1 INTRODUCTION

In the United States, the Veterans Health Administration (VA) faces a growing demand for long-term care among Veterans (Kinosian et al., 2007; Thomas et al., 2018). To help meet this need, VA contracts with non-VA providers allowing eligible Veterans to receive VA-financed care in privately owned community nursing homes hereafter referred to as CNHs (Intrator et al., 2020; Miller et al., 2015). VA purchasing of long-term care could expand amid reforms intended to increase Veterans' access to community providers (Kupfer et al., 2018; Veterans Access, Choice, and Accountability Act, 2014) creating a need to better understand the quality of VA-purchased nursing home care.

Despite antipsychotic medications being associated with serious adverse effects, including stroke and mortality (Maust et al., 2015; Randle et al., 2019; Schneider et al., 2005; Weintraub et al., 2016, 2017), their off-label prescribing (use for which there is no clinical indication) is common (Jeste et al., 2008; Kolanowski et al., 2006). A 2011 report estimated that 83% of Medicare antipsychotic drug claims for older adults in nursing homes were off-label (Levinson, 2011). There have been major quality improvement initiatives to reduce antipsychotic prescribing in nursing homes, including the 2012 launch of the National Partnership to Improve Dementia Care by Centers for Medicare and Medicaid Services (CMS) (Carnahan et al., 2017; Crystal et al., 2020; Lucas & Bowblis, 2017). From 2011Q4 to 2018Q4, antipsychotic use decreased from a national average of 23.9%–14.6% among long-stay nursing home residents, excluding those diagnosed with schizophrenia, Huntington's disease or Tourette's syndrome (Partnership, 2019).

Prior studies suggest that characteristics often associated with higher antipsychotic use in nursing homes are related to fewer resources and limited funding. Having a greater proportion of Medicaid-insured residents, for instance, has been associated with increased nursing home antipsychotic use (Fashaw et al., 2020). Given that Medicaid reimbursement rates are lower than private-pay, Medicaid-reliant facilities may have less resources. Therefore, antipsychotics could be serving as a cost-saving alternative to investing in specialized training and more highly trained nursing staff. Meanwhile, other studies reported that NHs with mental health staff are more likely to prescribe antipsychotics (Bonner et al., 2015; Hughes et al., 2000; Lucas et al., 2014). This could be explained by a greater share of residents with behavioral problems in facilities where mental health staff are more available or to the practices of mental health professionals (Cioltan et al., 2017). Although extensive literature has evaluated antipsychotic use in nursing homes, there is both limited information and a lack of contemporary data about the use of antipsychotics by Veterans (Gellad et al., 2012; Leslie et al., 2009). Antipsychotic use among Veterans who reside in CNHs has not been systematically evaluated. As the use of VA-financed care from privately owned nursing homes is expected to increase, it is unknown whether community rates of antipsychotic prescribing are associated with antipsychotic use when Veterans are admitted to CNHs.

CNHs are typically subject to CMS quality performance monitoring and reimbursement incentives (Lucas & Bowblis, 2017) that could influence antipsychotic prescribing practices. Since the incentives are designed to penalize only off-label antipsychotic use, an unintended consequence of these programs may be that residents receive diagnoses of psychiatric conditions such as schizophrenia for which antipsychotics are indicated—conditions that, under usual circumstances, are rarely diagnosed in older adults (Statement, 2017). Our objectives are to assess the influence of prevalent rates of antipsychotic use in CNHs on the likelihood of individual-level antipsychotic initiation and receipt of a new diagnosis excluded from nursing home antipsychotic reporting to CMS. We hypothesize that antipsychotic-naïve Veterans admitted to CNHs with higher rates of prevalent antipsychotic use are significantly more likely to initiate antipsychotic therapy and be diagnosed with antipsychotic indications than Veterans at CNHs with lower prevalent rates of antipsychotic use.

