Nursing Home Characteristics Associated with High and Low Levels of Antipsychotic, Benzodiazepine, and Opioid Prescribing to Residents with Alzheimer’s Disease and Related Dementias: A Cross-Sectional Analysis

Abstract

Objectives:

We examined the association between nursing home (NH) characteristics and whether NHs had high or low levels of antipsychotic, benzodiazepine, or opioid prescribing to residents with Alzheimer’s Disease and related dementias (ADRD). We then measured the likelihood that NHs who were high (low) prescribers of antipsychotics were also high (low) prescribers of benzodiazepines or opioids.

Design:

A retrospective, cross-sectional analysis.

Setting and Participants:

The sample included 448,128 Medicare beneficiaries diagnosed with ADRD, who resided in 13,151 NHs in 2017.

Methods:

Using Medicare claims, the Minimum Data Set, and LTCFocus, we measured the share of NH residents with ADRD who filled 1+ antipsychotic, benzodiazepine, or opioid prescription in 2017. Using linear probability models with state-clustered standard errors, we identified which NH characteristics were associated with being in the top (bottom) quartile of the prescribing distribution for each drug class. Finally, we measured whether NHs who were top-quartile (bottom-quartile) antipsychotic prescribers were more likely to be top-quartile (bottom-quartile) benzodiazepine or opioid prescribers.

Results:

Across NHs, an average of 29.1% of residents with ADRD received an antipsychotic, 30.2% received a benzodiazepine, and 40.9% received an opioid. Smaller NHs and NHs with a larger share of Medicaid-enrolled residents were more likely to be top-quartile prescribers; NHs with more registered nursing care were more likely to be bottom-quartile prescribers. Antipsychotic prescribing tracked closely with benzodiazepine prescribing, but not opioid prescribing.

Conclusions and Implications:

The overlap between antipsychotic and benzodiazepine prescribing and our finding that some NH characteristics were consistently associated with prescribing across drug classes may support the idea of an organizational culture of prescribing in NHs, which could inform efforts to improve prescribing quality in NHs. Our results also highlight benzodiazepine and opioid use for ADRD, which were more commonly prescribed than antipsychotics in NHs but have received less regulatory attention.

Brief Summary:

Some nursing home characteristics are consistently associated with high and low levels of antipsychotic, benzodiazepine, and opioid prescribing to residents with dementia, which may suggest the presence of an organizational culture of prescribing.

Introduction

In 2016, nearly half of the over 1.3 million nursing home (NH) residents in the U.S. had Alzheimer’s disease or a related dementia (ADRD).¹ The use of psychotropic medications among NH residents overall, and among residents with ADRD in particular, has been the focus of significant regulatory and research efforts since the Omnibus Budget Reconciliation Act of 1987, which included provisions prohibiting the use of psychotropic medications by NHs as chemical restraints to control resident behavior.² While the use of any chemical restraint should be minimized, attention has predominantly focused on antipsychotic prescribing, which is often used as a proxy for the quality of ADRD care.^3,4

Studies of prescribing to NH residents with ADRD often consider patient-level characteristics (e.g., gender, race/ethnicity, patient acuity) and/or facility-level characteristics (e.g., payer mix, nursing staffing, for-profit status), but tend to focus on the relationship between these characteristics and a single drug class.^5–9 Citing the extent of unexplained NH-level variation in prescribing, even after accounting for patient- and facility-level characteristics, researchers have proposed the idea that an organizational culture of prescribing may serve as a potentially important determinant of prescribing quality.^10–14 Tjia, Gurwitz, and Briesacher (2012), who refer to “prescribing cultures,” note that NHs are “complex institutions with a wide variety of organizational models and health care professionals” and that the values, beliefs, and assumptions of the professionals who work within NHs are relevant when attempting to improve health care outcomes, which could include medication use.¹⁵

