the HISA program is (to our knowledge) the first of its kind, and given the exploratory nature of this analysis, we did not have specific predictions for the effects of any one given variable. Nevertheless, some of the results were surprising, and we believe they warrant additional study. In particular, the opposing direction of effects for access to care and quality of care variables were hard to reconcile.

The county percent of uninsured adults (an access to care variable, specifically, a proxy for poor access to care) was negatively associated with county-level HISA utilization rate, whereas the county rate of preventable hospital stays (a quality of care variable, but also potentially an access to care variable, and specifically, proxies for poor quality of care or poor access to care) was positively associated with county-level HISA utilization rate. To describe the relationships more generally, one coefficient in the regression model indicated that the poorer the access to care, the lower the HISA utilization rate (higher percent of uninsured adults predicts lower HISA utilization rate), while another coefficient in the regression model indicated the poorer the quality of and access to care, the higher the HISA utilization rate (higher rate of preventable hospital stays predicts higher HISA utilization rate). Future study is warranted to disentangle and reconcile the various community-level predictors of this service.

Housing quality measures (eg, percent of households with high housing costs, percent of households with overcrowding, percent of households with lack of kitchen or plumbing, percent of households with severe housing cost burden, and percent of homeownership) are important in the consideration of whether a HM will be performed or should be performed. For example, if a person is cost burdened by the amount of expenditure spent in housing there will be little discretionary funds to perform a HM. Individuals who do not own their home may experience complications in obtaining permission from landlords to perform a HM. County-level predictors of housing quality (percent of households with high housing costs, overcrowding, and lack of kitchen or plumbing) were not significantly associated with county-level HISA utilization rate but are also nevertheless relevant to the discussion of home modifications. Of particular interest is the percent of households with lack of kitchen or plumbing variable, which was positively related to county-level HISA utilization rate although not statistically significant. HM elements related to the kitchen (eg, heighten countertop) add to the accessibility of the home allowing for the performing of activities of daily living such as cooking. Between FY 2015 and FY 2018, we discovered 131 prescriptions for kitchen (n = 90) and plumbing (n = 41) HMs, which is a very small proportion of the 30,780 total HMs (there were 24,397 bathroom HMs). The nonsignificant coefficient for this variable may reflect the small number of veterans that obtained these HM.

Limitations

The potentially conflicting direction of effects for a significant access to care variable (percent uninsured adults) and a significant access to care and quality of care variable (preventable hospital stays) are interesting and warrant additional study, but the inability to interpret or explain this apparent inconsistency constitutes a limitation of the current data and analyses presented here. Another limitation is that this analysis uses county-level predictors for what is ultimately an individual-level outcome. It would have been ideal to have both individual- and county-level data to conduct a multilevel analysis; in particular, individual-level data within counties of individuals (both veterans and nonveterans) who did not receive a HISA award (including both those who applied and were denied, and who did not apply) would be highly valuable.

Conclusions

Our continuing research into veterans’ use of HM fills a gap in the literature about the characteristics of HISA users, the impact of county-level variables on the use of HISA, and the geographic distribution and use of HISA within the VHA. While it is important to examine the influence of broader systems on individual outcomes, there could be myriad other factors that are more proximal and more closely related to whether any one individual applies for, let alone receives, a HISA award. Indeed, a low overall adjusted model R² indicates that there is considerable variability in county-level HISA utilization rate that was not accounted for by the current model; this further speaks to warranted additional study.

More research is needed to understand and account for geographical variation in HISA utilization rate across the US. However, this work serves as an exploratory first step at quantifying and predicting HISA utilization rate at a broad level, with the ultimate goal of increasing access to HMs for veterans with disabilities.

Acknowledgments

This research was supported by grant 15521 from the US Department of Veterans Affairs, Office of Rural Health. Furthermore, the research was supported in part by grant K12 HD055929 from the National Institutes of Health. We want to acknowledge Cheri E. Knecht, Project Coordinator, for her assistance throughout all aspects of our research study and for her thoughtful contributions during the writing of this manuscript.

This article is part of a series of articles on the Home Improvements and Structural Alterations program (HISA), a home modification (HM) program within the Veterans Health Administration (VHA). HISA is a benefit awarded to veterans with disabilities (VWDs) and is instrumental in affording physical accessibility and structural alterations to veterans’ homes.¹ The overarching goals of this project are to describe and understand HISA use by VWDs. Previous work has shown geographical variation in the number of HISA prescriptions across counties in the US (Figure 1).¹ The current work seeks to describe and predict the county-level rates of HISA use. Information about what predicts HISA utilization at the county level is important because it enhances understanding of program utilization at a national level. The long-term goal of the series is to provide knowledge about HM services within VHA to improve community-based independent living of VWDs by increasing awareness and utilization of HM services. 

Background

A health care professional (HCP) approves a HM support award by evaluating the practicality of the support to improve the built environment of a given veteran’s disability.^1,2 Previously we detailed some of the preliminary research into the HISA program, including HISA user demographic and clinical characteristics, types of HMs received, user suggestions for improvement, and geospatial analysis of HISA prescriptions concentration.^1-4

The geospatial analyses of HISA prescriptions revealed clusters of high numbers of HISA users (hot spots) and low numbers of HISA users (cold spots), indicating that HISA is either not prescribed or uniformly used across the US. The previous research prompted investigation into county-level variables that may impact HISA utilization rates. This inquiry focuses on county characteristics associated with HISA use rates, such as measures of clinical care and quality of care (eg, access to health services variables, lack of insurance, preventable hospital stays), physical environment, and sociodemographic characteristics. Clinical care and quality of care measures promote the interaction with HCPs. Moreover, access to health care is an important indicator of health outcomes.^5,6 An individual’s capacity to access health services, such as a HM program, greatly impacts well-being, safety, independence, and health.^2,4 Well-being, safety, independence, and health become compromised if individuals cannot access care, if needed care is lacking in their area, if HCPs are not available, or are unwilling to provide care due to lack of insurance coverage.^7-12 In locations where health care services are minimal due to lack of specialists or health care facilities, the quality of (or access to) care may be compromised, resulting in preventable conditions becoming problematic.^13,14 These conditions may result in unnecessary hospitalizations for conditions that could have been treated during routine care. Financial barriers to care particularly among low-income people and the uninsured have proven detrimental to health.^15,16 On the other hand, preventable hospital stays are a quality of care measure (ie, a proxy for poor quality of care). HISA operates within a health care system; thus, it is imperative to include these measures impacting health. 

In this study, we sought to identify county-level predictors—in particular, county-level proxies for access to care—that may be associated with county-level HISA use. We define HISA utilization rate as the percentage of a county’s VHA patients who have received a HISA award.

Methods

This study used data from the National Prosthetics Patient Database (NPPD), US Department of Veterans Affairs (VA) medical database inpatient and outpatient datasets, VHA Support Service Center (VSSC) data cubes, and the County Health Rankings database (CHRD). First, the study cohort was identified from NPPD users who have obtained a HISA award from fiscal years (FY) 2015 to 2018. Analysis started with FY 2015 following new regulations (38 CFR § 17) governing the operations of the HISA program.² The study cohort was matched with records from NPPD and VA inpatient and outpatient datasets to obtain information about the veterans’ demographic characteristics and their HM characteristics and costs. The number of VHA end-of-year (EOY) patients per county was extracted from the VSSC Current Enrollment Cube, which was used in calculation of the county-level HISA utilization rate.¹⁷ Finally, zip code–based locational data were used to calculate approximate drive time and distance from the HISA user’s approximate location to the facility where they received their HM prescription. Drive times and drive distances were calculated with Esri ArcGIS Pro (v2.6.3) by placing zip code centroid and VHA facilities on a nationwide road network that contains both road speeds and distances.

Calculations

Patient-level data were aggregated up to county-level variables by calculating the sum, mean, or percent per county. HISA user sample characteristics, including sex, race, rurality (urban, rural), marital status, and Class 1 vs Class 2 disability-related eligibility groups, were aggregated to the county level by calculating percentages of HISA users of the given characteristics out of total HISA users in the county. Disability-related eligibility groups (Class 1 vs Class 2 HISA users) determines the maximum lifetime award dollar amount. Specifically, those with service-connected disabilities or those with a ≥ 50% disability rating (regardless of whether or not their disability is service connected) are classified as Class 1 HISA users and are eligible to receive a maximum lifetime award of $6800. Those with a recorded disability that is not connected to their military service, and who have a disability rating of < 50% are classified as Class 2 HISA users and are eligible to receive a lifetime maximum award of $2000.

