Department of Economics
Response to Regulatory Stringency: The Case of
Antipsychotic Medication use in Nursing Homes
John R. Bowblis
Judith A. Lucas
Working Paper # - 2010-02
Response to regulatory stringency: The case of antipsychotic medication use
in nursing homes
John R. Bowblis†
Department of Economics
Scripps Gerontology Center
Institute for Health, Health Care Policy, and Aging Research
Center for Gerontology and Health Care Research
Providence VA Medical Center
Judith A. Lucas
Institute for Health, Health Care Policy, and Aging Research
Keywords: Regulation, Nursing Homes, Antipsychotics, Deficiencies
Acknowledgements: We would like to thank Robert Applebaum, Christopher Brunt, Jennifer Troyer, and
participants of the 2010 ASHEcon conference for helpful comments. We also like to thank Charlene
Harrington for information on minimum staffing requirements. Stephen Crystal and Judith A. Lucas
acknowledge funding for this work from the Agency for Healthcare Research and Quality (U18-
HS016097) and the Retirement Research Foundation (RRF# 2007-152).
† Corresponding Author: John R. Bowblis; Address: Department of Economics, Miami University, 3044
Farmer School of Business, Oxford, OH 45056; Phone 1 513 529 6180; Email: firstname.lastname@example.org
Abstract: This paper studies the impact of regulatory stringency, as measured by the statewide deficiency
citation rate over the past year, on the quality of care provided in a national sample of nursing homes from
2000 to 2005. The quality measure used is the proportion of residents who are using antipsychotic
medication. Although the changing case-mix of nursing home residents accounts for some of the increase
in the use of antipsychotics, we find that reliance on antipsychotics by nursing homes is responsive to
state regulatory enforcement. Nursing homes reduce their use of antipsychotics in response to the number
and type of deficiencies received by facilities in the state.
The quality of care provided by nursing homes has been a recurring concern for consumers, health
care professionals, and policy makers. States and the federal government have attempted to regulate
nursing home quality through multiple mechanisms (Walshe, 2001; Wiener, 2003). One regulatory
mechanism used to evaluate quality is the annual survey process conducted by states to determine if a
nursing home is compliant with federally-mandated standards of care. Those facilities that do not meet
these standards are given deficiency citations to indicate noncompliance (Spector and Drugovich, 1989).
A review of these state nursing home enforcement systems can be found in Harrington et al. (2004).
Multiple studies have looked at the relationship between the quality of care provided and the number
of deficiencies received by a nursing home. Studies that focus on nursing staff as a quality measure find
that facilities with lower staffing levels are more likely to receive a deficiency and receive more
deficiencies (See Harrington et al., 2000; Konetzka et al., 2004; Kim et al., 2009). Additionally, facilities
with poor quality in one dimension are also more likely to receive a deficiency in that dimension. For
example, Graber and Sloane (1995) find facilities with more physically restrained residents are more
likely to receive a physical restraint deficiency, while Castle and Engberg (2007) find similar results for
medication related deficiencies.
Although past research finds an association between the number of deficiency citations and quality,
many of these studies only measure the contemporaneous relationship between deficiencies and quality.
Contemporaneous deficiency citations and quality measures can be used to determine if deficiencies are
correlated with quality, but they do not measure the impact of these deficiencies on future quality.
Further, there is significant variation in the application and enforcement of nursing home standards across
states, as measured by the number and type of deficiencies issued (Harrington and Carrillo, 1999;
Harrington et al., 2006) and facilities may change care practices in an anticipatory fashion, responding to
the overall regulatory climate in the state. Prior studies have not addressed how this regulatory stringency
impacts the quality of care provided by nursing homes.
This paper studies the impact of regulatory stringency, as measured by the statewide deficiency
citation rate over the past year, on the quality of care provided by nursing homes from 2000 to 2005. The
quality measure studied is the proportion of residents who are using antipsychotic medication.
Antipsychotics are studied because the widespread reliance on these medications has been a long-standing
issue in nursing home quality. Further, there is significant variation in the use of antipsychotics across
states and their use grew rapidly, from 16.0% of nursing home residents in 1996 to over 27.6% by 2001
(Office of Inspector General, 2001; Briesacher et al., 2005). Although multiple factors, including changes
in resident case-mix and perceived safer side effect profiles of second generation antipsychotics are some
of the reasons for the increased use of antipsychotics, states exercise considerable discretion in the
number and type of deficiency citations they impose on nursing homes. We study how these deficiencies
impact the use of antipsychotic medications by nursing homes.
