Medicare Part D’s Exclusion of Benzodiazepines and Fracture Risk in Nursing Homes

Abstract

BACKGROUND

Medicare Part D excludes benzodiazepines from coverage and some state Medicaid programs also limit coverage. We assessed whether such policies decrease the risk of fractures in the elderly living in nursing homes.

METHODS

This is a quasi-experimental study with interrupted time-series estimation and extended Cox proportional hazards models comparing changes in outcomes pre- and post-Part D in a nationwide sample of nursing home residents in 48 states. The study included 1,068,104 residents and a subsample of 50,874 residents with fracture data from one pharmacy provider. We assessed monthly prescribing rates of benzodiazepines and potential substitutes over the period 2005- June 2007, and hazard ratios for incident hip fracture and falls, adjusted for age, sex, and race. Estimates were stratified by concurrent Medicaid limits on benzodiazepines: no coverage (n=1 state) or partial coverage (n=6 states), versus complete coverage (n=41 states).

RESULTS

The no coverage policy resulted in an immediate and significant reduction of 10 absolute points in benzodiazepine use (27% to 17%) after Part D (95% confidence interval [CI]: −.11- −.09, p<.0001). Benzodiazepine use remained stable in the partial and complete coverage states. Hazard ratios for incident hip fracture were 1.60 (95% CI: 1.05-2.45, p=.030) in the no coverage state after Part D, and 1.17 (95% CI: 0.93-1.46, p=.179) in the partial coverage states, relative to complete coverage states.

CONCLUSIONS

This study found that drug coverage exclusion policies affect the medication use of nursing home residents and may not decrease their fracture risk.

Introduction

Controversy continues to surround the safety of benzodiazepines in nursing homes (NH), even as up to 30% of residents receive these agents.¹ Benzodiazepines have well-documented sedative effects that can impair cognition, balance, and produce daytime somnolence.² Benzodiazepines have also been associated with an increased risk for falls and fractures in NHs although the findings are inconsistent.^3-6 Additionally, abrupt discontinuation of benzodiazepines, especially long-term use, has been associated with confusion, elevated systolic blood pressure, and, in rare cases, seizures.^7-9 However, no data have indicated that policies to limit benzodiazepine use in older institutionalized adults result in reduced risk of falls and injury.^{7, 10}

In 2006, the Medicare Part D prescription drug program became the main source of drug coverage for nursing home residents, but with a restrictive drug policy that excluded all benzodiazepines from reimbursement. State Medicaid programs may offer supplemental coverage for benzodiazepines, and all states except for Tennessee cover all or some benzodiazepines. Approximately 80% of Medicare-eligible NH residents enrolled in Medicare Part D in 2006, and 62%-69% of NH residents are eligible for state Medicaid supplemental coverage.¹¹

Medicare’s restrictive drug policy and the variation of states’ supplemental coverage policies offer a rare opportunity to assess the impact of policies limiting benzodiazepines and their impact on fracture risk in the NH setting. This vulnerable population is rarely available for policy evaluation and the combination of federal and state policies creates the conditions for a large-scale natural experiment. We hypothesize that an abrupt decrease in benzodiazepine prescribing will coincide with the implementation of Medicare Part D. However, this decrease will be smaller in states with partial supplemental coverage and smallest in states with complete supplemental coverage. Compensatory increases will occur in substitute medications reimbursable by Medicare. Furthermore, a reduction in the risk for falls and hip fractures will be associated with decreases in benzodiazepine use.

Methods

Design

We used a quasi-experimental research design that employed both a comparator group and pre-tests in a longitudinal cohort of NH residents. We compared changes in resident outcomes pre- and post-Part D in states with partial or no supplemental coverage for benzodiazepines to the same changes in outcome in states with complete supplemental coverage (comparator).

