Evaluation of Antipsychotic Reduction Efforts in Patients With Dementia in Veterans Health Administration Nursing Homes

The U.S. Food and Drug Administration (FDA) has not approved any drugs for behavioral and psychological symptoms of dementia (BPSDs), although the widespread prescribing of psychotropic and opioid medications to persons with dementia could represent off-label attempts to address BPSDs (1–3). Based on mortality concerns, the FDA issued black box warnings for mortality risk associated with atypical antipsychotics (2005) and conventional antipsychotics (2008) when used to treat dementia-related psychosis (4). Given these safety concerns, providers may prescribe alternative CNS-active medications in lieu of antipsychotics. However, commonly used alternatives, such as benzodiazepines and antiepileptics, have even poorer risk-to-benefit ratios (5). Ultimately, among psychotropic medications, atypical antipsychotics have the strongest evidence base for treatment of BPSDs, although their benefits may prove moderate at best (6). Numerous expert bodies and professional organizations, including the American Geriatrics Society, the American Association of Geriatric Psychiatry, and the American Psychiatric Association, have recommended nonpharmacological strategies as first-line treatments for addressing BPSDs (7, 8).

Prior to the first black box warning in 2005, 24%–32% of nursing home residents received antipsychotics (9–11). The Centers for Medicare and Medicaid Services (CMS) launched the National Partnership to Improve Dementia Care in Nursing Homes (CMSNP) in 2012 to address concerns about persistently high rates of antipsychotic use and the quality of care provided to nursing home residents with dementia (12). The CMSNP included training materials and online resources with the goal of reducing antipsychotic prescribing in nursing homes. The CMSNP appeared to lower antipsychotic prescribing in nursing homes from 24% in 2011 to 15% in 2018 (13). Starting in 2015, the CMS Five Star Quality Rating System for nursing homes included a publicly reported measure of antipsychotic drug use by nursing homes through the Nursing Home Compare website, thus incentivizing facilities to reduce their antipsychotic prescribing (14).

Whether the CMSNP recommendations translated into reductions in antipsychotic prescribing within the Veterans Health Administration (VHA), which provides care to many veterans with dementia in VHA nursing homes, named Community Living Centers (CLCs), remains unknown. The VHA maintains over 130 CLCs across the United States, which provide both short- and long-term nursing and medical care for veterans (15). In addition to the potential impact of the CMSNP within the VHA, in December 2013, the VHA launched the Psychotropic Drug Safety Initiative (PDSI), a nationwide psychopharmacology quality improvement initiative. From 2013 through 2017, the PDSI included two targets relevant to veterans with dementia: the proportion for whom antipsychotics were prescribed and the proportion for whom benzodiazepines were prescribed. Further, similar to community nursing homes, CLCs complete the CMS Minimum Data Set 3.0 for clinical assessment of veterans’ health and behaviors as well as to report CMS quality indicators. In 2018, the VHA developed a similar publicly reported rating system adapted from the CMS Nursing Home Compare website, called CLC Compare, to help evaluate the quality of care received by veterans in CLCs (16).

Therefore, both Medicare and the VHA have used antipsychotic use as a proxy measure for quality of care among patients with dementia in nursing home settings (17). In community settings, antipsychotic reduction led to an increase in other types of CNS-active medication prescribing (18); it remains unknown whether this also occurred within the VHA. In this study, we sought to evaluate national trends in antipsychotic and other CNS-active medication prescriptions for veterans with dementia residing in VHA nursing homes as well as to understand how the use of specific agents has changed over time.

Methods

We used patient-level data from the VHA Corporate Data Warehouse (CDW) to evaluate changes in antipsychotic and other CNS-active medication prescriptions from FY2009 through FY2018 for veterans with dementia residing in VHA nursing homes. The Veterans Affairs Ann Arbor Healthcare System institutional review board approved the study.