An organizational culture of prescribing offers an alternative model by which to improve the quality of prescribing by considering the NH as a system, rather than focusing on individual prescribers within a NH. This model could also be relevant when considering prescribing across drug classes, and may suggest a proclivity toward medication use overall. The drug class that has received the most attention from researchers and policymakers, at least when it comes to ADRD care in NHs, is antipsychotics.¹⁶ However, medications like benzodiazepines (BZDs) are commonly used to treat behavioral symptoms of ADRD and are associated with adverse outcomes.^17–20 Another drug class of interest is opioids. Like BZDs, opioids are controlled substances in the U.S. and there are ongoing questions about the appropriateness of their use (although studies suggest that individuals with ADRD may face barriers to pain management).^21–24 Whether NHs who are high (low) prescribers of antipsychotics are also high (low) prescribers of BZDs or opioids, and whether the NH characteristics that are associated with high (low) levels of prescribing overlap, is unclear.

In this study, we explore the idea of an organizational culture of prescribing across drug classes by measuring the overlap between antipsychotic, BZD, and opioid prescribing to NH residents with ADRD. We construct binary outcomes of whether NHs were in the top or bottom quartile of the share of residents with ADRD and at least one antipsychotic, BZD, and opioid prescription fill during 2017, then measure associations between prescribing outcomes and NH characteristics, including payer mix, nursing staffing, for-profit status, and patient acuity. We also examine whether NHs in the top (bottom) quartile of antipsychotic prescribing are more or less likely of being in the top (bottom) quartile of BZD or opioid prescribing. We hypothesize that an organizational culture of prescribing that supports higher rates of antipsychotic prescribing will be reflected through higher rates of BZD and opioid prescribing.

Methods

Study Design

In this retrospective, cross-sectional analysis, we used three administrative data sources to build a NH-level dataset that included antipsychotic, BZD, and opioid prescribing rates to residents with ADRD and a host of NH-level characteristics. We used the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines for cross-sectional studies. This study was approved by the Institutional Review Board at the University of Pennsylvania.

Data Sources

To compile the analytical dataset, we relied on a unique beneficiary identifier to merge Medicare claims, which included information on ADRD diagnoses and prescription drug fills, with the Minimum Data Set 3.0 (MDS), which included regular clinical assessments of all NH residents in Medicare and Medicaid-certified NHs. Using the unique facility identifier in the MDS, we were able to group Medicare beneficiaries into NHs and aggregate our resident-level data to the NH level.

We then merged in NH-level data from LTCFocus, which contains yearly data regarding organizational characteristics and quality and is sponsored by the National Institute on Aging (1P01AG027296) through a cooperative agreement with the Brown University School of Public Health.²⁵ LTCFocus includes a variety of NH-level characteristics and aggregates data from the MDS as well as Nursing Home Compare and the Online Survey Certification Automated Records, a data network that tracks quality in NHs.^26,27

We also collected resident information from the Medicare Beneficiary Summary File to determine demographic characteristics of the ADRD study sample, which was aggregated to the NH level and included average age, the share of female residents, and the share of Black residents, Hispanic residents, and residents of other races.

Setting and Participants

Our study population included Medicare beneficiaries with a diagnosis of ADRD during the calendar year 2017 who were residents of a NH. We included Medicare beneficiaries ≥65 years of age with continuous fee-for-service, i.e., the beneficiary had Medicare Part A (inpatient), Medicare Part B (outpatient), and Medicare Part D (pharmacy) enrollment for all of 2017 or up until the month of their death. We excluded Medicare beneficiaries with Medicare Part C, also known as Medicare Advantage, which refers to private health plans who are contracted out by Medicare to cover inpatient and outpatient care. ADRD diagnoses were determined by the presence of an International Classification of Disease, 10^th Edition (ICD-10) code on at least one Medicare inpatient or outpatient claim in 2017.²⁸ The list of ADRD diagnoses is available in the supplemental appendix (Table S1).

Using the MDS, we limited the sample to Medicare beneficiaries with ADRD with NH stays longer than 90 days, determined by the presence of at least one quarterly or annual MDS assessment.²⁹ Medicare beneficiaries who met the above criteria were assigned to a NH using unique facility-level identifiers in the MDS. We excluded those residents in NHs with fewer than 10 residents with ADRD during the study period.