Study population

We analyzed January 2005-June 2007 prescription drug dispensing records merged with Minimum Data Set (MDS) records. The prescription data come from over 2.5 million unique individuals living in approximately 16,000 NHs arrayed across 48 states. These individuals have a variety of prescription drug plans, including private insurance, Medicaid, Medicare Part D, or no drug coverage, but their medications come from a single pharmacy provider. The drug dispensing data include all medications prescribed and administered to residents, including over-the-counter drugs and PRN medications (as needed). Research using these data has been previously described.¹¹ The data elements include the product identification code (national drug code), date of fill, days’ supply, quantity dispensed, and payment source, as well as basic demographics, limited to gender and age, and the state where the NH was located. Linkable MDS records come from the same source and are available for about a third of individuals with prescription data. MDS is a nearly universal health assessment tool used in NHs and it captures over 300 items on residents’ physical and cognitive functioning. Full assessments occur upon admission, significant change in status, and at annual re-evaluations.¹²

For the sample, we excluded individuals who were not Medicare eligible (n=597,211) or observed for less than 4 months (n=1,870,850). Residents whose stay in the NH was less than 4 months may have their medications paid for by Medicare Part A, rather than Part D, as part of bundled per diem payments to the facility. In addition, prior research shows that short-stayers in NHs differ significantly from long-stay residents.¹³ We also identified a subsample of newly-admitted NH residents (n=50,874) whose drug dispensing data were linkable to MDS records that included full assessments of fracture outcomes for selected analyses.

State of residence came from the location of the nursing facility. Each state was characterized by its Medicaid reimbursement policy for benzodiazepines as: 1) complete supplemental coverage (n=41 states); 2) partial supplemental coverage (n=6 states: Alabama, California, Georgia, Illinois, Kansas and Missouri); and no supplemental coverage (n=1 state: Tennessee.) Partial supplemental coverage included the following restrictive policies on benzodiazepines: prior authorization, quantity limits, and preferred drug lists. All of the states implemented their benzodiazepine supplemental coverage policy to coincide with initiation of the Part D program.

Outcome measures

Use of study medications was measured as any dispensing during the month and also average number of prescriptions dispensed per month of use. Benzodiazepines included: alprazolam, clonazepam, estazolam, flurazepam, halazepam, lorazepam, oxazepam, quazepam, temazepam, triazolam, chlordiazepoxide, clorazepate, diazepam, and diazepam combinations. Potential substitute medications included the Medicare-covered categories of non-benzodiazepine sedative/hypnotics (e.g., zolpidem tartrate, phenobarital, eszopiclone, zaleplon, melatonin, and chloral hydrate), other anxiolytics (buspirone, hydroxyzine, and meprobamate), and antipsychotics (e.g., aripiprazole, clozapine, olanzapine, and haloperidol), as identified in previous research.¹⁴ (Complete list of drug names available from authors).

Outcomes of interest were measured using the MDS: falls in the past 180 days, hip fractures in last 180 days, and other factures in the last 180 days. The MDS documentation of hip fracture has demonstrated high concordance with medical claims data (>80%), while the documentation of falls has shown fair concordance with chart abstractions of 65% to 75%.^{15, 16}

Statistical analysis

We conducted descriptive analyses comparing baseline demographics and pre-Part D drug prescribing by state supplemental coverage policy. Interrupted times series estimation with segmented regression methods and autoregressive correlations of the first order were used for testing changes in the trend (slope and level) of medication use following the implementation date of the Part D program, and controlling for pre-policy trends.¹⁷ The basic model includes a constant summarizing the baseline level, and three terms. The first term estimates monthly changes per person in the pre-policy period, the second term estimates the average level change per person in the first month after Part D, and the third term is the post-Part D trend relative to the pre-Part D trend. This model tests the population-level effects of Part D regardless of enrollment status, thus avoiding the confounding introduced by comparing effects across non-randomly assigned groups.¹⁷ We assume that the timing of the new Part D program is independent of the factors that determine treatment assignment and confound simplistic outcome assessments. If Part D had any impact, the population-level observations after January 2006 will display a distinctly different pattern from the pre-Part D pattern. Furthermore, the difference will be greatest where there is no Medicaid supplemental coverage to moderate the effect.

