Nursing homes underreport antipsychotic prescribing

Abstract

Objective

Determine the accuracy of nursing home self-reported antipsychotic prescribing before and after implementation of a Medicare campaign to reduce use.

Methods

Quasi-experimental study comparing trends in self-reported antipsychotic prescribing relative to claims-based prescribing. Setting is a nationwide sample of 11,912 facilities, 2011–2013. Participants are long-stay nursing home residents (n=586,281) with prescribing data in Medicare Minimum Data Set 3.0 and Medicare Part D claims database. Verified with a pharmacy dispensing database. Main outcomes are the discrepancies in quarterly prevalence of antipsychotic prescribing between nursing home self-reports and claims data and the characteristics of facilities and residents where discrepancies were identified.

Results

Nursing homes underreport their antipsychotic prescribing levels, on average, by 1 percentage point per quarter relative to Medicare Part D claims (0.013, 95% confidence interval (CI), 0.012–0.015; p<.001). After the Medicare campaign, the underreporting gap increased by another half a percentage point (0.004, 95% CI .003–.005; p = .012). Nursing home residents with dementia, Alzheimer’s disease or bipolar disorders were at the highest risk for underreported antipsychotic prescribing before the campaign (Adjusted Odds ratio (AOR) 1.385, 95% CI: 1.330–1.444; AOR 1.234, 95% CI: 1.172–1.300; AOR 1.574, 95% CI: 1.444–1.716, respectively) and afterwards. After the launch of the Medicare campaign, underreported antipsychotic prescribing occurred most in for-profit nursing homes (AOR 1.088, 95% CI: 1.005–1.178) and facilities in the US South (AOR 1.262, 95% CI: 1.145–1.391). Agreement was high between claims and dispensing data (99.7%).

Conclusion

Nursing homes did not identify up to 6,000 residents per calendar quarter as having received antipsychotics despite these prescriptions being paid by Medicare and dispensed by a pharmacy. Nursing home rates of antipsychotic prescribing from self-reported data may be biased.

Introduction

Antipsychotic use remains common in the 700,000 US nursing home patients with Alzheimer’s disease or related dementias despite strong evidence of marginal clinical benefits and serious adverse effects including death (Harris- Kojetin et al., 2016). Since May 2012, the Centers for Medicare and Medicaid Service’s (CMS) has undertaken a national campaign to educate, train and enforce regulations about unnecessary antipsychotic prescribing. The campaign called the National Partnership to Improve Dementia Care in Nursing Homes entailed a well-publicized and multimodal campaign of social media videos, on-line training materials, and outreach to state-level stakeholders to disseminate materials and increase enforcement of regulations (Kuehn, 2013). By July 2012, public reporting began of facility-level antipsychotic prescribing rates and between 2012 to 2013 the number of CMS regulatory violations for inappropriate antipsychotic use increased by nearly 20%. According to CMS within 6 month of the initiative, antipsychotic prescribing in nursing homes decreased by 9% nationwide, and five years later, the levels had been reduced by 33% (Gurwitz, Bonner, & Berwick, 2017).

Our primary understanding of antipsychotic prescribing in nursing homes comes from self-reported rates, which are not independently verified. Nursing homes are required by federal regulation to report all antipsychotic use on standardized resident-level surveys, which are used to calculate summary measures of facility-level antipsychotic prescribing. Facility-level antipsychotic prescribing is factored into quality ratings and displayed on publicly-available website ‘Nursing Home Compare.’ Facilities with high antipsychotic prescribing are targeted for outreach by CMS and state coalitions. State surveyors review a 5% sample for compliance with federal regulations; however, there are no additional audits to ensure completeness of reporting. Subjective reporting of provider performance in public forums is vulnerable to misreporting (Berenson & Rice, 2015). This study compares nursing home self-reported antipsychotic prescribing to objective data of actual antipsychotic consumption before and after CMS’s antipsychotic campaign.

Methods

Using a quasi-experimental population-based study design, we assessed the accuracy of nursing home reported antipsychotic use before the CMS campaign (January 1, 2011–April 30, 2012) and afterwards (May 1, 2012–December 31, 2013). We examined three data sources: 1. Minimum Data Set (MDS) 3.0: federally-mandated surveys of nursing home resident health and care that occur on admission, with significant change in status, and at quarterly reevaluations; 2. Medicare Part D claims, which capture all Medicare-paid prescriptions; and 3. all-payer prescription dispensing database from the largest pharmacy serving the long-term care market. This database contains prescribing information from about half of all US nursing homes. Prescription data are uniquely comprehensive because they include all drugs dispensed regardless of payer (e.g., Medicare Part D, private insurance, and out-of-pocket). We linked these data together and added information on facility characteristics from CMS’s Certification and Survey Provider Enhance Report (Nursing Home Compare Datasets, 2018).