Outcome Variables

Antipsychotic, BZD, and opioid prescription fills were identified using generic names in Medicare pharmacy claims; the list of generic names was based on the American Hospital Formulary System (Table S2). To ensure that we measured prescriptions filled during NH stays, we restricted prescriptions to those with a fill date that occurred after the NH entry date and, when appropriate, before the discharge date, which were available in the MDS. Because our analysis was at the NH level, we computed the share of residents with ADRD in NHs who received at least one prescription for each drug class during 2017.

In regression analyses, our dichotomous outcome variables identified whether the share of ADRD residents at a given NH with an antipsychotic, BZD, or opioid prescription was in the top or bottom quartile of the NH prescribing distributions in 2017. For example, we computed the share of ADRD residents at each NH with at least one BZD fill in 2017 and calculated quartiles of the NH-level share of residents prescribed BZDs, then derived two binary outcomes: one identifying NHs in the top 25% of the share of ADRD residents with a BZD (1 if in the top quartile; 0 otherwise) and another identifying NHs in the bottom 25% of the BZD prescribing distribution (1 if in the bottom quartile; 0 otherwise). These steps were repeated for opioids and antipsychotics, resulting in six binary NH-level outcomes indicating if NHs were top-quartile or bottom-quartile prescribers of antipsychotics, BZDs, or opioids. We chose to use quartiles as distinct outcomes to assess whether the relationships between NH characteristics and prescribing outcomes were symmetric.³⁰

Covariates of Interest

NH characteristics were selected based on the following criteria: (1) a conceptual link to prescribing outcomes, as established by the literature; (2) the extent of missing data (LTCFocus requires 4,500 patient-bed days for variables to be populated); and (3) collinearity with other covariates. The final set of NH characteristics included total beds, occupancy rate, payer mix (%Medicaid), for-profit status, whether the NH was hospital-based or part of a chain, whether the NH had an Alzheimer’s unit, the ratio of registered nurses to registered nurses plus licensed practical nurses, and the mean hours of direct care provided to residents. We also included the NH’s acuity index, which was calculated by LTCFocus using the share of eating-assisted residents, toileting-assisted residents, transfer-assisted residents, bedfast residents, chairbound residents, residents receiving respiratory care, residents receiving suctioning, and residents receiving intravenous therapy, tracheostomy care, tube feedings, and physical/occupational/speech therapy.

Continuous covariates were standardized to mean zero and a standard deviation of one to allow us to directly compare magnitudes. We did not standardize the four binary covariates (i.e., whether the NH had an Alzheimer’s unit, whether the NH was for-profit, whether the NH was part of a chain, and whether the NH was hospital-based), so coefficients reflect the difference in likelihood that the NH is in the top or bottom quartile of NH-level prescribing relative to the reference group (i.e., NHs without an Alzheimer’s unit, not-for-profit NHs, NHs that were part of a chain, and hospital-based NHs).

Statistical Analysis

First, we measured the share of residents with ADRD in NHs who filled at least one antipsychotic, BZD, or opioid prescription during 2017. We compared prescribing outcomes and covariates across the four quartiles of antipsychotic prescribing and assessed group differences using analysis of variance (ANOVA).

To identify which NH characteristics were associated with prescribing for a given drug class, we estimated linear probability models (LPMs), which are becoming increasingly common in the clinical and health services literature.^31–34 Unless an outcome variable is highly skewed, LPMs perform similarly to logistic regressions.^35,36 An advantage is that LPM coefficients represent risk differences, which are easier to interpret than odds ratios as, for a given covariate, they reflect the change in the probability that Y = 1 for a one-unit change in the covariate, holding everything else constant. For a continuous variable, the coefficient therefore represents the difference in risk of the outcome (e.g., being in top prescribing quartile) for every one unit increase in the predictor; for categorical variables, the coefficient represents the difference in risk of the outcome comparing the given level of the variable to the reference group.