Extended Cox proportional hazards models with heaviside function were used to estimate hazard ratios for fractures before and after the implementation of Medicare Part D.¹⁸ The basic model includes time fixed terms for gender, age, race, ethnicity, and a time-varying term for supplemental coverage policy. A heaviside function is used to provide two hazard ratios for supplemental coverage policy that correspond to the time intervals of pre-Part D (January 2005-December 2005) and post-Part D (January 2006-December 2006). Unlike a standard Cox model, this model allows for changes in the proportional hazards over time.¹⁸

All multivariate models were estimated separately for each dependent variable (monthly proportion of residents who received benzodiazepines and each of the substitute drug categories, monthly average number of prescriptions dispensed, incidence of falls, and incidence of hip fracture, and incidence of other fracture. All multivariate analyses were conducted using STATA 10.0. The institutional review board of the University of Massachusetts Medical School exempted this research from review.

Results

We identified 1,068,104 long-stay Medicare enrollees in NHs nationwide who generated 145,254,104 prescription records during the study observation. Table 1 shows the characteristics of the study population and the baseline prescribing of the study medications. In 2005, over 80% of the study NH residents lived in states that would offer complete supplemental coverage of benzodiazepines after implementation of Part D, and 16% lived in states offering partial coverage; only Tennessee did not offer any supplemental coverage (n=15,733). A comparison of age and gender showed similar distributions across the three policy groups.

Table 1.

Characteristics of the Study Population, 2005

	Complete Supplemental Coverage	Partial Supplemental Coverage	No Supplemental Coverage
All, (n)	882,266	170,105	15,733
Age, (%)
<65	12.0	18.0	10.7
65-74	14.8	15.5	17.9
75-84	36.7	34.9	39.2
85+	36.5	31.5	32.2
Gender, (%)
Male	30.8	32.6	30.1
Female	69.2	67.4	69.9
Baseline Use of Medications (January 2005)
Prevalence of Benzodiazepines, (%)	15.6	16.4	26.0
Average # of Rx per user, (SE)	1.6 (.01)	1.3 (.01)	2.0 (.03)
Prevalence of Other Anxiolytics, (%)	3.3	3.3	3.7
Average # of Rx per person (SE)	1.2 (.01)	1.2 (.01)	1.1 (.03)
Prevalence of Sedatives, (%)	4.4	4.0	6.4
Average # of Rx per person (SE)	1.4 (.01)	1.2 (.01)	1.8 (.05)
Prevalence of Antipsychotics, (%)	26.6	33.4	29.9
Average # of Rx per person (SE)	1.5 (.00)	1.7 (.01)	1.5 (.02)

^*Percentages may not equal 100% due to rounding

At baseline, the prevalence of the study medications varied by the state’s supplemental coverage policy. In January 2005, the proportion of study NH residents on benzodiazepine therapy was 26.0% in the no coverage state, 15.6% in complete coverage states and 16.4% in partial coverage states, p<.001. The proportion of study NH residents on non-benzodiazepine sedative/hypnotics was 4.4% in the complete coverage states, 4.0 in the partial coverage states and 6.4% in the no coverage state, p<.001. The prevalence of antipsychotic use was 26.6% in the complete coverage state, 33.4% in the partial coverage state, and 29.9% in the no coverage state, p<.001. Lastly, the baseline prescribing of other anxiolytics ranged from 3.3% to 3.7% among the three groups, p=.564.

Figure 1 shows the changes that occurred in the monthly prevalence of benzodiazepine use and potential substitutes during the observation period. Times-series analyses showed a large and significant decrease of 10 percentage points (27% to 17%) in the proportion of benzodiazepine recipients immediately following the implementation of Part D in the no coverage state (−.10 change in prevalence; 95% confidence interval [CI], −.11- −.09, p<.0001). This large change did not occur in the partial coverage states (−.01 change in prevalence; 95% CI, −.014- −.006, p<.001) or the complete supplemental coverage states (−.01 change in prevalence; 95% CI, −.014- −.004, p<.001). The average monthly number of benzodiazepine prescriptions dispensed per user did not change before and after Part D in any of the three state reimbursement policy groups (data not shown).

Figure 1.