The county-level number of HISA users was used as the numerator for calculation of county-level HISA utilization rate. Counties with zero HISA users were excluded. The number of EOY VHA patients per county in FY 2018 was divided by 1000 and used as the denominator in the calculation of county-level HISA utilization rate. Thus, the outcome variable is HISA utilization rate per 1000 VHA patients in FY 2018 (HISA utilization rate).

County-Level Variables

County-level variables were downloaded from the 2020 CHRD.^5,6 An explanation of the CHRD model and the factors used in this study are shown in the eAppendix (available at doi: 10.12788/fp.0279).⁶ County-level aggregated HISA user data and the CHRD data were matched using county Federal Information Processing Standards codes. Access to care measures collected from CHRD included percentages uninsured and ratios of population to primary care physicians, dentists, mental health professionals, and other primary care professionals. Other CHRD measures included those for quality of care (rate of preventable hospital stay) and housing quality (percent of households with high housing costs, percent of households with overcrowding, percent of households with lack of kitchen or plumbing, percent of households with severe housing cost burden, percent of homeownership). Of secondary interest was county population rurality, as previous research findings showed the annual average of HISA users who are from rural areas ranged from 30 to 35%.

Analysis Methods

SAS (v9.4), R (v4.0.2), and RStudio (v1.3.1093) were used for data preparation and analysis.¹⁸ Multiple regression analysis was used to predict county-level utilization rate from county-level variables. Sociodemographic characteristics from CHRD and HISA data were included as important control predictors in the regression model, though our focus for this paper are the access to care and housing quality predictors.

Model diagnostics (examination of model residuals, Breusch-Godfrey test, Breusch-Pagan test) revealed significant heteroskedasticity of the model; thus, robust standard errors and associated P values were computed using the R estimatr package (v0.30.2).¹⁹ Some predictor variables of interest (eg, ratio of mental health professionals) were removed during the model building process either due to problems of multicollinearity or excessive missingness that would have resulted in listwise deletion.

Results

County-level HISA utilization rate per 1000 EOY VHA patients ranged from 0.09 to 59.7%, with a 6.6% mean and 5% median (Figure 2). The data were highly positively skewed. The final model included 33 total predictor variables (Table 1). The final regression model was a significantly better predictor of county-level HISA utilization rate than a null model (F[33-2184], 10.18; P < .001). The adjusted model R² showed that the overall model accounted for approximately 23% of variance in county-level HISA utilization rate (Table 2).

Among the primary variables of interest, percent uninsured adults and rate of preventable hospital stays emerged as significant predictors of county-level HISA utilization rate. Specifically, county percentage of uninsured adults was negatively related to county-level HISA utilization rate (b = -8.99, P = .005), indicating that the higher the proportion of uninsured adults—with all other predictors held constant—the lower the HISA utilization rate. Percent uninsured adults ranged from 2.7 to 42.4% across counties, with a mean (SD) of 12.7% (5.8%) and 11.4% median.

Discussion

This study examined whether county-level characteristics, specifically variables for access to care, quality of care, and housing quality, were predictive of a county’s HISA utilization rate. Given that this series of work on the HISA program is (to our knowledge) the first of its kind, and given the exploratory nature of this analysis, we did not have specific predictions for the effects of any one given variable. Nevertheless, some of the results were surprising, and we believe they warrant additional study. In particular, the opposing direction of effects for access to care and quality of care variables were hard to reconcile.

The county percent of uninsured adults (an access to care variable, specifically, a proxy for poor access to care) was negatively associated with county-level HISA utilization rate, whereas the county rate of preventable hospital stays (a quality of care variable, but also potentially an access to care variable, and specifically, proxies for poor quality of care or poor access to care) was positively associated with county-level HISA utilization rate. To describe the relationships more generally, one coefficient in the regression model indicated that the poorer the access to care, the lower the HISA utilization rate (higher percent of uninsured adults predicts lower HISA utilization rate), while another coefficient in the regression model indicated the poorer the quality of and access to care, the higher the HISA utilization rate (higher rate of preventable hospital stays predicts higher HISA utilization rate). Future study is warranted to disentangle and reconcile the various community-level predictors of this service.

Housing quality measures (eg, percent of households with high housing costs, percent of households with overcrowding, percent of households with lack of kitchen or plumbing, percent of households with severe housing cost burden, and percent of homeownership) are important in the consideration of whether a HM will be performed or should be performed. For example, if a person is cost burdened by the amount of expenditure spent in housing there will be little discretionary funds to perform a HM. Individuals who do not own their home may experience complications in obtaining permission from landlords to perform a HM. County-level predictors of housing quality (percent of households with high housing costs, overcrowding, and lack of kitchen or plumbing) were not significantly associated with county-level HISA utilization rate but are also nevertheless relevant to the discussion of home modifications. Of particular interest is the percent of households with lack of kitchen or plumbing variable, which was positively related to county-level HISA utilization rate although not statistically significant. HM elements related to the kitchen (eg, heighten countertop) add to the accessibility of the home allowing for the performing of activities of daily living such as cooking. Between FY 2015 and FY 2018, we discovered 131 prescriptions for kitchen (n = 90) and plumbing (n = 41) HMs, which is a very small proportion of the 30,780 total HMs (there were 24,397 bathroom HMs). The nonsignificant coefficient for this variable may reflect the small number of veterans that obtained these HM.

Limitations

The potentially conflicting direction of effects for a significant access to care variable (percent uninsured adults) and a significant access to care and quality of care variable (preventable hospital stays) are interesting and warrant additional study, but the inability to interpret or explain this apparent inconsistency constitutes a limitation of the current data and analyses presented here. Another limitation is that this analysis uses county-level predictors for what is ultimately an individual-level outcome. It would have been ideal to have both individual- and county-level data to conduct a multilevel analysis; in particular, individual-level data within counties of individuals (both veterans and nonveterans) who did not receive a HISA award (including both those who applied and were denied, and who did not apply) would be highly valuable.

Conclusions

Our continuing research into veterans’ use of HM fills a gap in the literature about the characteristics of HISA users, the impact of county-level variables on the use of HISA, and the geographic distribution and use of HISA within the VHA. While it is important to examine the influence of broader systems on individual outcomes, there could be myriad other factors that are more proximal and more closely related to whether any one individual applies for, let alone receives, a HISA award. Indeed, a low overall adjusted model R² indicates that there is considerable variability in county-level HISA utilization rate that was not accounted for by the current model; this further speaks to warranted additional study.

More research is needed to understand and account for geographical variation in HISA utilization rate across the US. However, this work serves as an exploratory first step at quantifying and predicting HISA utilization rate at a broad level, with the ultimate goal of increasing access to HMs for veterans with disabilities.

Acknowledgments

This research was supported by grant 15521 from the US Department of Veterans Affairs, Office of Rural Health. Furthermore, the research was supported in part by grant K12 HD055929 from the National Institutes of Health. We want to acknowledge Cheri E. Knecht, Project Coordinator, for her assistance throughout all aspects of our research study and for her thoughtful contributions during the writing of this manuscript.

This article is part of a series of articles on the Home Improvements and Structural Alterations program (HISA), a home modification (HM) program within the Veterans Health Administration (VHA). HISA is a benefit awarded to veterans with disabilities (VWDs) and is instrumental in affording physical accessibility and structural alterations to veterans’ homes.¹ The overarching goals of this project are to describe and understand HISA use by VWDs. Previous work has shown geographical variation in the number of HISA prescriptions across counties in the US (Figure 1).¹ The current work seeks to describe and predict the county-level rates of HISA use. Information about what predicts HISA utilization at the county level is important because it enhances understanding of program utilization at a national level. The long-term goal of the series is to provide knowledge about HM services within VHA to improve community-based independent living of VWDs by increasing awareness and utilization of HM services. 