2 METHODS 2.1 Study design and data sources

We conducted a retrospective study of VA-enrolled Veterans admitted to CNHs between 2013 and 2016. We used data from VA's Corporate Data Warehouse, including enrollment files, inpatient and outpatient encounters, and non-VA care in the community financed by VA. We used Medicare enrollment and claims to obtain additional data on Veterans' healthcare utilization. Assessments from CMS′ Resident Assessment Instrument/Minimum Data Set (MDS) were linked to Veterans' CNH stays. All Veterans included in this analysis had MDS data. To identify measures of quality and nursing home characteristics we linked Medicare's Nursing Home Compare quarterly reports. Linkages of VA and Medicare healthcare databases were developed by the Veteran's Affairs Information Resource Center (VIReC), a resource center of the VA Health Services Research & Development. VIReC maintains a crosswalk file with at least one unique patient identifier (e.g., Social Security Number [SSN]) included in both datasets to facilitate exact linkages (Hynes et al., 2018; Zhu et al., 2009). Additionally, VIReC receives all MDS assessment data for VA enrolled Veterans and links the assessments to VA data using scrambled SSNs. VIReC makes linked VA, Medicare, and MDS data available to VA researchers with approved research protocols.

2.2 Study sample

We analyzed data from Veterans with long stays at CNHs. We identified long-stay residents by the presence of an MDS assessment conducted around 100 days after the admission date. Stays were excluded if the Veteran had previously resided in a nursing home financed by Medicare or the VA in the prior 180 days, lacked a valid residential zip code, lived outside the 50 states, or filled prescriptions for antipsychotic medications 6 months prior to CNH admission. Prescription fills were determined using VA pharmacy data and Medicare Part D drug claims where available. Similarly, we relied on a combination of VA and Medicare claims to identify clinical diagnoses at baseline. We deemed the lack of pharmacy fills or diagnostic codes of interest to indicate non-use of those medications and absence of those conditions over a specified look-back period. At baseline, defined as the 12 months before admission, we classified Veterans by whether they had been diagnosed with an indication for receipt of antipsychotics. We regarded schizophrenia, Tourette's syndrome, and Huntington's disease as indications because these conditions are excluded from antipsychotic quality measure reporting based on CMS Surveyor Guidance (“CMS”). Our main analysis focuses on Veterans without CMS indications for antipsychotic use at admission.

2.3 Measures 2.3.1 Exposure variable

We ensured that the antipsychotic exposure measure was determined in the period before a Veteran entered a CNH. The study exposure was the prevalent antipsychotic prescribing rate, which we obtained from Nursing Home Compare, measured at the nursing home level and during the quarter preceding an individual Veteran's admission date. We assigned each CNH to a quintile, with quintile 1 as the lowest level of antipsychotic prescribing and quintile 5 the highest.

2.3.2 Outcome variables

We measured two outcomes using nursing home MDS clinical assessment data collected approximately 100 days after CNH admission: (1) new antipsychotic medication use, and (2) new diagnosis of schizophrenia, Tourette's syndrome, or Huntington's disease among those without an indication at baseline. The MDS items used in these outcome definitions were N0450A (antipsychotic medication review), I6000 (schizophrenia), I5250 (Huntington's disease), and I5350 (Tourette's syndrome).

2.3.3 Covariates

Demographic characteristics of Veterans included age, sex, race, and ethnicity, marital status, and a service-connected disability that entitles the Veteran to VA-financed long-term care (i.e., priority level 1). A Veteran's home residence was determined to be rural if it was in a rural-urban commuting area with codes 8, 9, or 10. To control for Veteran access to a VA-operated community living center (CLC), we included an indicator if there was one located within 30 miles of their home. Medicare and Medicaid insurance coverage were derived from the Medicare enrollment file. Clinical characteristics included hospitalization in the prior year, hospitalization immediately before CNH admission, and various physical and psychiatric comorbidities. All comorbidities were identified using ICD9/10 diagnosis codes in VA, community care financed by VA, and Medicare claims from the year prior to admission. We adjusted for hospice care during the CNH stay to account for potential antipsychotic use to treat terminal delirium. To avoid model overspecification we adjusted for limited nursing home characteristics including the overall star rating during the admission month and the number of other Veterans residing in the CNH at time admission.