2. Antipsychotic Use in Nursing Homes
In an attempt to improve quality in nursing homes, Congress passed nursing home reform legislation
as part of the federal Omnibus Budget Reconciliation Act (OBRA) of 1987. Part of this legislation
focused on the overuse of psychoactive medications as a form of “chemical restraint” and mandated the
establishment of guidelines to be used by state surveyors in sanctioning nursing homes for unnecessary
drug use. These guidelines define unnecessary drug use as excessive dose, excessive duration, without
adequate monitoring or indication, continued in the presence of adverse consequences, and without
specific target symptoms (Office of Inspector General, 2001). The proportion of nursing home residents
receiving antipsychotic medications declined after the passage of the legislation and as the antipsychotic
guidelines were developed through the early 1990s (Shorr et al., 1994; McKenzie et al., 1999).
Shortly after the passage of OBRA, the Food and Drug Administration (FDA) approved a new
generation of antipsychotics called atypical antipsychotics. Although the first of these drugs, clozapine,
was found to have serious side-effects, the introduction of risperidone in 1994, olanzapine in 1996, and
quetiapine in 1997 accelerated a switch from conventional antipsychotic medications to atypical
antipsychotics reflecting perceptions of greater safety of the atypicals. While schizophrenia and bipolar
disorder were the principal conditions for which FDA indications were approved for use of these drugs,
the perceived safety profile of these new atypical antipsychotics lead to their wide use for “off-label”
purposes (Crystal et al., 2009). In particular, nursing homes often used atypical antipsychotics to manage
behavioral symptoms associated with dementia (Briesacher et al, 2005; Kim and Whall, 2006). For
residents without schizophrenia or bipolar disorder, unless the resident has symptoms of psychosis or
certain persistent and severe behavioral symptoms of dementia (e.g., aggressive behavior), antipsychotic
use is inconsistent with federal interpretive guidelines promulgated to guide nursing facilities’ practices
and surveyors’ assessments (Centers for Medicare and Medicaid Services (CMS), 2004).
With the rapid increase in the use of atypical antipsychotics, evidence accumulated that risks for this
class of medications were greater than initially perceived. Atypical antipsychotics were found to be
associated with the adverse side effects of weight gain, hyperlipidemia, and diabetes (Gianfrancesco et al.,
2002; Koro et al., 2002: Lund et al., 2001). For elderly patients with dementia, who make up a large
percentage of nursing home residents, the evidence of risks associated with taking antipsychotics is
mounting (Crystal et al., 2009; Trufuro et al., 2009). Using a meta-analysis of randomized clinical trials,
Schneider et al. (2005) found the absolute mortality risk for nursing home residents with dementia is
about two percent higher for nursing home residents treated eight to twelve weeks with an antipsychotic
compared to a placebo. This led the FDA to issue a public health advisory on April 11, 2005 that
cautioned that atypical antipsychotics were associated with increased risk of death for persons with
dementia. In 2008, this warning was extended to all antipsychotics.
3. Deficiency Citations and Their Impact on Antipsychotic Use
Nursing homes are complex producers of long term care which operate in a highly regulated industry.
One regulatory tool that policy-makers can use to impact nursing home quality is to assign regulatory
sanctions called deficiency citations. Deficiencies are issued by state surveyors as part of the required
annual Medicare and Medicaid re-certification process that evaluates whether the nursing home is
meeting minimum regulatory standards. For research purposes, these deficiencies can be organized to
reflect quality of care, quality of life, and administrative process standards (Harrington et al., 2000).
The standards of care for which deficiencies are given are set at the federal level but how states
interpret, implement, and enforce these regulations can be different for each state. This has led to
significant variation in the number and type of deficiency citations given by each state (Harrington and
Carrillo, 1999; Harrington et al., 2006). The variation in deficiency citations across states provides
identification of regulatory stringency, with a higher number of deficiency citations implying a more
stringent regulatory environment within the state. Since the use of antipsychotics can be viewed as an
input in the production of nursing home care and as a quality measure (Mor et al., 2004), the literature on
how deficiency citations impact quality provides a conceptual framework to make predictions on how
regulatory stringency impacts the use of antipsychotic medication.
The first of these regulatory stringency measures is the total number of deficiencies issued by state
surveyors. Since the re-certification review provides an external evaluation of quality for all facilities, the
total number of deficiency citations is often taken as a measure of overall quality. Facilities that receive a
high number of deficiencies are aware they have a number of quality issues and may attempt to evaluate
and address multiple areas of quality concerns. States that assign a high average number of deficiencies
per survey, consistent with high regulatory stringency, could cause facilities to self-evaluate all aspects of
quality, including antipsychotic prescribing practices, and may reduce the use of antipsychotics before
regulatory reviews. However, a high number of deficiencies might cause the facility to thinly spread
resources to improve multiple dimensions of quality concern instead focusing on a few key areas. This
could lead to facilities increasing their use of antipsychotics.