Monthly Use of Benzodiazepines and Potential Substitutes, January 2005-June 2007

No supplemental coverage Partial supplemental coverage Complete supplemental coverage

Times-series analyses of the reimbursable anxiolytics, sedative/hypnotics, and antipsychotics showed some evidence of potential substitution for the benzodiazepines in the no coverage state after implementation of Part D (see Figure 1). For instance, the use of other anxiolytics immediately increased significantly in 2006 (+.02 change in prevalence; 95% CI, .005-.03, p=.007) in the no coverage state relative to 2005. The average monthly number of antipsychotics and other anxiolytics dispensed also increased immediately in 2006 in the no coverage state relative to 2005; antipsychotics, +.66 change in monthly fills; 95% CI, .65-.67, p<.0000; and other anxiolytics, +.57 change in monthly fills; 95% CI, .55-.58, p=.0002. In comparison, these changes did not occur in the other states (data not shown).

Table 2 shows the falls and fracture outcomes for the subgroup of newly-admitted NH residents. Overall, this group experienced 9426 incident fractures (4632 pre-Part D and 4794 post-Part D) and 23,601 incident falls (10,722 pre-Part D and 12,879 post-Part D) while observed. Before implementation of Part D, the rates of fractures were similar to or lower in the no coverage state compared to the rates in the other states. For instance, in 2005 the rate of hip fractures was 6.4 per 100 person-years in the no coverage state compared to 8.9 per 100 person-years in the complete supplemental coverage state and 9.6 per 100 person-years in the partial coverage state, p=.007. After Part D, the rates of fractures increased significantly in the no coverage state compared to the rates in the other states. For instance, the rate of hip fractures doubled from 6.4 in 2005 to 12.4 in 2006 per 100 person-years in the no coverage state. In comparison, there were modest increases of 8.9 in 2005 to 9.9 in 2006 per 100 person-years in the complete supplemental coverage state, and 9.6 to 10.7 per 100 person-years in the partial coverage state, p=.002. The incidence rate of falls showed a similar pattern, although all three groups experienced increases after Part D.

Table 2.

Fracture Outcomes Pre- and Post-Part D, 2005-2006

	Complete Supplemental Coverage		Partial Supplemental Coverage		No Supplemental Coverage
	Pre Part D	Post Part D	Pre Part D	Post Part D	Pre Part D	Post Part D
Individuals, (n)	24,521	22,108	1,551	1,687	546	461
Incident Fractures (any), (n)	4269	4334	275	350	88	110
Incidence rate, per 100 person  years	17.4	19.6	17.7	20.7	16.1	23.9
Incident Hip Fractures, (n)	2189	2185	149	181	35	57
Incidence rate, per 100 person  years	8.9	9.9	9.6	10.7	6.4	12.4
Incident Falls, (n)	9753	11706	739	902	230	271
Incidence rate, per 100 person  years	39.8	52.9	47.6	53.5	42.1	58.8

Multivariate analyses for fracture outcomes showed some significant changes in the hazards ratios after implementation of Part D, but none were in the previously hypothesized direction. Where the large decrease in benzodiazepine use occurred (no coverage state), the hazard ratio for incident hip fractures in NHs increased from 0.74 (95% CI, 0.53-1.0) before Part D to 1.60 (95% CI 1.05-2.45, p=.030) after Part D, compared to the states with stable benzodiazepines use (complete coverage states), see Figure 2. The hazard ratios for falls did not change. In the partial coverage states with no change in benzodiazepine use, the hazard ratios for falls decreased from 1.09 (95% CI, 1.01-1.17, p=.04) to .99 (95% CI, 0.89-1.09) relative to the rates in complete coverage states, although the hazard ratios for hip fractures did not change.

Figure 2.

Adjusted* Hazard Ratios of Fracture Outcomes Pre- and Post-Part D, 2005-2006

*Estimates come from Cox Models adjusted for age, gender, and race/ethnicity. Reference group is NH residents in states with complete supplemental coverage.