2.3.4 Antipsychotic prescribing ratio

We compared prevalent rates of antipsychotic prescribing among VA-contracted CNHs relative to all other community nursing homes in a VA Medical Center's (VAMC) local market. We first defined a set of potential CNHs in a Veteran's community based on the distance from the Veteran's home address. For non-rural Veterans, a nursing home was included if it was within 25 and 35 miles for rural Veterans. If there were fewer than 15 nursing homes within that radius, we added the nearest nursing home in terms of distance until the Veteran's choice set included at least 15 facilities. To be included in the Veteran's local market, the nursing home needed to be operating in the month of admission. We included the CNH where the Veteran was admitted in the choice set. After assignment at the Veteran level, the local market was aggregated to the VAMC level. There was no adjustment done for the size of VAMCs. The antipsychotic prescribing ratio for each VAMC market was calculated as the mean prevalent antipsychotic use rate in contracted nursing homes divided by the mean prevalent antipsychotic use among non-contracted nursing homes.

2.4 Statistical analysis

We used a logit model to estimate a resident's probability of new antipsychotic prescription around 100 days after admission. We adjusted for the aforementioned covariates in the model with VAMC and year fixed effects. We estimated a separate model of the probability of being newly diagnosed with schizophrenia, Tourette's syndrome, or Huntington's disease after admission to the CNH.

2.5 Secondary analysis

We additionally analyzed a sample of Veterans with an indication for antipsychotic use at baseline in order to contrast the extent of the association with incident antipsychotic use against that observed among Veterans without an indication at baseline. Because there were too few Veterans who were antipsychotic-naïve who were diagnosed with schizophrenia, Tourette's syndrome, and Huntington's disease at baseline (n = 375), we additionally included FDA-approved adult indications (bipolar disorder, bipolar depression, schizoaffective disorder, major depressive disorder) to create a subgroup with indications at baseline based on a broad definition (Indications, 2015).

All analyses were performed using SAS Enterprise Guide 7.1 (SAS Institute Inc) and STATA 15. The study was approved by the Providence VAMC Institutional Review Board, and all data were acquired through data use agreements for VA and VA/CMS data provided by the Department of Veterans Affairs and VIReC.

3 RESULTS 3.1 Descriptive results

The final analytic sample in the main analysis comprised 8201 (77.9%) Veterans who were antipsychotic-naïve and had not been diagnosed with schizophrenia, Tourette's syndrome, or Huntington's disease before CNH admission. Veterans were distributed across 1035 unique CNHs and had mean length of stay (standard deviation) of 324 (275) days. The prevalent rates of antipsychotic use at CNHs were 0%–10.9% (quintile 1), 11%–14.7% (quintile 2), 14.8%–19.2% (quintile 3), 19.3%–25.6% (quintile 4), and 25.7%–91.4% (quintile 5).

Overall, among Veterans without an antipsychotic indication at admission, 89.8% were at least 65 years of age, and 97.1% male. Non-Hispanic Whites accounted for 74.5% of Veterans; whereas 15.7% were Black, and 3.2% were Hispanic (Table 1). Demographically, Veterans tended to be younger and more likely to be Hispanic as the prevalent rate of antipsychotic use increased. The proportion of Veterans with a service-connected disability was higher with increasing prevalent antipsychotic rates as were the proportions enrolled in Medicare or hospitalized in the year prior to CNH admission. Veterans admitted to CNHs with high antipsychotic use rates were less likely to have a home residence within 30 miles of a CLC: 26.7% in quintile 1 versus 17.9% in quintile 5. The distributions of comorbidities were similar across quintiles for many conditions except for myocardial infarction, dementia, chronic pulmonary disease, diabetes without complications, and cancer. These relationships were often nonlinear where they existed with exception of dementia which was generally more prevalent in CNHs with higher antipsychotic use rates.