Besides the total number of deficiencies, it is possible to look at specific deficiencies. States that
assign more facilities a specific deficiency can be attempting to improve a targeted quality area. For
example, states that have a high proportion of surveys with a deficiency for physical restraints may be
focusing on reducing the use of physical restraints. However, the impact of specific deficiencies on
antipsychotic use rates could be positive or negative depending on the specific citation. That is, more
deficiency citations related to antipsychotic use should reduce the use of antipsychotics, but more
deficiency citations in which antipsychotic use is a potential substitute (i.e., restraint use) may increase
the use of antipsychotics.
This study uses six specific deficiency citations that can impact the use of antipsychotics. These
deficiencies include F221 (free from physical restraint), F222 (free from chemical restraint), F319 (receipt
of mental health services for mental or psychosocial adjustment difficulty), F329 (unnecessary drug use),
F330 (free from antipsychotic use without approved conditions), and F331 (efforts to reduce dosage and
Federal guidelines state that “physical or chemical restraints” are inappropriate when used for the
purpose of “discipline or convenience, and not required to treat the resident’s medical symptoms.” If a
facility is found to be violating this regulation by using physical restraints, deficiency F221 is given; if the
violation is for “chemical restraints” then deficiency F222 is given. In both cases, if regulators are
assigning a high number of F221 or F222 deficiencies, a facility could believe that regulators are targeting
the use of restraints. Facilities in states with a higher number of F221 deficiencies may reduce or increase
the use of antipsychotics depending on whether the facility views antipsychotics as a substitute for
physical restraints or believes regulators will focus on both physical and chemical restraints. A higher
number of F222 deficiencies should reduce the use of antipsychotic medications.
The deficiency F319 is given if a resident with psychosocial or mental adjustment difficulty does not
receive mental health treatment or services for his or her conditions. Residents who have problems in
adjustment are required to receive a psychiatric evaluation and appropriate medical interventions, such as
individual, family, or group psychotherapy, drug therapy, or other rehabilitative therapies. Deficiency
citations in this area may reflect antipsychotics being used to treat residents with adjustment difficulty
without proper evaluation. More statewide F319 deficiencies are expected to be associated with lower
The final set of regulatory deficiencies (F329, F330, and F331) pertains to unnecessary medication
use. In particular, CMS guidelines related to deficiency F329 require that each resident’s drug regimen be
free from all unnecessary drug use while deficiencies F330 and F331 specifically address antipsychotics
(CMS, 2004). Deficiency F330 requires that the resident be free from antipsychotic use without approved
conditions, while F331 requires there be a tapering of dosage when antipsychotics are used. These
deficiencies are expected to be associated with reduced use of antipsychotics, although which deficiency
has the largest effect is an empirical question.
4. Econometric Methods
4.1 Data Sources
The data source used in this analysis is the Online Survey Certification and Reporting (OSCAR)
System. OSCAR is a uniform database of state nursing home regulatory reviews and contains
information on facility characteristics, including nursing home structure, case-mix, and deficiency
citations. All Medicare/Medicaid certified facilities are required to report facility, census, and staffing
information as part of their yearly re-certification review process. Data are validated during on-site
surveys completed by state surveyors, operating under CMS oversight, every nine to fifteen months with
an average period of twelve months between surveys. Survey data are entered into the OSCAR system.
OSCAR is the only national source for information on deficiencies for the period studied. Although
OSCAR data have been criticized, and some studies have preferred to use cost reports, many studies have
found that OSCAR measures are appropriate for research (Intrator et al., 2005; Harrington et al., 2006;
Feng et al., 2005).
The OSCAR system allows for construction of a panel dataset with the nursing home facility as the
unit of observation. In order to construct the sample used for analysis, all standard OSCAR surveys at
least 180 days apart for U.S., non-hospital based nursing homes in the forty-eight contiguous states that
occurred between January 1, 1999 and December 31, 2005 are obtained (N=94,680). Since regulatory
variables are measured using data from all the surveys in the prior year, data from 1999 are only used to
construct analytical variables. Additionally, the regression method accounts for serial correlation. This
method causes the first observation of each facility to be dropped from the sample and requires each
included facility to have at least three surveys in the study period. The resulting sample contains 14,743
unique nursing facilities with the average time between surveys being slightly over 365 days (N= 64,711).
These restrictions are unlikely to lead to any significant selection bias because most of the observations
that are dropped are from 1999 or reflect the first observation of a facility in the dataset after 1999.
Further, these data are supplemented with information from two additional sources to obtain the state
Medicaid reimbursement rates and the state minimum nursing direct care staffing requirements. State
Medicaid reimbursement rates are obtained from Grabowski et al. (2004a; 2004b; 2008) and are adjusted
for inflation to 2005 dollars using the CPI-U index. The minimum required state nursing direct care
hours per resident day (HPRD) are constructed from multiple sources. First, information on nurse
staffing requirements are obtained from Harrington (2001; 2008). Since these sources only provide a
cross-sectional perspective of minimum staffing rates, state statutes and regulations on state websites are
reviewed with follow-up phone calls to state agencies/associations to identify and confirm required
minimum staffing rates for each specific year from 1999 to 2005.