We fit the following model:

$${\mathbf{\text{Rx}}}_{\text{hs}}=\text{a}+{\text{b}\mathbf{\text{F}}}_{\text{hs}}+{\text{cS}}_{\mathbf{\text{S}}}+{\text{e}}_{\text{s}}$$

where Rx_hs is a vector of NH-level prescribing outcomes of interest (i.e., binary indicators of a given NH being in the top or bottom quartile of antipsychotic, BZD, and opioid prescribing) for nursing home h in state s, and F is a vector of NH characteristics as described above. State fixed effects, captured by S, were included to account for unmeasured state-level variation that may explain differences in prescribing rates across states, as states are responsible for monitoring NHs in compliance with federal standards.³⁷ We clustered standard errors at the state-level to further account for correlation in prescribing within the same state.³⁸ In supplemental analyses, we estimated logistic regressions with state-clustered standard errors and instead presented odds ratios that measured the association between NH characteristics and the six prescribing outcomes.

Finally, we explored if NHs in the top (bottom) quartile of antipsychotic prescribing were more likely to be in the top (bottom) quartile of BZD and opioid prescribing. We plotted the distribution of quartiles of BZD and opioid prescribing across the four quartiles of antipsychotic prescribing (e.g., among NHs in each quartile of antipsychotic prescribing, we present the percentage that were in each quartile of BZD prescribing). To gauge whether relationships were significant, we also fit a LPM where the outcome was being in the top (bottom) quartile of BZD or opioid prescribing and the independent variable of interest was whether it was in the top (bottom) quartile of antipsychotic prescribing, adjusting for the covariates included in the model above.

Data were cleaned and analyzed using SAS 9.4 and Stata 16.1. All statistical tests were two sided and alpha was set at 0.05.

Results

After excluding NHs with fewer than 10 residents (n=1,360, 9.4% of NHs) and NHs with missing characteristics (n=24, 0.2% of NHs), our study sample included 448,128 Medicare beneficiaries diagnosed with ADRD residing in 13,151 NHs across the U.S.

During 2017, the share of residents with ADRD receiving an antipsychotic prescription was 29.1% on average across NHs; 30.2% of residents with ADRD received a BZD and 40.9% of residents with ADRD received an opioid (Table 1). The average NH-level BZD prescribing rate increased across quartiles of antipsychotic prescribing, but opioid prescribing did not. There were significant differences in most NH characteristics across antipsychotic prescribing quartiles.

Table 1.

Summary Statistics of Prescribing Rates and Characteristics of Nursing Homes used by Residents with Alzheimer’s Disease and Related Dementias in 2017, Overall and by Quartile of Antipsychotic Prescribing

		Distribution of Antipsychotic Prescribing
	All N=13,151	Bottom Quartile N=3,056	Quartile 2 N=3,495	Quartile 3 N=3,484	Top Quartile N=3,116	p-value
	Mean (SD)
Prescribing Outcomes
Share of ADRD residents with 1+ antipsychotic prescription fill	29.1% (14.1)	12.5% (4.9)	23.4% (2.5)	32.3% (2.8)	48.4% (10.5)	<0.001
Share of ADRD residents with 1+ benzodiazepine prescription fill	30.2% (13.5)	26.1% (13.0)	29.6% (12.7)	31.1% (12.8)	33.9% (14.4)	<0.001
Share of ADRD residents with 1+ opioid prescription fill	40.9% (15.3)	41.6% (15.6)	42.3% (14.6)	41.0% (14.8)	38.6% (16.0)	<0.001
Resident Characteristics
Number of unique residents with ADRD	37.1 (23.8)	33.3 (21.4)	40.1 (24.6)	40.8 (25.4)	33.3 (22.2)	<0.001
average age	83.5 (3.4)	84.9 (2.9)	84.2 (2.8)	83.3 (2.9)	81.3 (3.7)	<0.001
% Black	12.1 (18.8)	11.0 (19.0)	11.1 (18.0)	12.5 (18.8)	14.0 (19.3)	<0.001
% Hispanic	2.0 (5.3)	2.2 (5.7)	1.8 (4.8)	1.9 (5.2)	2.0 (5.4)	0.043
% other race	2.9 (8.8)	4.7 (13.8)	2.7 (7.4)	2.3 (6.8)	2.2 (4.8)	<0.001
% female	71.7 (13.1)	73.9 (13.1)	74.3 (11.0)	72.2 (11.6)	66.2 (15.1)	<0.001
Nursing Home Characteristics
acuity index	12.2 (1.2)	12.3 (1.2)	12.3 (1.1)	12.3 (1.1)	12.0 (1.4)	<0.001
% Medicaid	61.9 (20.4)	56.1 (21.5)	59.9 (19.5)	63.1 (19.2)	68.7 (19.4)	<0.001
RN:RN+LPN ratio	0.33 (0.19)	0.36 (0.19)	0.34 (0.18)	0.33 (0.18)	0.31 (0.18)	<0.001
hours of direct care	3.56 (0.83)	3.72 (0.85)	3.61 (0.83)	3.53 (0.78)	3.39 (0.84)	<0.001
% occupancy	81.2 (14.1)	82.4 (13.7)	82.1 (13.4)	81.2 (14.0)	79.1 (15.2)	<0.001
total beds	113.4 (61.0)	106.3 (60.9)	116.8 (62.0)	117.7 (59.1)	111.7 (61.3)	<0.001
% Alzheimer’s unit	15.7 (36.4)	10.9 (31.1)	15.9 (36.5)	18.4 (38.7)	17.2 (37.7)	<0.001
% multiple facilities	58.2 (49.3)	56.5 (49.6)	57.6 (49.4)	59.3 (49.1)	59.1 (49.2)	0.083
% for-profit	71.3 (45.2)	65.3 (47.6)	68.3 (46.5)	72.4 (44.7)	79.4 (40.5)	<0.001
% hospital-based	2.9 (16.7)	3.2 (17.6)	3.2 (17.6)	2.4 (15.3)	2.7 (16.3)	0.120