TABLE 1. Characteristics of Veterans without exclusionary diagnoses (schizophrenia, Tourette's syndrome, and Huntington's disease) at admission to VA-contracted community nursing homes—overall and by quintile of prevalent rates of antipsychotic use Characteristic, n (%) Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 Total p-value CNH prevalent antipsychotic use rate 0%–10.9% 11.0%–14.7% 14.8%–19.2% 19.3%–25.6% 25.7%–91.4% 18.6% (SD 9.8%) Number of Veterans, N 1654 1654 1664 1638 1591 8201 Age, years <55 27 (1.6) 31 (1.9) 32 (1.9) 32 (2.0) 40 (2.5) 162 (2.0) <0.001 56–64 129 (7.8) 137 (8.3) 117 (7.0) 123 (7.5) 165 (10.4) 671 (8.2) 65–74 558 (33.7) 503 (30.4) 518 (31.1) 539 (32.9) 551 (34.6) 2669 (32.5) 75–84 310 (18.7) 336 (20.3) 351 (21.1) 320 (19.5) 323 (20.3) 1640 (20.0) 85+ 630 (38.1) 647 (39.1) 646 (38.8) 624 (38.1) 512 (32.2) 3059 (37.3) Male 1601 (96.8) 1607 (97.2) 1626 (97.7) 1593 (97.3) 1538 (96.7) 7965 (97.1) 0.40 Race and ethnicity White non-Hispanic 1185 (71.6) 1254 (75.8) 1281 (77.0) 1233 (75.3) 1156 (72.7) 6109 (74.5) <0.001 Black 303 (18.3) 251 (15.2) 238 (14.3) 252 (15.4) 243 (15.3) 1287 (15.7) Hispanic 46 (2.8) 49 (3.0) 46 (2.8) 52 (3.2) 68 (4.3) 261 (3.2) Other 61 (3.7) 38 (2.3) 29 (1.7) 24 (1.5) 20 (1.3) 172 (2.1) Unknown 59 (3.6) 62 (3.7) 70 (4.2) 77 (4.7) 104 (6.5) 372 (4.5) Married 859 (51.9) 891 (53.9) 852 (51.2) 850 (51.9) 734 (46.1) 4186 (51.0) <0.001 Priority group 1 (service-connected disability > 50%) 1313 (79.4) 1322 (79.9) 1352 (81.3) 1355 (82.7) 1363 (85.7) 6705 (81.8) <0.001 2–4 (other disabled) 131 (7.9) 131 (7.9) 139 (8.4) 120 (7.3) 111 (7.0) 632 (7.7) 5–8 (no disability) 210 (12.7) 201 (12.2) 173 (10.4) 163 (10.0) 117 (7.4) 864 (10.5) Medicaid 49 (3.0) 47 (2.8) 58 (3.5) 56 (3.4) 52 (3.3) 262 (3.2) 0.79 Medicare 1046 (63.2) 1073 (64.9) 1166 (70.1) 1178 (71.9) 1231 (77.4) 5694 (69.4) <0.001 Hospitalized in year prior to admission 1123 (67.9) 1130 (68.3) 1137 (68.3) 1118 (68.3) 1124 (70.6) 5632 (68.7) 0.45 Admitted from a hospital stay 1035 (62.6) 1050 (63.5) 1050 (63.1) 1050 (64.1) 1075 (67.6) 5260 (64.1) 0.027 CLC at home VAMC 441 (26.7) 413 (25.0) 388 (23.3) 288 (17.6) 284 (17.9) 1814 (22.1 <0.001 Comorbidities Myocardial infarction 241 (14.6) 261 (15.8) 290 (17.4) 285 (17.4) 205 (12.9) 1282 (15.6) <0.001 Peripheral vascular disease 627 (37.9) 638 (38.6) 620 (37.3) 639 (39.0) 567 (35.6) 3091 (37.7) 0.31 Dementia 602 (36.4) 620 (37.5) 624 (37.5) 683 (41.7) 661 (41.5) 3190 (38.9) 0.002 Chronic pulmonary disease 679 (41.1) 737 (44.6) 759 (45.6) 785 (47.9) 702 (44.1) 3662 (44.7) 0.002 Diabetes without complications 764 (46.2) 746 (45.1) 780 (46.9) 840 (51.3) 783 (49.2) 3913 (47.7) 0.003 Diabetes with complications 341 (20.6) 297 (18.0) 297 (17.8) 304 (18.6) 267 (16.8) 1506 (18.4) 0.067 Cancer 392 (23.7) 375 (22.7) 356 (21.4) 376 (23.0) 311 (19.5) 1810 (22.1) 0.041 Metastatic carcinoma 118 (7.1) 98 (5.9) 98 (5.9) 102 (6.2) 78 (4.9) 494 (6.0) 0.12 Post-traumatic stress disorder 223 (13.5) 208 (12.6) 221 (13.3) 218 (13.3) 203 (12.8) 1073 (13.1) 0.93 Substance use disorder 125 (7.6) 119 (7.2) 124 (7.5) 153 (9.3) 118 (7.4) 639 (7.8) 0.14 Major depressive disorder 755 (45.6) 770 (46.6) 795 (47.8) 815 (49.8) 790 (49.7) 3925 (47.9) 0.066 Bipolar disorder 82 (5.0%) 77 (4.7%) 109 (6.6%) 89 (5.4%) 150 (9.4%) 507 (6.2%) <0.001 Hospice care during CNH stay 60 (3.6) 77 (4.7) 87 (5.2) 108 (6.6) 131 (8.2) 463 (5.7) <0.001 Outcome measures Initiating antipsychotics 226 (12.7) 269 (15.3) 324 (18.5) 385 (22.2) 526 (31.8) 1730 (21.1) <0.001 Acquiring indicationa 40 (2.3) 28 (1.6) 51 (2.9) 51 (2.9) 113 (6.8) 283 (3.5) <0.001 Abbreviations: CLC, community living center; CNH, community nursing home; SD, standard deviation; VAMC, VA medical center. a The total diagnosed with schizophrenia was 274 and less than 10 were diagnosed with each of Tourette's syndrome and Huntington's disease.