Study population

The sample included nursing home residents with data in all three databases and a long-stay in the facility (>100 consecutive days). In accordance with CMS’ regulations of appropriate users of antipsychotics, we excluded residents diagnosed with schizophrenia, Huntington’s disease or Tourette syndrome from the main analyses. However we conducted sensitivity analyses in this population to assess potentially unintended changes in these diagnoses and their use of antipsychotics (see technical appendix) (RTI International, 2017).

This study was approved by the institutional review board at Northeastern University.

Primary outcome measures

We calculated quarterly rates of antipsychotic prescribing as the proportion of nursing home residents receiving antipsychotics over all eligible residents and compared the rates by data source. In the MDS data, antipsychotic use is captured as a > 0 reply to the question, ‘Indicate the number of days the resident received [antipsychotics] during last 7 days or since admission/entry or re-entry’ on any survey filled out during the quarter. In the Part D data, antipsychotic use is captured as at least 1 claim for any antipsychotic prescription within a given quarter. We also examined the ‘days supply’ field to account for prescriptions spanning more than 1 quarter. Underprescribing is the difference between MDS rates and Part D rates.

Statistical analyses

We estimated interrupted time series models to test for temporal changes between the quarterly MDS and Part D antipsychotic prescribing rates and relative to the timing of the CMS campaign. This quasi-experimental method controls for historical trends and major time-invariant threats to validity (Shadish, Cook, & Campbell, 2002). Generalized estimation equations models were estimated for clustered data to identify predictors of antipsychotic underreporting. Diagnostic analyses indicated missing data in the first quarter so models were conducted with and without this data point (See Supplemental Table S1, supplementary material). Preliminary analyses showed Part D claims lacked >0.3% of all antipsychotics dispensed so results from prescription dispensing database are reported in only the technical appendix (Supplemental Table S1, supplementary material).

Results

Table 1 shows our study sample of 586,281 nursing home residents in 11,912 facilities nationwide. The majority were aged 75 or older (76.3%), female (69.7%) and white (80.5%); nearly half had dementia (48.5%), with Alzheimer’s disease indicated specifically for 17.6%. Most of the nursing homes were in the southern US census region (34.4%), served 50–100 residents (44.8%), and were operating as for-profit entities (73.9%).

Table 1.

Characteristics of Study Nursing Nome Residents and Facilities, 2011–2013.

Resident demographics	No. (%) of residents (n = 586,281)	Facility characteristics	No. (%) of facilities (n = 11,912)
Age		US Census Region
<65	45,746 (7.80%)	South	4,100 (34.42%)
65–75	93,315 (15.92%)	West	1,798 (15.09%)
76–85	185,888 (31.71%)	Midwest	3,791 (31.83%)
>85	261,332 (44.57%)	Northeast	2,223 (18.66%)
Sex		Size * , No. of residents
Male	177,892 (30.34%)	<50	2,009 (17.13%)
Female	408,389 (69.66%)	50–100	5,254 (44.79%)
		101–250	4,246 (36.20%)
		>250	221 (1.88%)
Race/ethnicity		Total Residents on Medicaid*, %
White	471,854 (80.48%)	0%−24%	1,047 (9.00%)
Black	69,860 (11.92%)	25%−49%	1,673 (14.38%)
Hispanic	30,530 (5.21%)	50%−74%	5,835 (50.14%)
Other	14,037 (2.39%)	75%+	3,081 (26.48%)
Active diagnosis		Chain ownership *
Dementia	284,511 (48.53%)	Yes	7,094 (63.54%)
Diabetes	193,329 (32.98%)	No	4,069 (36.45%)
Heart failure	135,357 (23.09%)
Stroke	106,881 (18.23%)
Alzheimer’s	102,980 (17.56%)
Aphasia	26,186 (4.47%)
Bipolar disorder	19,046 (3.25%)
Hip fracture	12,170 (2.08%)
Multiple sclerosis	7,736 (1.32%)	Profit status *
		For profit	8,671 (73.92%)
		Non-profit	2,614 (22.28%)
		Government	445 (3.79%)

^*276 residents lacked complete information on NHs.