Notes. ADRD refers to Alzheimer’s disease and related dementias, RN refers to registered nurse, and LPN refers to licensed practical nurse. Standard deviations are in parentheses. The p-value column presents the p-value from the analysis of variance test examining if the expected characteristic (e.g., average age) differed across quartiles of antipsychotic prescribing rates (i.e., expected characteristic in at least one antipsychotic quartile differed from the others). The acuity index was based on the share of eating-assisted residents, toileting-assisted residents, transfer-assisted residents, bedfast residents, chairbound residents, residents receiving respiratory care, residents receiving suctioning, and residents receiving intravenous therapy, tracheostomy care, tube feedings, and physical/occupational/speech therapy. Nursing home prescribing rates and resident characteristics were computed only for residents with ADRD.

Sources. Prescribing outcomes are from 2017 Medicare pharmacy claims, resident characteristics are from the Medicare Beneficiary Summary File, and facility characteristics are from LTCFocus.

Selected NH characteristics were consistently associated with top-quartile prescribing across the three drug classes (Table 2). For example, in NHs with an average resident age that was one standard deviation older than the mean (i.e., 3.4 years), the probability of being in the top quartile of prescribing was 16.6 (p<0.001), 6.8 (p<0.001), and 1.3 (p=0.039) percentage points lower for antipsychotics, BZD, and opioids, respectively. Having more beds and a higher proportion of Black and other non-white residents with ADRD was also associated with lower probabilities of NHs being in the top quartile, while a larger share of Medicaid-enrolled residents was positively associated with the NH being in the top quartile.

Table 2:

Linear Probability Model Estimates of Nursing Home Characteristics Associated with Being in the Top Quartile of Antipsychotic, Benzodiazepine, or Opioid Prescribing to Residents with Alzheimer’s Disease and Related Dementias, 2017