Background

A health care professional (HCP) approves a HM support award by evaluating the practicality of the support to improve the built environment of a given veteran’s disability.^1,2 Previously we detailed some of the preliminary research into the HISA program, including HISA user demographic and clinical characteristics, types of HMs received, user suggestions for improvement, and geospatial analysis of HISA prescriptions concentration.^1-4

The geospatial analyses of HISA prescriptions revealed clusters of high numbers of HISA users (hot spots) and low numbers of HISA users (cold spots), indicating that HISA is either not prescribed or uniformly used across the US. The previous research prompted investigation into county-level variables that may impact HISA utilization rates. This inquiry focuses on county characteristics associated with HISA use rates, such as measures of clinical care and quality of care (eg, access to health services variables, lack of insurance, preventable hospital stays), physical environment, and sociodemographic characteristics. Clinical care and quality of care measures promote the interaction with HCPs. Moreover, access to health care is an important indicator of health outcomes.^5,6 An individual’s capacity to access health services, such as a HM program, greatly impacts well-being, safety, independence, and health.^2,4 Well-being, safety, independence, and health become compromised if individuals cannot access care, if needed care is lacking in their area, if HCPs are not available, or are unwilling to provide care due to lack of insurance coverage.^7-12 In locations where health care services are minimal due to lack of specialists or health care facilities, the quality of (or access to) care may be compromised, resulting in preventable conditions becoming problematic.^13,14 These conditions may result in unnecessary hospitalizations for conditions that could have been treated during routine care. Financial barriers to care particularly among low-income people and the uninsured have proven detrimental to health.^15,16 On the other hand, preventable hospital stays are a quality of care measure (ie, a proxy for poor quality of care). HISA operates within a health care system; thus, it is imperative to include these measures impacting health. 

In this study, we sought to identify county-level predictors—in particular, county-level proxies for access to care—that may be associated with county-level HISA use. We define HISA utilization rate as the percentage of a county’s VHA patients who have received a HISA award.

Methods

This study used data from the National Prosthetics Patient Database (NPPD), US Department of Veterans Affairs (VA) medical database inpatient and outpatient datasets, VHA Support Service Center (VSSC) data cubes, and the County Health Rankings database (CHRD). First, the study cohort was identified from NPPD users who have obtained a HISA award from fiscal years (FY) 2015 to 2018. Analysis started with FY 2015 following new regulations (38 CFR § 17) governing the operations of the HISA program.² The study cohort was matched with records from NPPD and VA inpatient and outpatient datasets to obtain information about the veterans’ demographic characteristics and their HM characteristics and costs. The number of VHA end-of-year (EOY) patients per county was extracted from the VSSC Current Enrollment Cube, which was used in calculation of the county-level HISA utilization rate.¹⁷ Finally, zip code–based locational data were used to calculate approximate drive time and distance from the HISA user’s approximate location to the facility where they received their HM prescription. Drive times and drive distances were calculated with Esri ArcGIS Pro (v2.6.3) by placing zip code centroid and VHA facilities on a nationwide road network that contains both road speeds and distances.

Calculations

Patient-level data were aggregated up to county-level variables by calculating the sum, mean, or percent per county. HISA user sample characteristics, including sex, race, rurality (urban, rural), marital status, and Class 1 vs Class 2 disability-related eligibility groups, were aggregated to the county level by calculating percentages of HISA users of the given characteristics out of total HISA users in the county. Disability-related eligibility groups (Class 1 vs Class 2 HISA users) determines the maximum lifetime award dollar amount. Specifically, those with service-connected disabilities or those with a ≥ 50% disability rating (regardless of whether or not their disability is service connected) are classified as Class 1 HISA users and are eligible to receive a maximum lifetime award of $6800. Those with a recorded disability that is not connected to their military service, and who have a disability rating of < 50% are classified as Class 2 HISA users and are eligible to receive a lifetime maximum award of $2000.

The county-level number of HISA users was used as the numerator for calculation of county-level HISA utilization rate. Counties with zero HISA users were excluded. The number of EOY VHA patients per county in FY 2018 was divided by 1000 and used as the denominator in the calculation of county-level HISA utilization rate. Thus, the outcome variable is HISA utilization rate per 1000 VHA patients in FY 2018 (HISA utilization rate).

County-Level Variables

County-level variables were downloaded from the 2020 CHRD.^5,6 An explanation of the CHRD model and the factors used in this study are shown in the eAppendix (available at doi: 10.12788/fp.0279).⁶ County-level aggregated HISA user data and the CHRD data were matched using county Federal Information Processing Standards codes. Access to care measures collected from CHRD included percentages uninsured and ratios of population to primary care physicians, dentists, mental health professionals, and other primary care professionals. Other CHRD measures included those for quality of care (rate of preventable hospital stay) and housing quality (percent of households with high housing costs, percent of households with overcrowding, percent of households with lack of kitchen or plumbing, percent of households with severe housing cost burden, percent of homeownership). Of secondary interest was county population rurality, as previous research findings showed the annual average of HISA users who are from rural areas ranged from 30 to 35%.

Analysis Methods

SAS (v9.4), R (v4.0.2), and RStudio (v1.3.1093) were used for data preparation and analysis.¹⁸ Multiple regression analysis was used to predict county-level utilization rate from county-level variables. Sociodemographic characteristics from CHRD and HISA data were included as important control predictors in the regression model, though our focus for this paper are the access to care and housing quality predictors.

Model diagnostics (examination of model residuals, Breusch-Godfrey test, Breusch-Pagan test) revealed significant heteroskedasticity of the model; thus, robust standard errors and associated P values were computed using the R estimatr package (v0.30.2).¹⁹ Some predictor variables of interest (eg, ratio of mental health professionals) were removed during the model building process either due to problems of multicollinearity or excessive missingness that would have resulted in listwise deletion.

Results

County-level HISA utilization rate per 1000 EOY VHA patients ranged from 0.09 to 59.7%, with a 6.6% mean and 5% median (Figure 2). The data were highly positively skewed. The final model included 33 total predictor variables (Table 1). The final regression model was a significantly better predictor of county-level HISA utilization rate than a null model (F[33-2184], 10.18; P < .001). The adjusted model R² showed that the overall model accounted for approximately 23% of variance in county-level HISA utilization rate (Table 2).

Among the primary variables of interest, percent uninsured adults and rate of preventable hospital stays emerged as significant predictors of county-level HISA utilization rate. Specifically, county percentage of uninsured adults was negatively related to county-level HISA utilization rate (b = -8.99, P = .005), indicating that the higher the proportion of uninsured adults—with all other predictors held constant—the lower the HISA utilization rate. Percent uninsured adults ranged from 2.7 to 42.4% across counties, with a mean (SD) of 12.7% (5.8%) and 11.4% median.

Discussion

This study examined whether county-level characteristics, specifically variables for access to care, quality of care, and housing quality, were predictive of a county’s HISA utilization rate. Given that this series of work on the HISA program is (to our knowledge) the first of its kind, and given the exploratory nature of this analysis, we did not have specific predictions for the effects of any one given variable. Nevertheless, some of the results were surprising, and we believe they warrant additional study. In particular, the opposing direction of effects for access to care and quality of care variables were hard to reconcile.

The county percent of uninsured adults (an access to care variable, specifically, a proxy for poor access to care) was negatively associated with county-level HISA utilization rate, whereas the county rate of preventable hospital stays (a quality of care variable, but also potentially an access to care variable, and specifically, proxies for poor quality of care or poor access to care) was positively associated with county-level HISA utilization rate. To describe the relationships more generally, one coefficient in the regression model indicated that the poorer the access to care, the lower the HISA utilization rate (higher percent of uninsured adults predicts lower HISA utilization rate), while another coefficient in the regression model indicated the poorer the quality of and access to care, the higher the HISA utilization rate (higher rate of preventable hospital stays predicts higher HISA utilization rate). Future study is warranted to disentangle and reconcile the various community-level predictors of this service.