Study Cohort

We used data from the VHA CDW for all veterans from October 1, 2008, to September 31, 2018 (FY2009–FY2018). We first constructed quarterly cohorts of long-stay residents at VHA nursing facilities (i.e., CLCs). In community settings, “long-stay” is typically defined as a length of stay >100 nursing home days (19). However, only 28.9% of veterans with dementia had a CLC length of stay >100 days. We therefore chose a length of stay >30 days to represent long-stay CLC stays, which represents 53.0% of veterans with dementia in CLCs.

We included a veteran in a given quarterly cohort if they met the long-stay criterion (>30 days) for at least 1 day during the quarter, were ≥65 years of age during the quarter, and had a dementia diagnosis (see Table S1 in the online supplement for ICD-9 and ICD-10 codes) (20). We determined comorbid clinical conditions (e.g., depression, anxiety, bipolar disorder, chronic pain) based on an inpatient or outpatient diagnosis code in the previous 12 months (see Table S1 in the online supplement for ICD-9 and ICD-10 codes for clinical conditions; we identified chronic pain using previously established codes [21, 22]). We excluded veterans with a diagnosis of schizophrenia, Tourette’s syndrome, and Huntington’s disease to match the exclusion criteria that CMS uses for its antipsychotic quality measure (i.e., for Nursing Home Compare and the CMSNP) (23). A long-stay CLC resident remained in the cohort (denominator) until death, the end of the long-stay period (i.e., discharge from CLC), or the end of the study period, whichever came first.

Outcomes

We used Bar-Code Medication Administration data within CDW to identify medications of interest. For each veteran, we determined quarterly CNS-active medication administration, including antipsychotics, antidepressants, antiepileptics, anxiolytics (including benzodiazepines and sedative-hypnotics), opioids, and memory medications (i.e., cholinesterase inhibitors and memantine; see Table S2 in the online supplement) (18). Additionally, we included a composite measure of any non-antipsychotic psychotropic medication use, which included the following medication classes: antidepressants, antiepileptics, and anxiolytics, along with non-antipsychotic CNS-active medication polypharmacy, defined as three or more medications from these classes and/or opioids during a 30-day rolling window.

Statistical Analysis

We used an interrupted time-series design to evaluate quarterly trends in prescribing rates from FY2009 to FY2018. We began the analysis in FY2009, following the FDA black box warnings related to antipsychotic use among patients with dementia. To examine the association of the CMSNP and VHA PDSI recommendations with CNS-active medication prescribing, we divided time into three periods: period 1 covered the interval from FY2009 to the CMSNP (2012, first quarter [Q1]); period 2 covered the interval after the CMSNP (2012 Q2) to before the VHA PDSI (2013 Q3); and period 3 covered the interval from the VHA PDSI (2013 Q4) to the end of the study period (2018 Q4).

We calculated prescribing rates as the percentage of long-stay CLC residents with dementia in each quarter with a prescription of interest. We then fitted a three-phase interrupted time-series regression model to evaluate the association of the CMSNP and VHA PDSI, controlling for autocorrelation by assuming a first-order autoregressive process. We used quarterly percentage prescribed antipsychotics or other medications as our outcome of interest. Models included a linear time-trend variable, indicators for post-CMSNP and post-VHA PDSI, and terms for change in the linear time trend between the study periods (e.g., between the pre- and post-CMSNP periods). Lastly, we evaluated which specific medications increased or decreased over time by computing and plotting the percentage of veterans who had prescriptions for medication during each quarter. We determined statistical significance at the 0.05 level using two-tailed tests, and we used R, version 4.0.2, for all analyses.

Results

Our cohort included 35,742 veterans with who met the long-stay criterion (>30 days) in a CLC between FY2009 and FY2018 (Table 1). The cohort was predominantly male (97.9%) and non-Hispanic White (74.5%) and had a mean age of 80.5 years.