Discussion

In this population-based study, we found an abrupt and large discontinuation of benzodiazepines in NHs immediately following the exclusion of these agents by Medicare Part D in the state that did not offer supplemental coverage. Furthermore, NH residents in the same state experienced increases in the use of potential substitute agents, although the increases were temporary and not a complete offset. These responses are similar to those documented in the community-setting after restrictive drug policies were implemented. ^{10, 14} To our knowledge, this is the first analysis to demonstrate these relationships in the NH setting. In contrast, NH residents living in states covering all or some portion of benzodiazepines did not experience the same therapy changes observed in the most restrictive state.

We also showed that the large reductions in benzodiazepines use was not associated with decreased risk of fracture outcomes. Previous restrictive drug policies on the prescribing of benzodiazepines, including prescription limits, prior authorization protocols, and surveillance programs, have also failed to produce a reduction in the risk of fracture outcomes.¹⁰

There are several possible explanations for our finding of no reduction in fracture outcomes. First, the impact of Medicare Part D and state-level policies may have been broad and affected the use of a wide array of agents with a mixture of positive and negative associations with fracture risk. Substitute agents for benzodiazepines include medications with known safety risks for older adults, as in the anxiolytic meprobamate or antipsychotics, as well as medications with unknown safety risks. Newly-marketed non-benzodiazepine sedatives such as eszopiclone and zolpidem tartrate have not been extensively studied in populations of frail elders, limiting our understanding of the potential risks and benefits. Concurrent changes in prescribing of these other agents may have influenced fracture risk and may have masked any benefits from reducing the use of benzodiazepines.

Second, any benefit of reducing benzodiazepines may have been offset by the suddenness of the reduction. Abrupt discontinuation of benzodiazepines, especially long-term use, can pose risks to residents including confusion, elevated systolic blood pressure, and, in rare cases, seizures.^7-9 In ad hoc analysis stratified by duration of benzodiazepine use, we detected abrupt discontinuation in both long-term use (≥120 continuous days of use) and short-term use in our study sample.

Finally, benzodiazepines may not be associated with hip fractures, at least not to the extent reported in other studies. At least 7 studies to date, including studies with prospective cohort and longitudinal, quasi-experimental designs, have failed to find a relationship between benzodiazepine use and fracture risk.^{10, 19-24}

Limitations of the study include the following. First, our analyses were limited to NH residents whose prescriptions were filled by one long-term care pharmacy provider. However, our study reflects the observed experiences of nearly half of the entire US Medicare population living in NHs. Second, we did not have information on some potentially important clinical factors relating to fracture risk such as bone mineral density. Furthermore, we did not have enough information to comment on whether the reduced use of benzodiazepines was clinically appropriate. This limitation also applies to the substitute prescribing. In ad hoc analyses, we found increases in the prescribing of meprobamate in the no coverage state, which is a potentially inappropriate medication for older adults (1.6 dispensing per 1,000 NH residents in 2005 vs. 6.6 dispensing per 1,000 NH residents in 2006).²⁵ Third, we had MDS data for only a third of the sample, although we found no relationship between availability of these data and benzodiazepine use. Lastly, our category of partial supplemental coverage encompassed several restrictive policies that may vary in impacts.

Our approach has several strengths. First, we conducted the analyses of changes in benzodiazepine use using interrupted time-series analyses. The interrupted time-series approach is one of the most powerful quasi-experimental designs since it is robust to many of the threats to the validity of weaker observational designs, particularly in unmeasured changes in the composition of the study population and historical changes in benzodiazepine use. We also used an intention-to-treat analysis, without accounting for actual enrollment into Part D, meaning that we avoided introducing the selection bias inherent in comparing changes in drug prescribing between individuals who selected into Part D and those who did not. Furthermore, the state-level analysis also minimizes any threats to validity based on preferential prescribing patterns related to resident characteristics.

In conclusion, we found that the Medicare Part D’s reimbursement exclusion of benzodiazepines was associated with a significant and abrupt decrease in prescribing of these agents in NHs, if the state did not mitigate the impact by providing partial or complete supplemental coverage with state funds. The reimbursement restriction was not associated with any advantage in patient safety by reducing falls and fracture risk.