Housing quality measures (eg, percent of households with high housing costs, percent of households with overcrowding, percent of households with lack of kitchen or plumbing, percent of households with severe housing cost burden, and percent of homeownership) are important in the consideration of whether a HM will be performed or should be performed. For example, if a person is cost burdened by the amount of expenditure spent in housing there will be little discretionary funds to perform a HM. Individuals who do not own their home may experience complications in obtaining permission from landlords to perform a HM. County-level predictors of housing quality (percent of households with high housing costs, overcrowding, and lack of kitchen or plumbing) were not significantly associated with county-level HISA utilization rate but are also nevertheless relevant to the discussion of home modifications. Of particular interest is the percent of households with lack of kitchen or plumbing variable, which was positively related to county-level HISA utilization rate although not statistically significant. HM elements related to the kitchen (eg, heighten countertop) add to the accessibility of the home allowing for the performing of activities of daily living such as cooking. Between FY 2015 and FY 2018, we discovered 131 prescriptions for kitchen (n = 90) and plumbing (n = 41) HMs, which is a very small proportion of the 30,780 total HMs (there were 24,397 bathroom HMs). The nonsignificant coefficient for this variable may reflect the small number of veterans that obtained these HM.

Limitations

The potentially conflicting direction of effects for a significant access to care variable (percent uninsured adults) and a significant access to care and quality of care variable (preventable hospital stays) are interesting and warrant additional study, but the inability to interpret or explain this apparent inconsistency constitutes a limitation of the current data and analyses presented here. Another limitation is that this analysis uses county-level predictors for what is ultimately an individual-level outcome. It would have been ideal to have both individual- and county-level data to conduct a multilevel analysis; in particular, individual-level data within counties of individuals (both veterans and nonveterans) who did not receive a HISA award (including both those who applied and were denied, and who did not apply) would be highly valuable.

Conclusions

Our continuing research into veterans’ use of HM fills a gap in the literature about the characteristics of HISA users, the impact of county-level variables on the use of HISA, and the geographic distribution and use of HISA within the VHA. While it is important to examine the influence of broader systems on individual outcomes, there could be myriad other factors that are more proximal and more closely related to whether any one individual applies for, let alone receives, a HISA award. Indeed, a low overall adjusted model R² indicates that there is considerable variability in county-level HISA utilization rate that was not accounted for by the current model; this further speaks to warranted additional study.

More research is needed to understand and account for geographical variation in HISA utilization rate across the US. However, this work serves as an exploratory first step at quantifying and predicting HISA utilization rate at a broad level, with the ultimate goal of increasing access to HMs for veterans with disabilities.

Acknowledgments

This research was supported by grant 15521 from the US Department of Veterans Affairs, Office of Rural Health. Furthermore, the research was supported in part by grant K12 HD055929 from the National Institutes of Health. We want to acknowledge Cheri E. Knecht, Project Coordinator, for her assistance throughout all aspects of our research study and for her thoughtful contributions during the writing of this manuscript.

Suicide is the 10th leading cause of death in the US. In 2017, there were 47,173 deaths by suicide (14 deaths per 100,000 people), representing a 33% increase from 1999.¹ In 2017 veterans accounted for 13.5% of all suicide deaths among US adults, although veterans comprised only 7.9% of the adult population; the age- and sex-adjusted suicide rate was 1.5 times higher for veterans than that of nonveteran adults.^2,3

Among veteran users of Veterans Health Administration (VHA) services, mental health and substance use disorders, chronic medical conditions, and chronic pain are associated with an increased risk for suicide.³ About one-half of VHA veterans have been diagnosed with chronic pain.⁴ A chronic pain diagnosis (eg, back pain, migraine, and psychogenic pain) increased the risk of death by suicide even after adjusting for comorbid psychiatric diagnoses, according to a study on pain and suicide among US veterans.⁵

One-quarter of veterans received an opioid prescription during VHA outpatient care in 2012.⁴ Increased prescribing of opioid medications has been associated with opioid overdose and suicides.^6-10 Opioids are the most common drugs found in suicide by overdose.¹¹ The rate of opioid-related suicide deaths is 13 times higher among individuals with opioid use disorder (OUD) than it is for those without OUD.¹² The rate of OUD diagnosis among VHA users was 7 times higher than that for non-VHA users.¹³

In the US the age-adjusted rate of drug overdose deaths increased from 6 per 100,000 persons in 1999 to 22 per 100,000 in 2017.¹⁴ Drug overdoses accounted for 52,404 US deaths in 2015; 33,091 (63.1%) were from opioids.¹⁵ In 2017, there were 70,237 drug overdose deaths; 67.8% involved opioids (ie, 5 per 100,000 population represent prescription opioids).¹⁶

The VHA is committed to reducing opioid use and veteran suicide prevention. In 2013 the VHA launched the Opioid Safety Initiative employing 4 strategies: education, pain management, risk management, and addiction treatment.¹⁷ To address the opioid epidemic, the North Florida/South Georgia Veteran Health System (NF/SGVHS) developed and implemented a multispecialty Opioid Risk Reduction Program that is fully integrated with mental health and addiction services. The purpose of the NF/SGVHS one-stop pain addiction clinic is to provide a treatment program for chronic pain and addiction. The program includes elements of a whole health approach to pain care, including battlefield and traditional acupuncture. The focus went beyond replacing pharmacologic treatments with a complementary integrative health approach to helping veterans regain control of their lives through empowerment, skill building, shared goal setting, and reinforcing self-management.

The self-management programs include a pain school for patient education, a pain psychology program, and a yoga program, all stressing self-management offered onsite and via telehealth. Special effort was directed to identify patients with OUD and opioid dependence. Many of these patients were transitioned to buprenorphine, a potent analgesic that suppresses opioid cravings and withdrawal symptoms associated with stopping opioids. The clinic was structured so that patients could be seen often for follow-up and support. In addition, open lines of communication and referral were set up between this clinic, the interventional pain clinic, and the physical medicine and rehabilitation service. A detailed description of this program has been published elsewhere.¹⁸

The number of veterans receiving opioid prescription across the VHA system decreased by 172,000 prescriptions quarterly between 2012 and 2016.¹⁹ Fewer veterans were prescribed high doses of opioids or concomitant interacting medicines and more veterans were receiving nonopioid therapies.¹⁹ The prescription reduction across the VHA has varied. For example, from 2012 to 2017 the NF/SGVHS reported an 87% reduction of opioid prescriptions (≥ 100 mg morphine equivalents/d), compared with the VHA national average reduction of 49%.¹⁸

Vigorous opioid reduction is controversial. In a systematic review on opioid reduction, Frank and colleagues reported some beneficial effects of opioid reduction, such as increased health-related quality of life.²⁰ However, another study suggested a risk of increased pain with opioid tapering.²¹ The literature findings on the association between prescription opioid use and suicide are mixed. The VHA Office of Mental Health and Suicide Prevention literature review reported that veterans were at increased risk of committing suicide within the first 6 months of discontinuing opioid therapy.²² Another study reported that veterans who discontinued long-term opioid treatment had an increased risk for suicidal ideation.²³ However, higher doses of opioids were associated with an increased risk for suicide among individuals with chronic pain.¹⁰ The link between opioid tapering and the risk of suicide or overdose is uncertain.

Bohnert and Ilgen suggested that discontinuing prescription opioids leads to suicide without examining the risk factors that influenced discontinuation is ill-informed.⁷ Strong evidence about the association or relationship among opioid use, overdose, and suicide is needed. To increase our understanding of that association, Bohnert and Ilgen argued for multifaceted interventions that simultaneously address the shared causes and risk factors for OUD,⁷ such as the multispecialty Opioid Risk Reduction Program at NF/SGVHS.

Because of the reported association between robust integrated mental health and addiction, primary care pain clinic intervention, and the higher rate of opioid tapering in NF/SGVHS,¹⁸ this study aims to describe the pattern of overdose diagnosis (opioid overdose and nonopioid overdose) and pattern of suicide rates among veterans enrolled in NF/SGVHS, Veterans Integrated Service Network (VISN) 8, and the entire VA health care system during 2012 to 2016.The study reviewed and compared overdose diagnosis and suicide rates among veterans across NF/SGVHS and 2 other levels of the VA health care system to determine whether there were variances in the pattern of overdose/suicide rates and to explore these differences.