	Outcome
	Top Quartile of Antipsychotic Prescribing	Top Quartile of Benzodiazepine Prescribing	Top Quartile of Opioid Prescribing
	Coefficient (p-value)
Average age	−16.58***	−6.76***	−1.32*
	(<0.001)	(<0.001)	(0.039)
Share of Black residents	−3.67***	−9.02***	−5.44***
	(<0.001)	(<0.001)	(<0.001)
Share of Hispanic residents	0.23	−2.06*	−2.33***
	(0.766)	(0.020)	(<0.001)
Share of other race	−1.41*	−2.44***	−3.36***
	(0.035)	(<0.001)	(0.001)
Share of female residents	−3.84***	1.44*	4.29***
	(<0.001)	(0.016)	(<0.001)
Acuity index	−2.08**	1.36*	0.02
	(0.005)	(0.022)	(0.962)
% Medicaid	1.72**	1.78*	1.18*
	(0.003)	(0.012)	(0.028)
RN:RN+LPN ratio	−0.78	−0.70	0.14
	(0.118)	(0.328)	(0.827)
Hours of direct patient care	−0.52	0.91	1.30*
	(0.318)	(0.084)	(0.014)
Occupancy rate	−1.92***	−0.18	−0.61
	(<0.001)	(0.693)	(0.233)
Total beds	−3.03***	−2.26**	−1.27**
	(<0.001)	(0.002)	(0.004)
Alzheimer’s unit†	2.95**	2.21	−2.78**
	(0.008)	(0.165)	(0.008)
Multiple facilities†	−3.16**	−2.02*	1.56
	(0.005)	(0.036)	(0.079)
For-profit†	−0.21	−1.29	−0.78
	(0.803)	(0.325)	(0.519)
Hospital-based†	5.95*	2.79	10.05**
	(0.018)	(0.183)	(0.004)
Number of facilities	13,151	13,151	13,151

Notes:

^***p<0.001,

^**p<0.01,

^*p<0.05.

ADRD: Alzheimer’s disease and related dementias; RN: registered nurse; LPN: licensed practical nurse. Estimates are based on linear probability models with state-clustered standard errors. Continuous variables are standardized and coefficients are in percentage points. State fixed effects are included but not reported. Prescribing outcomes are from 2017 pharmacy claims, resident characteristics (age, race/ethnicity, and gender) are from the Medicare Beneficiary Summary File, and NH characteristics are from LTC Focus.

^†Variables are binary. Reference categories are NHs without an Alzheimer’s unit, NHs that are not part of a chain, not-for-profit NHs, and NHs that are not hospital-based.

Other NH characteristics were inconsistently associated with prescribing outcomes. The share of females among the ADRD population was positively associated with being in the top quartile of BZD and opioid prescribing (p=0.016 and p<0.001, respectively), but negatively associated with being in the top quartile of antipsychotic prescribing (p<0.001). Having an Alzheimer’s unit was associated with an increased probability of being a top-quartile antipsychotic NH (p=0.008), a decreased probability of being a top-quartile opioid facility (p=0.008), and was unrelated to BZD prescribing (p=0.165).

NH characteristics that correlated with NHs being in the bottom quartile of prescribing were mostly symmetric to the correlates of top-quartile prescribing (Table 3). One exception was nursing care. While the ratio of registered nurses (RNs) to RNs plus licensed practical nurses was not significantly associated with being a top-quartile prescribing NH for any drug class, a larger proportion of RNs was significantly associated with bottom-quartile prescribing of all three classes (antipsychotic [p=0.035], BZD [p=0.005], and opioid [p=0.017]). Findings were consistent when estimating logistic regressions (Tables S3 and S4).

Table 3:

Linear Probability Model Estimates of Nursing Home Characteristics Associated with Being in the Bottom Quartile of Antipsychotic, Benzodiazepine, or Opioid Prescribing to Residents with Alzheimer’s Disease and Related Dementias, 2017

	Outcome
	Bottom Quartile of Antipsychotic Prescribing	Bottom Quartile of Benzodiazepine Prescribing	Bottom Quartile of Opioid Prescribing
	Coefficient (p-value)
Average age	11.91***	5.98***	0.90
	(<0.001)	(<0.001)	(0.267)
% Black	4.07***	10.79***	6.75***
	(<0.001)	(<0.001)	(<0.001)
% Hispanic	−0.04	2.27*	6.05***
	(0.952)	(0.033)	(<0.001)
% other race	3.23***	6.74***	6.91***
	(<0.001)	(<0.001)	(<0.001)
% female	−0.73	−3.46***	−5.57***
	(0.237)	(<0.001)	(<0.001)
Acuity index	0.14	−0.88	−1.26
	(0.797)	(0.190)	(0.067)
% Medicaid	−2.43**	−1.33*	0.51
	(0.001)	(0.012)	(0.215)
RN:RN+LPN ratio	1.14*	1.78**	2.56*
	(0.035)	(0.005)	(0.017)
Hours of direct patient care	1.01	−0.24	−0.44
	(0.050)	(0.635)	(0.446)
Occupancy rate	0.95*	−0.36	−0.15
	(0.050)	(0.387)	(0.861)
Total beds	−1.10	−0.55	0.95
	(0.099)	(0.532)	(0.295)
Alzheimer’s unit†	−7.04***	−4.66***	1.57
	(<0.001)	(<0.001)	(0.232)
Multiple facilities†	2.31**	−0.03	−3.09*
	(0.005)	(0.967)	(0.031)
For-profit†	−1.23	0.78	2.67
	(0.319)	(0.458)	(0.056)
Hospital-based†	−3.32	−0.34	−5.26**
	(0.105)	(0.895)	(0.003)
Number of facilities	13,151	13,151	13,151