CNS-Active Medication Prescribing

Figure 1 illustrates the quarterly rates of CNS-active or memory medication prescribing for veterans with dementia residing in CLCs throughout the three study periods (Table S3 in the online supplement presents the quarterly rates represented in Figure 1). Table 2 depicts changes in the level and slope of medication prescribing before and after the start of the CMSNP and VHA PDSI.

Antipsychotics

At the start of the study (period 1) in FY2009, 33.7% of veterans had an antipsychotic medication prescription. By the end of the study (period 3), antipsychotic prescribing had decreased to 27.5%, representing a 6.2% absolute change from FY2009 to FY2018 (Figure 1A; see also Table S3 in the online supplement). The rate of antipsychotic prescribing decreased across all study periods, but only significantly declined following the start of the VA PDSI (slope=−0.13, p<0.001).

Quetiapine was the most commonly prescribed antipsychotic in all quarters, followed by haloperidol, risperidone, and olanzapine (Figure 2A). Prescribing of most antipsychotic agents remained stable, with the largest declines observed for risperidone (a decline from 9.9% to 6.5%) and haloperidol (a decline from 9.0% to 7.6%).

Antiepileptics

Among non-antipsychotic medications, antiepileptics showed the greatest growth in prescribing over the study period (Figure 1A). Among veterans with dementia, antiepileptic use increased from 26.8% at the beginning of the study to 43.3% by the end of period 3, representing a 16.5% absolute change. Antiepileptic prescribing increased significantly during period 1 (i.e., before the start of the CMSNP; slope=0.11, p=0.02) and continued to increase following the start of the CMSNP (period 2; slope change=0.28, p=0.03). Following the start of the VHA PDSI, antiepileptic prescribing continued to increase at a similar rate.

Prescribing of gabapentin, valproate derivatives, and levetiracetam had the largest increases during the study period (Figure 2B). The most prescribed antiepileptic was gabapentin, for which prescribing grew from 11.1% to 23.5% during the study period.

Antidepressants

Antidepressants were the most commonly prescribed CNS-active medication class at all study time points (Figure 1B). Overall use increased from 56.8% at the start of period 1 to 63.4% at the end of period 3 (6.6% increase). The rate of antidepressant growth was highest in period 3, after the start of the VHA PDSI (slope=0.07, p=0.04) (Table 2). By the post-VHA PDSI period, sertraline was the most commonly prescribed antidepressant among our study population (Figure 2C). Use of sedating antidepressants such as mirtazapine (from 10.2% to 14.5%) and trazodone (from 14.4% to 17.5%) increased, while citalopram declined markedly.

Anxiolytics

Anxiolytic prescribing (including benzodiazepines) declined throughout the study period, decreasing from 33.5% at the start of period 1 to 27.1% at the end of the study. The greatest decline in anxiolytic prescribing was observed in period 3 after the start of the VHA PDSI (slope change=−0.46, p=0.006).

Opioids

Prescription of opioids appeared common among long-stay veterans with dementia and increased significantly during period 1 (from 32.6% to 39.1%; slope=0.19, p=0.008) and period 2 (from 39.3% to 46.0%; slope change=0.44, p=0.04). After the start of period 3 (VHA PSDI), opioid prescribing began to decrease significantly, from 48.0% to 41.2% at the end of the study (slope change=−0.89, p<0.001).

Memory Medications

Prescribing of memory medications declined throughout the study, from 32.4% at the start of period 1 to 21.8% at the end of period 3, representing a 10.6% absolute decline. Memory medications were the least prescribed medication class for veterans with dementia throughout the study period.