Methods

In this retrospective study, aggregate data were obtained from several sources. First, the drug overdose data were extracted from the VA Support Service Center (VSSC) medical diagnosis cube. We reviewed the literature for opioid codes reported in the literature and compared these reported opioid International Classification of Diseases, Ninth Revision (ICD-9) and International Classification of Diseases, 10th Revision (ICD-10) codes with the local facility patient-level comprehensive overdose diagnosis codes. Based on the comparison, we found 98 ICD-9 and ICD-10 overdose diagnosis codes and ran the modified codes against the VSSC national database. Overdose data were aggregated by facility and fiscal year, and the overdose rates (per 1,000) were calculated for unique veteran users at the 3 levels (NF/SGVHS, VISN 8, and VA national) as the denominator.

Each of the 18 VISNs comprise multiple VAMCs and clinics within a geographic region. VISN 8 encompasses most of Florida and portions of southern Georgia and the Caribbean (Puerto Rico, US Virgin Islands), including NF/SGVHS.

In this study, drug overdose refers to the overdose or poisoning from all drugs (ie, opioids, cocaine, amphetamines, sedatives, etc) and defined as any unintentional (accidental), deliberate, or intent undetermined drug poisoning.²⁴ The suicide data for this study were drawn from the VA Suicide Prevention Program at 3 different levels: NF/SGVHS, VISN 8, and VHA national. Suicide is death caused by an intentional act of injuring oneself with the intent to die.²⁵

This descriptive study compared the rate of annual drug overdoses (per 1,000 enrollees) between NF/SGVHS, VISN 8, and VHA national from 2012 to 2016. It also compared the annual rate of suicide per 100,000 enrollees across these 3 levels of the VHA. The overdose and suicide rates and numbers are mutually exclusive, meaning the VISN 8 data do not include the NF/SGVHS information, and the national data excluded data from VISN 8 and NF/SGVHS. This approach helped improve the quality of multiple level comparisons for different levels of the VHA system.

Results

Figure 1 shows the pattern of overdose diagnosis by rates (per 1,000) across the study period (2012 to 2016) and compares patterns at 3 levels of VHA (NF/SGVHS, VISN 8, and VHA national). The average annual rate of overdose diagnoses for NF/SGVHS during the study was slightly higher (16.8 per 1,000) than that of VISN 8 (16 per 1,000) and VHA national (15.3 per 1,000), but by the end of the study period the NF/SGVHS rate (18.6 per 1,000) nearly matched the national rate (18.2 per 1,000) and was lower than the VISN 8 rate (20.4 per 1,000). Additionally, NF/SGVHS had less variability (SD, 1.34) in yearly average overdose rates compared with VISN 8 (SD, 2.96), and VHA national (SD, 1.69).

From 2013 to 2014 the overdose diagnosis rate for NF/SGVHS remained the same (17.1 per 1,000). A similar pattern was observed for the VHA national data, whereas the VISN 8 data showed a steady increase during the same period. In 2015, the NF/SGVHS had 0.7 per 1,000 decrease in overdose diagnosis rate, whereas VISN 8 and VHA national data showed 1.7 per 1,000 and 0.9 per 1,000 increases, respectively. During the last year of the study (2016), there was a dramatic increase in overdose diagnosis for all the health care systems, ranging from 2.2 per 1,000 for NF/SGVHS to 3.3 per 1,000 for VISN 8.

Figure 2 shows the annual rates (per 100,000 individuals) of suicide for NF/SGVHS, VISN 8, and VHA national. The suicide pattern for VISN 8 shows a cyclical acceleration and deceleration trend across the study period. From 2012 to 2014, the VHA national data show a steady increase of about 1 per 100,000 from year to year. On the contrary, NF/SGVHS shows a low suicide rate from year to year within the same period with a rate of 10 per 100,000 in 2013 compared with the previous year. Although the NF/SGVHS suicide rate increased in 2016 (10.4 per 100,000), it remained lower than that of VISN 8 (10.7 per 100,00) and VHA national (38.2 per 100,000).

Discussion

This study described and compared the distribution pattern of overdose (nonopioid and opioid) and suicide rates at different levels of the VHA system. Although VHA implemented systemwide opioid tapering in 2013, little is known about the association between opioid tapering and overdose and suicide. We believe a retrospective examination regarding overdose and suicide among VHA users at 3 different levels of the system from 2012 to 2016 could contribute to the discussion regarding the potential risks and benefits of discontinuing opioids.

First, the average annual rate of overdose diagnosis for NF/SGVHS during the study period was slightly higher (16.8 per 1,000) compared with those of VISN 8 (16.0 per 1,000) and VHA national (15.3 per 1,000) with a general pattern of increase and minimum variations in the rates observed during the study period among the 3 levels of the system. These increased overdose patterns are consistent with other reports in the literature.¹⁴ By the end of the study period, the NF/SGVHS rate (18.6 per 1,000) nearly matched the national rate (18.2 per 1,000) and was lower than VISN 8 (20.4 per 1,000). During the last year of the study period (2016), there was a dramatic increase in overdose diagnosis for all health care systems ranging from 2.2 per 1,000 for NF/SGVHS to 3.3 per 1,000 for VISN 8, which might be because of the VHA systemwide change of diagnosis code from ICD-9 to ICD-10, which includes more detailed diagnosis codes.

Second, our results showed that NF/SGVHS had the lowest average annual suicide rate (9.1 per 100,000) during the study period, which is one-fourth the VHA national rate and 2.6 per 100,000 lower than the VISN 8 rate. According to Bohnert and Ilgen,programs that improve the quality of pain care, expand access to psychotherapy, and increase access to medication-assisted treatment for OUDs could reduce suicide by drug overdose.⁷ We suggest that the low suicide rate at NF/SGVHS and the difference in the suicide rates between the NF/SGVHS and VISN 8 and VHA national data might be associated with the practice-based biopsychosocial interventions implemented at NF/SGVHS.

Our data showed a rise in the incidence of suicide at the NF/SGVHS in 2016. We are not aware of a local change in conditions, policy, and practice that would account for this increase. Suicide is variable, and data are likely to show spikes and valleys. Based on the available data, although the incidence of suicides at the NF/SGVHS in 2016 was higher, it remained below the VISN 8 and national VHA rate. This study seems to support the practice of tapering or stopping opioids within the context of a multidisciplinary approach that offers frequent follow-up, nonopioid options, and treatment of opioid addiction/dependence.

Limitations

The research findings of this study are limited by the retrospective and descriptive nature of its design. However, the findings might provide important information for understanding variations of overdose and suicide among VHA enrollees. Studies that use more robust methodologies are warranted to clinically investigate the impact of a multispecialty opioid risk reduction program targeting chronic pain and addiction management and identify best practices of opioid reduction and any unintended consequences that might arise from opioid tapering.²⁶ Further, we did not have access to the VA national overdose and suicide data after 2016. Similar to most retrospective data studies, ours might be limited by availability of national overdose and suicide data after 2016. It is important for future studies to cross-validate our study findings.

Conclusions

The NF/SGVHS developed and implemented a biopsychosocial model of pain treatment that includes multicomponent primary care integrated with mental health and addiction services as well as the interventional pain and physical medicine and rehabilitation services. The presence of this program, during a period when the facility was tapering opioids is likely to account for at least part of the relative reduction in suicide.