Notes:

^***p<0.001,

^**p<0.01,

^*p<0.05.

ADRD: Alzheimer’s disease and related dementias; RN: registered nurse; LPN: licensed practical nurse. Estimates are based on linear probability models with state-clustered standard errors. Continuous variables are standardized and coefficients are in percentage points. State fixed effects are included but not reported. Prescribing outcomes are from 2017 pharmacy claims, resident characteristics (age, race/ethnicity, and gender) are from the Medicare Beneficiary Summary File, and NH characteristics are from LTCFocus.

^†Variables are binary. Reference categories are NHs without an Alzheimer’s unit, NHs that are not part of a chain, not-for-profit NHs, and NHs that are not hospital-based.

The extent of BZD prescribing tracked closely with antipsychotic prescribing. Specifically, 34.4% of top-quartile antipsychotic NHs were also top-quartile BZD NHs and, among the bottom quartile of antipsychotic NHs, 34.0% were also bottom-quartile BZD NHs (Figure 1). Linear probability models demonstrated that NHs in the top quartile of antipsychotic prescribing were significantly more likely to be in the top quartile of BZD prescribing (coefficient=11.1 percentage points; p<0.001), while NHs in the bottom quartile of antipsychotic prescribing were significantly more likely to be in the bottom quartile of BZD prescribing (coefficient=8.6 percentage points; p<0.001).

Figure 1.

Distribution of Nursing Home Benzodiazepine (BZD) and Opioid Prescribing Quartiles by Antipsychotic Prescribing Quartile Among Older Adults with ADRD, 2017

Opioid prescribing did not track with antipsychotic prescribing as consistently. In the bottom quartile of antipsychotic prescribing, 22.3% of NHs were bottom-quartile opioid prescribers; in the top quartile of antipsychotic prescribing, 30.6% were bottom-quartile opioid prescribers. However, linear probability models showed that NHs in the top quartile of antipsychotic prescribing were less likely to be in the top quartile of opioid prescribing (coefficient=−4.4 percentage points; p=0.001), and NHs in the bottom quartile of antipsychotic prescribing were slightly less likely to be in the bottom quartile of opioid prescribing (coefficient=−2.3 percentage points; p=0.039).

Discussion

Conducting a cross-sectional analysis of Medicare beneficiaries with ADRD residing in NHs in 2017, we found that the average NH-level prescribing rate of BZDs to residents with ADRD was slightly higher than the average NH-level antipsychotic prescribing rate (30.2% vs. 29.1%), while the average rate of opioid prescribing was over 40%. We also found that selected NH characteristics were associated with high and low levels of prescribing across drug classes, and that antipsychotic prescribing overlapped more with BZD prescribing than opioid prescribing.

While other studies have demonstrated that NH characteristics are associated with the use of individual drug classes, we found that some characteristics were consistently associated with NH-level antipsychotic, BZD, and opioid prescribing. There were three notable results. First, larger NHs and NHs with fewer Medicaid-enrolled residents were less likely to be in the top quartile of prescribing. While the payer mix finding aligns with prior work focused on antipsychotics, our results demonstrate that the pattern extends to BZDs and opioids as well.³⁹ This may relate to resource constraints that smaller NHs face compared to the economies of scale of larger NHs, or to resource constraints in NHs with a higher share of Medicaid-enrolled residents due to lower reimbursement.⁴⁰ It could be that NHs with fewer resources find it more difficult to provide behavioral interventions or nonpharmacological pain management, opting instead for medications to address symptoms.