On average, 21.1% of long-stay Veterans initiated antipsychotic medications after CNH admission. The proportion initiating antipsychotics increased across quintiles and ranged from 12.7% (quintile 1) to 31.8% (quintile 5). Overall, 3.5% (n = 283) of Veterans received a new diagnosis of schizophrenia, Tourette's syndrome, or Huntington's disease after CNH admission. There were 274 Veterans newly diagnosed with schizophrenia representing a vast majority (96.8%) of the new diagnoses that were identified. The proportion of Veterans acquiring new indications was comparable across CNHs in quintiles 1–4 (2.3%–2.9%) but markedly higher in quintile 5 at 6.8%.

At the nursing home level, CNHs with higher prevalent rates of antipsychotic use had more residents on Medicaid, fewer registered nurses, greater total beds, lower overall star rating, and a larger number of Veterans with VA-paid care than CNHs with lower antipsychotic use (Table 2). On average, 46 nursing homes were in a Veteran's choice set with 5 of those facilities under contract with VA. No VAMC had nursing home markets that were all under VA contracts. There was considerable variation in the antipsychotic prescribing ratio across VAMC markets (Figure 1). Almost three-quarters (72.4%) of VAMC-contracted CNHs had greater prevalent antipsychotic rates than nursing homes VAMCs did not contract with.

TABLE 2. Characteristics of VA-contracted community nursing homes by quintiles of prevalent rates of antipsychotic use Characteristic Quintile 1 Quintile 2 Quintile 3 Quintile 4 Quintile 5 p-value CNH prevalent antipsychotic use rate 0%–10.9% 11%–14.7% 14.8%–19.2% 19.3%–25.6% 25.7%–91.4% Maximum number of veterans, mean (SD) 10.3 (7.1) 9.9 (7.2) 9.5 (6.7) 9.9 (7.3) 11.1 (8.1) <0.001 Distance to CNH from home residence (miles), mean (SD) 41.0 (228.7) 48.8 (327.7) 92.6 (328.1) 50.7 (315.8) 42.6 (340.8) <0.001 Distance to CNH from VAMC, mean (SD) 39.7 (52.3) 37.5 (52.2) 40.4 (57.6) 37.9 (44.8) 39.8 (60.2) 0.50 Total beds, mean (SD) 128.2 (48.1) 140.7 (59.4) 137.8 (56.9) 148.7 (66.5) 147.0 (58.8) <0.001 Part of a chain 930 (70.6%) 994 (69.4%) 1006 (70.0%) 959 (66.0%) 979 (61.6%) <0.001 For profit 1119 (85.0%) 1125 (78.5%) 1130 (78.6%) 1174 (80.9%) 1413 (89.0%) <0.001 Hospital based 14 (1.1%) 22 (1.5%) 25 (1.7%) 11 (0.8%) 4 (0.3%) <0.001 % Medicaid residents, mean (SD) 55.9 (17.2) 59.4 (14.6) 60.4 (14.6) 62.4 (14.4) 66.6 (13.5) <0.001 RN hours/Patient day, mean (SD) 0.5 (0.4) 0.5 (0.3) 0.5 (0.3) 0.5 (0.2) 0.4 (0.2) <0.001 LPN hours/Patient day, mean (SD) 0.9 (0.4) 0.8 (0.3)