Suicide is the 10th leading cause of death in the US. In 2017, there were 47,173 deaths by suicide (14 deaths per 100,000 people), representing a 33% increase from 1999.¹ In 2017 veterans accounted for 13.5% of all suicide deaths among US adults, although veterans comprised only 7.9% of the adult population; the age- and sex-adjusted suicide rate was 1.5 times higher for veterans than that of nonveteran adults.^2,3

Among veteran users of Veterans Health Administration (VHA) services, mental health and substance use disorders, chronic medical conditions, and chronic pain are associated with an increased risk for suicide.³ About one-half of VHA veterans have been diagnosed with chronic pain.⁴ A chronic pain diagnosis (eg, back pain, migraine, and psychogenic pain) increased the risk of death by suicide even after adjusting for comorbid psychiatric diagnoses, according to a study on pain and suicide among US veterans.⁵

One-quarter of veterans received an opioid prescription during VHA outpatient care in 2012.⁴ Increased prescribing of opioid medications has been associated with opioid overdose and suicides.^6-10 Opioids are the most common drugs found in suicide by overdose.¹¹ The rate of opioid-related suicide deaths is 13 times higher among individuals with opioid use disorder (OUD) than it is for those without OUD.¹² The rate of OUD diagnosis among VHA users was 7 times higher than that for non-VHA users.¹³

In the US the age-adjusted rate of drug overdose deaths increased from 6 per 100,000 persons in 1999 to 22 per 100,000 in 2017.¹⁴ Drug overdoses accounted for 52,404 US deaths in 2015; 33,091 (63.1%) were from opioids.¹⁵ In 2017, there were 70,237 drug overdose deaths; 67.8% involved opioids (ie, 5 per 100,000 population represent prescription opioids).¹⁶

The VHA is committed to reducing opioid use and veteran suicide prevention. In 2013 the VHA launched the Opioid Safety Initiative employing 4 strategies: education, pain management, risk management, and addiction treatment.¹⁷ To address the opioid epidemic, the North Florida/South Georgia Veteran Health System (NF/SGVHS) developed and implemented a multispecialty Opioid Risk Reduction Program that is fully integrated with mental health and addiction services. The purpose of the NF/SGVHS one-stop pain addiction clinic is to provide a treatment program for chronic pain and addiction. The program includes elements of a whole health approach to pain care, including battlefield and traditional acupuncture. The focus went beyond replacing pharmacologic treatments with a complementary integrative health approach to helping veterans regain control of their lives through empowerment, skill building, shared goal setting, and reinforcing self-management.

The self-management programs include a pain school for patient education, a pain psychology program, and a yoga program, all stressing self-management offered onsite and via telehealth. Special effort was directed to identify patients with OUD and opioid dependence. Many of these patients were transitioned to buprenorphine, a potent analgesic that suppresses opioid cravings and withdrawal symptoms associated with stopping opioids. The clinic was structured so that patients could be seen often for follow-up and support. In addition, open lines of communication and referral were set up between this clinic, the interventional pain clinic, and the physical medicine and rehabilitation service. A detailed description of this program has been published elsewhere.¹⁸

The number of veterans receiving opioid prescription across the VHA system decreased by 172,000 prescriptions quarterly between 2012 and 2016.¹⁹ Fewer veterans were prescribed high doses of opioids or concomitant interacting medicines and more veterans were receiving nonopioid therapies.¹⁹ The prescription reduction across the VHA has varied. For example, from 2012 to 2017 the NF/SGVHS reported an 87% reduction of opioid prescriptions (≥ 100 mg morphine equivalents/d), compared with the VHA national average reduction of 49%.¹⁸

Vigorous opioid reduction is controversial. In a systematic review on opioid reduction, Frank and colleagues reported some beneficial effects of opioid reduction, such as increased health-related quality of life.²⁰ However, another study suggested a risk of increased pain with opioid tapering.²¹ The literature findings on the association between prescription opioid use and suicide are mixed. The VHA Office of Mental Health and Suicide Prevention literature review reported that veterans were at increased risk of committing suicide within the first 6 months of discontinuing opioid therapy.²² Another study reported that veterans who discontinued long-term opioid treatment had an increased risk for suicidal ideation.²³ However, higher doses of opioids were associated with an increased risk for suicide among individuals with chronic pain.¹⁰ The link between opioid tapering and the risk of suicide or overdose is uncertain.

Bohnert and Ilgen suggested that discontinuing prescription opioids leads to suicide without examining the risk factors that influenced discontinuation is ill-informed.⁷ Strong evidence about the association or relationship among opioid use, overdose, and suicide is needed. To increase our understanding of that association, Bohnert and Ilgen argued for multifaceted interventions that simultaneously address the shared causes and risk factors for OUD,⁷ such as the multispecialty Opioid Risk Reduction Program at NF/SGVHS.

Because of the reported association between robust integrated mental health and addiction, primary care pain clinic intervention, and the higher rate of opioid tapering in NF/SGVHS,¹⁸ this study aims to describe the pattern of overdose diagnosis (opioid overdose and nonopioid overdose) and pattern of suicide rates among veterans enrolled in NF/SGVHS, Veterans Integrated Service Network (VISN) 8, and the entire VA health care system during 2012 to 2016.The study reviewed and compared overdose diagnosis and suicide rates among veterans across NF/SGVHS and 2 other levels of the VA health care system to determine whether there were variances in the pattern of overdose/suicide rates and to explore these differences.

Methods

In this retrospective study, aggregate data were obtained from several sources. First, the drug overdose data were extracted from the VA Support Service Center (VSSC) medical diagnosis cube. We reviewed the literature for opioid codes reported in the literature and compared these reported opioid International Classification of Diseases, Ninth Revision (ICD-9) and International Classification of Diseases, 10th Revision (ICD-10) codes with the local facility patient-level comprehensive overdose diagnosis codes. Based on the comparison, we found 98 ICD-9 and ICD-10 overdose diagnosis codes and ran the modified codes against the VSSC national database. Overdose data were aggregated by facility and fiscal year, and the overdose rates (per 1,000) were calculated for unique veteran users at the 3 levels (NF/SGVHS, VISN 8, and VA national) as the denominator.

Each of the 18 VISNs comprise multiple VAMCs and clinics within a geographic region. VISN 8 encompasses most of Florida and portions of southern Georgia and the Caribbean (Puerto Rico, US Virgin Islands), including NF/SGVHS.

In this study, drug overdose refers to the overdose or poisoning from all drugs (ie, opioids, cocaine, amphetamines, sedatives, etc) and defined as any unintentional (accidental), deliberate, or intent undetermined drug poisoning.²⁴ The suicide data for this study were drawn from the VA Suicide Prevention Program at 3 different levels: NF/SGVHS, VISN 8, and VHA national. Suicide is death caused by an intentional act of injuring oneself with the intent to die.²⁵

This descriptive study compared the rate of annual drug overdoses (per 1,000 enrollees) between NF/SGVHS, VISN 8, and VHA national from 2012 to 2016. It also compared the annual rate of suicide per 100,000 enrollees across these 3 levels of the VHA. The overdose and suicide rates and numbers are mutually exclusive, meaning the VISN 8 data do not include the NF/SGVHS information, and the national data excluded data from VISN 8 and NF/SGVHS. This approach helped improve the quality of multiple level comparisons for different levels of the VHA system.

Results

Figure 1 shows the pattern of overdose diagnosis by rates (per 1,000) across the study period (2012 to 2016) and compares patterns at 3 levels of VHA (NF/SGVHS, VISN 8, and VHA national). The average annual rate of overdose diagnoses for NF/SGVHS during the study was slightly higher (16.8 per 1,000) than that of VISN 8 (16 per 1,000) and VHA national (15.3 per 1,000), but by the end of the study period the NF/SGVHS rate (18.6 per 1,000) nearly matched the national rate (18.2 per 1,000) and was lower than the VISN 8 rate (20.4 per 1,000). Additionally, NF/SGVHS had less variability (SD, 1.34) in yearly average overdose rates compared with VISN 8 (SD, 2.96), and VHA national (SD, 1.69).

From 2013 to 2014 the overdose diagnosis rate for NF/SGVHS remained the same (17.1 per 1,000). A similar pattern was observed for the VHA national data, whereas the VISN 8 data showed a steady increase during the same period. In 2015, the NF/SGVHS had 0.7 per 1,000 decrease in overdose diagnosis rate, whereas VISN 8 and VHA national data showed 1.7 per 1,000 and 0.9 per 1,000 increases, respectively. During the last year of the study (2016), there was a dramatic increase in overdose diagnosis for all the health care systems, ranging from 2.2 per 1,000 for NF/SGVHS to 3.3 per 1,000 for VISN 8.

Figure 2 shows the annual rates (per 100,000 individuals) of suicide for NF/SGVHS, VISN 8, and VHA national. The suicide pattern for VISN 8 shows a cyclical acceleration and deceleration trend across the study period. From 2012 to 2014, the VHA national data show a steady increase of about 1 per 100,000 from year to year. On the contrary, NF/SGVHS shows a low suicide rate from year to year within the same period with a rate of 10 per 100,000 in 2013 compared with the previous year. Although the NF/SGVHS suicide rate increased in 2016 (10.4 per 100,000), it remained lower than that of VISN 8 (10.7 per 100,00) and VHA national (38.2 per 100,000).