Given that one of the most important resources in NHs is nursing care, a second notable finding was the positive association between the ratio of RNs to RNs plus licensed practical nurses and being in the bottom quartile of prescribing for all three drug classes. However, the inverse was not true—this ratio was not associated with being in the top prescribing quartile—which shows how the relationship between NH characteristics and prescribing may be inconsistent across its distribution. These findings suggest that more (registered) nursing care could offset the use of medications, perhaps by allowing for more behavioral interventions or nonpharmacological pain management.

A third finding was the relationship between NH resident characteristics and prescribing outcomes. There were stark discrepancies by age, gender, and race/ethnicity, which may reflect disparities in the quality of care delivered. Aligning with prior work focused on antipsychotics, we found that a larger proportion of non-white residents was associated with a lower likelihood of top-quartile antipsychotic prescribing, which extended to BZD and opioid prescribing.³⁹ For BZDs and opioids, both of which are controlled substances, there is evidence of under-prescribing to Black and Hispanic NH residents, even after accounting for clinical factors, which could point to potential racial biases among clinicians.^41,42 While there are safety concerns related to the use of these medications in older adults, lower levels of prescribing to Black and Hispanic residents should not necessarily be interpreted as more appropriate care.

In terms of an organizational culture of prescribing across drug classes, there was a clear relationship between antipsychotic and BZD prescribing at the NH level. Overlap in the proclivity toward certain drug classes has implications for interventions to improve appropriate prescribing, and further highlights the importance of not monitoring antipsychotics in a vacuum. In contrast, opioid prescribing did not appear to follow antipsychotic prescribing closely, suggesting that the factors that influence pain management differ from those that influence the treatment of behavioral symptoms. It could also be that more effective pain management, as demonstrated by appropriate opioid use, may result in less antipsychotic use by ameliorating behavioral and psychological symptoms of dementia.^43,44

We face several limitations. Given the cross-sectional analyses, we are only able to estimate associations between NH characteristics and prescribing outcomes and do not assess whether the association is dynamic over time or whether the effect sizes are clinically meaningful. There may also be interaction effects between different NH characteristics, such as nursing care and Medicaid payer mix, that are worth exploring. Because we are using claims, we further assume that prescription fills can proxy for medication use and are unable to discern prescribing indications and the appropriateness of medication use. While we do control for patient acuity at the NH level, the selection of residents into NHs may be driven by clinical need, with use of these medications driven by the types of residents that select particular NHs. For example, higher rates of antipsychotic prescribing in facilities with an Alzheimer’s unit may reflect a resident population with more behavioral symptoms.

Furthermore, our generalizability is limited by several aspects of our cohort design, which includes only older adults with ADRD diagnoses who were enrolled in Medicare fee-for-service with Part D coverage and who resided in NHs with at least 10 residents with ADRD. Smaller NHs may have uniquely different organization cultures that are a function of their size. On one hand, staff in NHs with few residents with ADRD may be less familiar with appropriate care for this patient population, leading to more prescribing; alternatively, in a small NH, staff may know patient needs particularly well and prescribe less. We also note that estimating models at the NH level could create aggregation bias; future studies conducted at the resident-level could help contribute to our understanding of the organizational culture of prescribing.

Conclusions and Implications

Using a national sample of NHs from 2017, we found that BZD prescribing to residents with ADRD was slightly more common than antipsychotic prescribing, while the rate of opioid prescribing was over ten percentage points higher. We also found that certain NH characteristics were consistently associated with high and low rates of prescribing across drug classes, and that there was substantial overlap in antipsychotic and BZD prescribing. These findings contribute to the broader literature on prescribing in NHs by providing support for the idea of an organizational culture of prescribing that spans multiple drug classes. Future initiatives focused on prescribing quality might benefit from more investigations that explore the use of multiple drug classes to best identify targets for quality improvement.