Figure 1 shows that nursing homes consistently underreport their level of antipsychotic prescribing compared to the prescribing captured in Medicare Part D claims, and this discrepancy widened after the CMS campaign. During the study period, the absolute level of antipsychotic prescribing decreased from 28.5% to 23.7% in the MDS and from 29.1% to 25.4% in Part D. This approximate 1% difference per quarter translates to approximately 6,000 nursing home residents. The statistical model shows these differences are statistically significant and change over time. Before the campaign, nursing homes underreported antipsychotic use, on average, by one percentage point relative to Medicare Part D levels (95% confidence interval (CI), 0.012–0.015; p<.001). Immediately after the CMS campaign, the gap in underreporting widened by another half a percentage point (0.004, 95% CI 0.003–0.005; p = .012). The extra underreporting of antipsychotic prescribing remained consistent throughout the rest of the post-campaign period, although the trend showed a gradual reduction toward baseline levels (−0.002, 95% CI −0.002 to −0.001); p = .011). (See Supplemental Figure S1, Supplemental Table S2, and Supplemental Table S3 in supplementary material.)

Figure 1.

Percentage of Long-stay Nursing Home Residents Receiving Antipsychotic Medications by Data Source 2011–2013. Figure 1 depicts the percentage of long-stay nursing home residents who are receiving antipsychotic medications as reported by the Medicare Part D Claims and MDS self-reports each quarter from 2011 through 2013.

Table 2 shows the characteristics of nursing homes and residents most likely to be associated with underreported antipsychotic prescribing. Before the CMS campaign, nursing homes in the US southern region were at highest risk for underreporting their antipsychotic prescribing (Adjusted Odds Ratio (AOR) 1.256, 95% CI: 1.159–1.362) relative to facilities in other parts of the US; this risk persisted throughout the entire period. Underreported antipsychotic prescribing was also more likely to occur in for-profit nursing homes both immediately after initiation of the CMS campaign (AOR 1.088, 95% CI: 1.005–1.178) and afterwards (AOR 1.094, 95% CI: 1.031–1.161). Nursing home residents with dementia, Alzheimer’s disease or bipolar disorders were at highest risk for underreported antipsychotic prescribing both before the CMS campaign (AOR 1.385, 95% CI: 1.330–1.444; AOR 1.234, 95% CI: 1.172–1.300; AOR 1.574, 95% CI: 1.444–1.716, respectively) and afterwards.

Table 2.

Facility and resident characteristics associated with underreported antipsychotic prescribing.

	Pre-CMS campaign			Launch of CMS campaign			Post CMS campaign
	AOR	95% CI lower limit	95% CI upper limit	AOR	95% CI lower limit	95% CI upper limit	AOR	95% CI lower limit	95% CI upper limit
Region
Midwest	1.031	0.950	1.119	1.089	0.987	1.202	1.123	1.044	1.209
South	1.256	1.159	1.362	1.262	1.145	1.391	1.292	1.201	1.390
Northeast	1.080	0.993	1.176	1.070	0.963	1.187	1.065	0.979	1.159
West	1.0 (ref)
Size, No. of residents
<50	1.118	0.984	1.271	1.119	0.943	1.327	0.997	0.883	1.127
50–100	1.029	0.923	1.147	1.005	0.868	1.164	0.910	0.817	1.014
101–250	1.030	0.926	1.145	1.019	0.883	1.176	0.917	0.825	1.019
>250	1.0 (ref)
% Residents on medicaid
0–24%	1.0 (ref)
25%–49%	0.890	0.790	1.003	0.895	0.776	1.033	0.949	0.855	1.054
50%–74%	0.955	0.861	1.059	0.957	0.847	1.081	0.960	0.874	1.054
75%+	1.008	0.903	1.125	0.997	0.875	1.135	1.027	0.930	1.134
Owned by a chain	0.969	0.916	1.025	0.953	0.891	1.019	1.006	0.956	1.059
For profit status
For Profit	1.052	0.984	1.124	1.088	1.005	1.178	1.094	1.031	1.161
Government	1.068	0.941	1.212	0.936	0.790	1.110	0.936	0.812	1.079
Non-profit	1.0 (ref)
Active diagnosis
Dementia	1.385	1.330	1.444	1.458	1.382	1.538	1.383	1.330	1.437
Diabetes	1.002	0.962	1.044	0.963	0.916	1.012	0.970	0.933	1.009
Heart failure	0.903	0.861	0.947	0.941	0.887	0.998	0.875	0.836	0.916
Stroke	0.898	0.853	0.945	0.965	0.908	1.025	0.879	0.837	0.924
Alzheimer’s	1.234	1.172	1.300	1.224	1.154	1.299	1.233	1.177	1.291
Aphasia	0.673	0.601	0.754	0.606	0.527	0.697	0.639	0.573	0.712
Bipolar	1.574	1.444	1.716	1.882	1.705	2.076	1.655	1.521	1.800
Hip fracture	1.116	1.006	1.239	1.100	0.979	1.236	0.960	0.879	1.049
Multiple sclerosis	0.673	0.552	0.820	0.515	0.401	0.662	0.525	0.430	0.642