Discussion

This study described and compared the distribution pattern of overdose (nonopioid and opioid) and suicide rates at different levels of the VHA system. Although VHA implemented systemwide opioid tapering in 2013, little is known about the association between opioid tapering and overdose and suicide. We believe a retrospective examination regarding overdose and suicide among VHA users at 3 different levels of the system from 2012 to 2016 could contribute to the discussion regarding the potential risks and benefits of discontinuing opioids.

First, the average annual rate of overdose diagnosis for NF/SGVHS during the study period was slightly higher (16.8 per 1,000) compared with those of VISN 8 (16.0 per 1,000) and VHA national (15.3 per 1,000) with a general pattern of increase and minimum variations in the rates observed during the study period among the 3 levels of the system. These increased overdose patterns are consistent with other reports in the literature.¹⁴ By the end of the study period, the NF/SGVHS rate (18.6 per 1,000) nearly matched the national rate (18.2 per 1,000) and was lower than VISN 8 (20.4 per 1,000). During the last year of the study period (2016), there was a dramatic increase in overdose diagnosis for all health care systems ranging from 2.2 per 1,000 for NF/SGVHS to 3.3 per 1,000 for VISN 8, which might be because of the VHA systemwide change of diagnosis code from ICD-9 to ICD-10, which includes more detailed diagnosis codes.

Second, our results showed that NF/SGVHS had the lowest average annual suicide rate (9.1 per 100,000) during the study period, which is one-fourth the VHA national rate and 2.6 per 100,000 lower than the VISN 8 rate. According to Bohnert and Ilgen,programs that improve the quality of pain care, expand access to psychotherapy, and increase access to medication-assisted treatment for OUDs could reduce suicide by drug overdose.⁷ We suggest that the low suicide rate at NF/SGVHS and the difference in the suicide rates between the NF/SGVHS and VISN 8 and VHA national data might be associated with the practice-based biopsychosocial interventions implemented at NF/SGVHS.

Our data showed a rise in the incidence of suicide at the NF/SGVHS in 2016. We are not aware of a local change in conditions, policy, and practice that would account for this increase. Suicide is variable, and data are likely to show spikes and valleys. Based on the available data, although the incidence of suicides at the NF/SGVHS in 2016 was higher, it remained below the VISN 8 and national VHA rate. This study seems to support the practice of tapering or stopping opioids within the context of a multidisciplinary approach that offers frequent follow-up, nonopioid options, and treatment of opioid addiction/dependence.

Limitations

The research findings of this study are limited by the retrospective and descriptive nature of its design. However, the findings might provide important information for understanding variations of overdose and suicide among VHA enrollees. Studies that use more robust methodologies are warranted to clinically investigate the impact of a multispecialty opioid risk reduction program targeting chronic pain and addiction management and identify best practices of opioid reduction and any unintended consequences that might arise from opioid tapering.²⁶ Further, we did not have access to the VA national overdose and suicide data after 2016. Similar to most retrospective data studies, ours might be limited by availability of national overdose and suicide data after 2016. It is important for future studies to cross-validate our study findings.

Conclusions

The NF/SGVHS developed and implemented a biopsychosocial model of pain treatment that includes multicomponent primary care integrated with mental health and addiction services as well as the interventional pain and physical medicine and rehabilitation services. The presence of this program, during a period when the facility was tapering opioids is likely to account for at least part of the relative reduction in suicide.

Suicide is the 10th leading cause of death in the US. In 2017, there were 47,173 deaths by suicide (14 deaths per 100,000 people), representing a 33% increase from 1999.¹ In 2017 veterans accounted for 13.5% of all suicide deaths among US adults, although veterans comprised only 7.9% of the adult population; the age- and sex-adjusted suicide rate was 1.5 times higher for veterans than that of nonveteran adults.^2,3

Among veteran users of Veterans Health Administration (VHA) services, mental health and substance use disorders, chronic medical conditions, and chronic pain are associated with an increased risk for suicide.³ About one-half of VHA veterans have been diagnosed with chronic pain.⁴ A chronic pain diagnosis (eg, back pain, migraine, and psychogenic pain) increased the risk of death by suicide even after adjusting for comorbid psychiatric diagnoses, according to a study on pain and suicide among US veterans.⁵

One-quarter of veterans received an opioid prescription during VHA outpatient care in 2012.⁴ Increased prescribing of opioid medications has been associated with opioid overdose and suicides.^6-10 Opioids are the most common drugs found in suicide by overdose.¹¹ The rate of opioid-related suicide deaths is 13 times higher among individuals with opioid use disorder (OUD) than it is for those without OUD.¹² The rate of OUD diagnosis among VHA users was 7 times higher than that for non-VHA users.¹³

In the US the age-adjusted rate of drug overdose deaths increased from 6 per 100,000 persons in 1999 to 22 per 100,000 in 2017.¹⁴ Drug overdoses accounted for 52,404 US deaths in 2015; 33,091 (63.1%) were from opioids.¹⁵ In 2017, there were 70,237 drug overdose deaths; 67.8% involved opioids (ie, 5 per 100,000 population represent prescription opioids).¹⁶

The VHA is committed to reducing opioid use and veteran suicide prevention. In 2013 the VHA launched the Opioid Safety Initiative employing 4 strategies: education, pain management, risk management, and addiction treatment.¹⁷ To address the opioid epidemic, the North Florida/South Georgia Veteran Health System (NF/SGVHS) developed and implemented a multispecialty Opioid Risk Reduction Program that is fully integrated with mental health and addiction services. The purpose of the NF/SGVHS one-stop pain addiction clinic is to provide a treatment program for chronic pain and addiction. The program includes elements of a whole health approach to pain care, including battlefield and traditional acupuncture. The focus went beyond replacing pharmacologic treatments with a complementary integrative health approach to helping veterans regain control of their lives through empowerment, skill building, shared goal setting, and reinforcing self-management.

The self-management programs include a pain school for patient education, a pain psychology program, and a yoga program, all stressing self-management offered onsite and via telehealth. Special effort was directed to identify patients with OUD and opioid dependence. Many of these patients were transitioned to buprenorphine, a potent analgesic that suppresses opioid cravings and withdrawal symptoms associated with stopping opioids. The clinic was structured so that patients could be seen often for follow-up and support. In addition, open lines of communication and referral were set up between this clinic, the interventional pain clinic, and the physical medicine and rehabilitation service. A detailed description of this program has been published elsewhere.¹⁸

The number of veterans receiving opioid prescription across the VHA system decreased by 172,000 prescriptions quarterly between 2012 and 2016.¹⁹ Fewer veterans were prescribed high doses of opioids or concomitant interacting medicines and more veterans were receiving nonopioid therapies.¹⁹ The prescription reduction across the VHA has varied. For example, from 2012 to 2017 the NF/SGVHS reported an 87% reduction of opioid prescriptions (≥ 100 mg morphine equivalents/d), compared with the VHA national average reduction of 49%.¹⁸

Vigorous opioid reduction is controversial. In a systematic review on opioid reduction, Frank and colleagues reported some beneficial effects of opioid reduction, such as increased health-related quality of life.²⁰ However, another study suggested a risk of increased pain with opioid tapering.²¹ The literature findings on the association between prescription opioid use and suicide are mixed. The VHA Office of Mental Health and Suicide Prevention literature review reported that veterans were at increased risk of committing suicide within the first 6 months of discontinuing opioid therapy.²² Another study reported that veterans who discontinued long-term opioid treatment had an increased risk for suicidal ideation.²³ However, higher doses of opioids were associated with an increased risk for suicide among individuals with chronic pain.¹⁰ The link between opioid tapering and the risk of suicide or overdose is uncertain.