Non-Antipsychotic Psychotropic Prescribing Overall and CNS-Active Polypharmacy

The prescribing of non-antipsychotic psychotropic medication overall (i.e., antiepileptics, antidepressants, and anxiolytics) increased in period 1, steadied following the start of the CMSNP in period 2, and increased significantly in period 3, following the VHA PDSI (Figure 1B; Table 2). A total of 81.1% of long-stay veterans had prescriptions for a non-antipsychotic psychotropic at the end of the study period (increased from 75.0% at the start of period 1). Prescribing of non-antipsychotic CNS-active polypharmacy (i.e., three or more non-antipsychotic prescriptions for antidepressants, antiepileptics, anxiolytics, and opioids) increased during the first and second periods but declined during the third period, following the VHA PDSI (Table 2; see also Figure S2 in the online supplement). Prescribing of CNS-active polypharmacy overall increased from 32.1% at the start of period 1 to 41.3% at the end of period 3, representing an absolute increase of 9.2 percentage points.

Discussion

This study provides data on national trends in CNS-active medication prescribing for veterans with dementia in VHA CLCs. We found that antipsychotic use among veterans with dementia declined over time, and the decline was most significant following the VHA PDSI. Antidepressant prescribing increased following the VHA PDSI while antiepileptic prescribing increased following both the CMSNP and the VHA PDSI. Overall prescribing of non-antipsychotic psychotropic medications in VHA CLCs remained high, with over 80% of veterans with dementia receiving prescriptions for a non-antipsychotic psychotropic medication at the end of the study period. Gabapentin, valproate derivatives, and sedative antidepressants had the largest contributions to the growth in CNS-active medication prescribing. Anxiolytic prescribing declined following the VHA PDSI, and memory medication prescribing declined across all study periods and had a lower prescribing rate than any of the other medication classes observed. Lastly, opioid prescribing remained common and increasing prior to and during the CMSNP, only decreasing following the start of the VHA PDSI.

Our findings in the VHA are consistent with previous studies in non-VHA settings (one in the community, two in nursing homes) (2, 18, 24), which found that initiatives focused on reducing antipsychotic use may have contributed to providers shifting to alternative medication classes with similar risks and even less evidence of benefit. A Canadian study (24) showed that decreased antipsychotic use in nursing homes between 2004 and 2013 was offset by increased sedative antidepressant and antiepileptic use. A recent U.S. study examining Medicare beneficiaries found that antiepileptic prescriptions for nursing home residents increased and accelerated after the initiation of the CMSNP (18). It appears that these trends also extend to the VHA CLC population, where antiepileptic prescribing occurred at even higher rates compared with rates outside of the VHA.

When evaluating change in specific medications, we found that quetiapine was the most commonly prescribed antipsychotic among patients with dementia, consistent with findings from community long-term care populations (25). While quetiapine has been found to have a lower mortality risk in dementia (26), studies also suggest that it may have less evidence of benefit in reducing BPSDs (27). Increases in antidepressant use were largely driven by prescribing of sertraline and sedating antidepressants (e.g., trazodone and mirtazapine), which may be used for treatment of depression or BPSDs. Notable decreases in citalopram were observed following the 2011 FDA drug safety warning regarding concerns about QT prolongation with high-dose citalopram use, as has been reported in other studies (28, 29).

The growth of antiepileptic use was largely driven by gabapentin, for which prescribing doubled during the study period, likely off-label for pain, anxiety, or BPSDs, rather than for seizure disorders (25). Significant increases in prescription opioid use were observed before and during the CMSNP. This is consistent with studies demonstrating high rates of opioid prescribing for patients with dementia in the community (2) as well as population-based studies demonstrating significant growth in opioid prescribing for patients with dementia (3). While such treatment may reflect improved recognition and treatment of pain in dementia or palliative approaches to dementia end-of-life care, it may also reflect another attempt to reduce BPSDs (3, 30). In addition to measures of the percentage of long-stay residents with dementia for whom an antipsychotic is prescribed, VHA CLCs also have a publicly reported quality measure of percentage of long-stay CLC residents who self-report moderate or severe pain (16), further incentivizing treatment of pain.