^*AOR=Adjusted Odds Ratio. Adjusted for gender, race, age, Activities of Living Score (Morris, Fries, & Morris,1999), Cognitive Function Scale score (Thomas, Dosa, Wysocki, & Mor,2017), and missing data.

Discussion

Compared to Medicare Part D claims, nursing homes substantially underreported their antipsychotic prescribing despite federal regulations mandating full disclosure. The difference is not trivial—nursing homes failed to report the antipsychotic use of up to 6,000 residents each quarter, and yet these medications were paid for by the Medicare program. Furthermore, the underreporting is not random. The likelihood of the underreported antipsychotic prescribing increased for facilities located in the US southern states and operating as for-profit entities. Residents with dementia, Alzheimer’s disease or bipolar disorders were at the highest risk. Lastly, the underreporting showed a temporal response to the CMS campaign: underreporting grew at the same time there was increased pressure to reduce antipsychotic prescribing. From our understanding, this is the first study to examine the accuracy of self-reported antipsychotic prescribing in nursing homes.

Our findings contribute to a growing body of research challenging the extent that the CMS campaign improved the quality of dementia care. Our analyses confirmed that nursing homes prescribed less antipsychotics over the 3-year study period regardless of data source: MDS prevalence decreased from 28.5% to 23.7% and Part D prevalence decreased from 29.1% to 25.4%. As a result, 16,000 fewer nursing home residents received antipsychotics over our evaluation period. However, some of this decline is attributed to greater underreporting of the antipsychotic prescribing rather than actual reductions in use. Other research has also found the rate of antipsychotic prescribing in nursing homes was already declining well before the CMS campaign began suggesting a negligible impact (Maust, Kim, Chiang, & Kales, 2018). There is also evidence of subsequent increases in drug substitutions for the antipsychotics with medications that also pose risks to older adults (Maust et al., 2018).

In addition, our sensitivity analyses suggest unintended consequences. The prevalence of conditions associated with appropriately-prescribed indications (e.g., schizophrenia) increased from 7.3% (Q1 2011) to 8.2% (Q4 2013) suggesting documentation changes to accommodate the CMS policy (see technical appendix). Other research has detected similarly implausible increases in these diagnoses after initiation of the CMS campaign (Winter, Kerns, Winter, & Sabo, 2017). Statements by leading long-term care organizations including the American Geriatrics Society, American Health Care Association, American Medical Directors Association and the American Psychiatric Association have cautioned clinicians to avoid labelling older patients with new diagnoses to justify use of medications (Gerontological Advanced Practice Nurses Association, 2017). Furthermore, we found antipsychotic use in appropriately-prescribed residents decreased from 87% (Q1 2011) to 83% (Q4 2013) even though the CMS campaign excludes them. See Supplemental Table S4 in supplementary materials.

The study had several limitations. Our study was observational, although we minimized bias by applying a quasi-experimental research design. We did not evaluate all US nursing homes; however, 76% were still represented. Underreported antipsychotic prescribing levels may be different for patients not enrolled in Medicare Part D, although over 80% of nursing home residents have Medicare drug coverage (Briesacher, Soumerai, Field, Fouayzi, & Gurwitz, 2009). Underreported prescribing levels in the MDS may not be limited to only the antipsychotics. The MDS requires capture of several other drug classes including anticoagulants and antidepressants; however, it is our understanding that the accuracy of those reporting levels has never been assessed.

Conclusion

Nursing home facilities underreport their antipsychotic prescribing despite federal regulations mandating full disclosure. Future CMS regulations and research should consider the feasibility of using Medicare Part D claims rather than MDS data for more accurate information on antipsychotic prescribing.