Bohnert and Ilgen suggested that discontinuing prescription opioids leads to suicide without examining the risk factors that influenced discontinuation is ill-informed.⁷ Strong evidence about the association or relationship among opioid use, overdose, and suicide is needed. To increase our understanding of that association, Bohnert and Ilgen argued for multifaceted interventions that simultaneously address the shared causes and risk factors for OUD,⁷ such as the multispecialty Opioid Risk Reduction Program at NF/SGVHS.

Because of the reported association between robust integrated mental health and addiction, primary care pain clinic intervention, and the higher rate of opioid tapering in NF/SGVHS,¹⁸ this study aims to describe the pattern of overdose diagnosis (opioid overdose and nonopioid overdose) and pattern of suicide rates among veterans enrolled in NF/SGVHS, Veterans Integrated Service Network (VISN) 8, and the entire VA health care system during 2012 to 2016.The study reviewed and compared overdose diagnosis and suicide rates among veterans across NF/SGVHS and 2 other levels of the VA health care system to determine whether there were variances in the pattern of overdose/suicide rates and to explore these differences.

Methods

In this retrospective study, aggregate data were obtained from several sources. First, the drug overdose data were extracted from the VA Support Service Center (VSSC) medical diagnosis cube. We reviewed the literature for opioid codes reported in the literature and compared these reported opioid International Classification of Diseases, Ninth Revision (ICD-9) and International Classification of Diseases, 10th Revision (ICD-10) codes with the local facility patient-level comprehensive overdose diagnosis codes. Based on the comparison, we found 98 ICD-9 and ICD-10 overdose diagnosis codes and ran the modified codes against the VSSC national database. Overdose data were aggregated by facility and fiscal year, and the overdose rates (per 1,000) were calculated for unique veteran users at the 3 levels (NF/SGVHS, VISN 8, and VA national) as the denominator.

Each of the 18 VISNs comprise multiple VAMCs and clinics within a geographic region. VISN 8 encompasses most of Florida and portions of southern Georgia and the Caribbean (Puerto Rico, US Virgin Islands), including NF/SGVHS.

In this study, drug overdose refers to the overdose or poisoning from all drugs (ie, opioids, cocaine, amphetamines, sedatives, etc) and defined as any unintentional (accidental), deliberate, or intent undetermined drug poisoning.²⁴ The suicide data for this study were drawn from the VA Suicide Prevention Program at 3 different levels: NF/SGVHS, VISN 8, and VHA national. Suicide is death caused by an intentional act of injuring oneself with the intent to die.²⁵

This descriptive study compared the rate of annual drug overdoses (per 1,000 enrollees) between NF/SGVHS, VISN 8, and VHA national from 2012 to 2016. It also compared the annual rate of suicide per 100,000 enrollees across these 3 levels of the VHA. The overdose and suicide rates and numbers are mutually exclusive, meaning the VISN 8 data do not include the NF/SGVHS information, and the national data excluded data from VISN 8 and NF/SGVHS. This approach helped improve the quality of multiple level comparisons for different levels of the VHA system.

Results

Figure 1 shows the pattern of overdose diagnosis by rates (per 1,000) across the study period (2012 to 2016) and compares patterns at 3 levels of VHA (NF/SGVHS, VISN 8, and VHA national). The average annual rate of overdose diagnoses for NF/SGVHS during the study was slightly higher (16.8 per 1,000) than that of VISN 8 (16 per 1,000) and VHA national (15.3 per 1,000), but by the end of the study period the NF/SGVHS rate (18.6 per 1,000) nearly matched the national rate (18.2 per 1,000) and was lower than the VISN 8 rate (20.4 per 1,000). Additionally, NF/SGVHS had less variability (SD, 1.34) in yearly average overdose rates compared with VISN 8 (SD, 2.96), and VHA national (SD, 1.69).

From 2013 to 2014 the overdose diagnosis rate for NF/SGVHS remained the same (17.1 per 1,000). A similar pattern was observed for the VHA national data, whereas the VISN 8 data showed a steady increase during the same period. In 2015, the NF/SGVHS had 0.7 per 1,000 decrease in overdose diagnosis rate, whereas VISN 8 and VHA national data showed 1.7 per 1,000 and 0.9 per 1,000 increases, respectively. During the last year of the study (2016), there was a dramatic increase in overdose diagnosis for all the health care systems, ranging from 2.2 per 1,000 for NF/SGVHS to 3.3 per 1,000 for VISN 8.

Figure 2 shows the annual rates (per 100,000 individuals) of suicide for NF/SGVHS, VISN 8, and VHA national. The suicide pattern for VISN 8 shows a cyclical acceleration and deceleration trend across the study period. From 2012 to 2014, the VHA national data show a steady increase of about 1 per 100,000 from year to year. On the contrary, NF/SGVHS shows a low suicide rate from year to year within the same period with a rate of 10 per 100,000 in 2013 compared with the previous year. Although the NF/SGVHS suicide rate increased in 2016 (10.4 per 100,000), it remained lower than that of VISN 8 (10.7 per 100,00) and VHA national (38.2 per 100,000).

Discussion

This study described and compared the distribution pattern of overdose (nonopioid and opioid) and suicide rates at different levels of the VHA system. Although VHA implemented systemwide opioid tapering in 2013, little is known about the association between opioid tapering and overdose and suicide. We believe a retrospective examination regarding overdose and suicide among VHA users at 3 different levels of the system from 2012 to 2016 could contribute to the discussion regarding the potential risks and benefits of discontinuing opioids.

First, the average annual rate of overdose diagnosis for NF/SGVHS during the study period was slightly higher (16.8 per 1,000) compared with those of VISN 8 (16.0 per 1,000) and VHA national (15.3 per 1,000) with a general pattern of increase and minimum variations in the rates observed during the study period among the 3 levels of the system. These increased overdose patterns are consistent with other reports in the literature.¹⁴ By the end of the study period, the NF/SGVHS rate (18.6 per 1,000) nearly matched the national rate (18.2 per 1,000) and was lower than VISN 8 (20.4 per 1,000). During the last year of the study period (2016), there was a dramatic increase in overdose diagnosis for all health care systems ranging from 2.2 per 1,000 for NF/SGVHS to 3.3 per 1,000 for VISN 8, which might be because of the VHA systemwide change of diagnosis code from ICD-9 to ICD-10, which includes more detailed diagnosis codes.

Second, our results showed that NF/SGVHS had the lowest average annual suicide rate (9.1 per 100,000) during the study period, which is one-fourth the VHA national rate and 2.6 per 100,000 lower than the VISN 8 rate. According to Bohnert and Ilgen,programs that improve the quality of pain care, expand access to psychotherapy, and increase access to medication-assisted treatment for OUDs could reduce suicide by drug overdose.⁷ We suggest that the low suicide rate at NF/SGVHS and the difference in the suicide rates between the NF/SGVHS and VISN 8 and VHA national data might be associated with the practice-based biopsychosocial interventions implemented at NF/SGVHS.

Our data showed a rise in the incidence of suicide at the NF/SGVHS in 2016. We are not aware of a local change in conditions, policy, and practice that would account for this increase. Suicide is variable, and data are likely to show spikes and valleys. Based on the available data, although the incidence of suicides at the NF/SGVHS in 2016 was higher, it remained below the VISN 8 and national VHA rate. This study seems to support the practice of tapering or stopping opioids within the context of a multidisciplinary approach that offers frequent follow-up, nonopioid options, and treatment of opioid addiction/dependence.

Limitations

The research findings of this study are limited by the retrospective and descriptive nature of its design. However, the findings might provide important information for understanding variations of overdose and suicide among VHA enrollees. Studies that use more robust methodologies are warranted to clinically investigate the impact of a multispecialty opioid risk reduction program targeting chronic pain and addiction management and identify best practices of opioid reduction and any unintended consequences that might arise from opioid tapering.²⁶ Further, we did not have access to the VA national overdose and suicide data after 2016. Similar to most retrospective data studies, ours might be limited by availability of national overdose and suicide data after 2016. It is important for future studies to cross-validate our study findings.

Conclusions

The NF/SGVHS developed and implemented a biopsychosocial model of pain treatment that includes multicomponent primary care integrated with mental health and addiction services as well as the interventional pain and physical medicine and rehabilitation services. The presence of this program, during a period when the facility was tapering opioids is likely to account for at least part of the relative reduction in suicide.