A recent international panel of geriatric psychiatry and dementia care experts noted that antipsychotics are the most effective pharmacotherapy for BPSDs, including psychosis, agitation, and aggression (31). Studies reviewing the use of other medication classes, such as antidepressants, benzodiazepines, antiepileptics, and opioids, for treatment of BPSDs show minimal benefit in reducing such symptoms (5, 30, 32). Additionally, these alternative medication classes substituted for antipsychotics can yield significant risks. Benzodiazepines can contribute to falls, worsened cognition, respiratory depression, and paradoxical disinhibition (33). Antiepileptics such as valproate derivatives can cause gait disturbances, tremor, cognitive changes, and mortality (26, 34). Gabapentin use can lead to ataxia, dizziness, drowsiness, fatigue, falls, and increased risk of respiratory suppression when combined with opioids (35). Reductions in prescribing of memory medications may reflect provider concerns about limited therapeutic benefit for patients with advanced dementia as well as concerns regarding side effects, including nausea, loss of appetite, and diarrhea (36).

The VHA may have achieved declines in several classes of CNS-active medications as a result of its multipronged efforts. In contrast with the CMSNP, which focused on antipsychotic use only, the VHA PDSI, in addition to an antipsychotic prescribing measure, addresses benzodiazepine use, which may explain the observed anxiolytic reductions in the VHA. Further VHA initiatives, including the 2013 Opioid Safety Initiative, may explain reductions in opioid prescribing as well as increased recognition of the dangers of opioid prescribing in patients with dementia (37). Additionally, in 2010, the VHA launched an initiative to adapt an evidenced-based behavioral approach (Staff Training in Assisted Living Residences, or STAR [38]) for the VHA to increase uptake of nonpharmacological strategies for BPSDs. We do not know whether CLCs with STAR implementation decreased antipsychotic use, and if so, whether the decrease reflected increased use of nonpharmacological strategies (39). Reports from long-term care staff outside of the VHA suggest considerable challenges in implementing such interventions given staff turnover as well as the time and resources needed to implement nonpharmacological strategies (40).

Our study has several limitations. First, our results on prescription rates may not be directly comparable to those in community nursing homes given the differences in the structure of VHA CLCs, including the presence of a national formulary and the range of psychosocial services available to veterans (15). Next, our sample includes a high proportion of men, and our study results may not generalize to other clinical populations. Third, prescriptions filled from pharmacy data can be an imprecise measure of actual drug exposure; medication fills may not reflect day-to-day use. However, this study focused on the impact of initiatives that influence prescribing rather than medication use. Next, we do not know the indication for medication prescribing (e.g., gabapentin could be prescribed for pain, seizures, or BPSDs) or the appropriateness of such prescribing. However, the potential for medication-related harms remains regardless of clinical indication. Further, prescribing trends may reflect changes in patient characteristics over time; however, our approach is consistent with the CMSNP and VHA PDSI, neither of which use case-mix adjustment. Lastly, administrative data do not include reliable measures of nonpharmacological interventions, limiting our ability to examine whether increased use of such interventions was associated with declines in antipsychotic prescribing.

Conclusions

Although antipsychotic prescribing to veterans with dementia in VHA nursing homes declined following both the CMSNP and the PDSI, concomitant increases in antiepileptic, antidepressant, and opioid prescribing occurred. Within the VHA, rates of anxiolytic and opioid prescribing to veterans with dementia declined following the PDSI, but the rate of any non-antipsychotic psychotropic use overall increased. Initiatives focused on improving quality of care for nursing home residents with BPSDs both in the VHA and in the community should 1) monitor use of all CNS-active medication and other potentially sedating treatments used for sedation in dementia; and 2) consider how to incentivize and measure use of recommended evidence-based nonpharmacological alternatives (6, 7). The latter link to policy changes to stimulate a reorganization of dementia care, where providers are compensated for time spent in elucidating and addressing modifiable triggers to BPSDs, and not perversely incentivized to utilize other medications with even less evidence for benefit in dementia.