The dependent variable used in the regression is the proportion of residents in the facility who are
receiving an antipsychotic medication. This proportion is calculated as the total number of residents that
are receiving an antipsychotic medication at the time of the survey divided by the total number of
residents in the facility. The total number of residents receiving antipsychotics is determined from the
Minimum Dataset (MDS), which uses clinical records of each patient, and this measure is not risk-
adjusted. The remaining discussion in this section describes the construction of regulatory deficiency and
control variables included in the model.
Regulatory Deficiency Variables
Deficiency citations are a measure of how facilities are achieving minimum quality standards. The
guidelines used to determine if a facility should receive a deficiency are set by the federal government,
but the actual on-site surveys of compliance by nursing facilities are performed by state surveyors within
particular survey regions. Because of this, there is significant variation in the number and types of
deficiencies issued to facilities (see Table 1). This variation in number and type of deficiencies found in
the OSCAR system reflects variation in state and local survey region regulatory stringency; for example,
a state that issues a high number of deficiencies for physical restraints may have prioritized this area as a
focus for enforcement efforts.
Deficiencies can be measured either in terms of the total number of deficiencies or in terms of
specific types of deficiencies. To capture regulatory enforcement effort, both total deficiencies and
specific deficiencies that could impact antipsychotic use are included in the model. The specific
deficiencies used include F221 (free from physical restraint), F222 (free from chemical restraint), F319
(receipt of mental health services for mental or psychosocial adjustment difficulty), F329 (unnecessary
drug use), F330 (free from antipsychotic use without approved conditions), and F331 (efforts to reduce
dosage and discontinue antipsychotics). F329 citations reflect a broader measure of regulatory activity
related to medication use and are not limited to antipsychotic use.
In the regression model, state-level deficiency variables (e.g., total and specific) are used to capture
how a facility responds to overall regulatory stringency as reflected in the statewide rate of deficiency
citations. If a facility observes that other facilities in the state are receiving certain deficiencies, the
facility may be induced to focus on that aspect of care. Further, the use of state-level variables captures
variation in enforcement efforts by states. For total number of deficiencies, the state-level deficiency
variable is the statewide average number of deficiency citations for all regulatory reviews in the prior
twelve month period. For the state-level specific deficiency variables, the regulatory variable is the
proportion of all statewide regulatory reviews with the specific deficiency in the prior twelve month
In order to completely model the impact of regulation on nursing homes’ antipsychotic use, it is
important to account for other characteristics that may affect the use of antipsychotics. Control variables
include a set of time dummies, facility-specific heterogeneity, and time-varying controls. The time
dummies are indicator variables for each calendar year. The facility-specific heterogeneity accounts for
facility-specific variables that are constant over time and may impact antipsychotic use. Both observable
(e.g. state indicators, facility bed size) and unobservable (e.g., floor plan, unobservable care practices)
facility-specific heterogeneity are controlled in the regression model by using a fixed effects estimation
technique that is discussed further in the next section. Finally, time-varying control variables are included
in the model to capture changes in facilities over time that could impact the use of antipsychotics. These
time-varying control variables are broken into five categories.
The first time-varying control variable is the use of physical restraints (proportion of residents who
are physically restrained in the prior regulatory review). Physical restraints are any physical or
mechanical device that restricts the freedom of movement, and are an input that nursing homes may use to
avoid harm to the resident or other persons. Although there may be justification for restraining selected
residents for short periods in a limited set of circumstances, reducing the use of physical restraints has
been a priority since the passage of OBRA. Therefore, some nursing homes may have reduced the use of
physical restraints only to substitute increased use of antipsychotic medications. Two different
regressions are estimated, one focusing on restraint use among all residents and the other on facility-
acquired use among residents who did not have orders for restraints prior to admission.
The second set of time-varying control variables are facility operational characteristics. These
operational characteristics of the nursing home include payer-mix and occupancy rate. The
reimbursement facilities receive for providing care varies significantly by source. For example, Medicaid
has consistently paid low reimbursement rates and facilities that are more dependent on Medicaid have
been found to provide lower quality of care (Cohen and Spector, 1996; Grabowski, 2004). Payer mix
categories include the proportion of residents funded by Medicaid and the proportion of residents paid for
by Medicare, with the reference category the proportion of residents paid for by all other sources (e.g.,
self-pay, private insurance). Finally, occupancy rate is included as a control variable as it has important
influence on treatment patterns; for example, a facility with a low proportion of occupied beds may not
have enough revenue to cover the fixed costs of production.
The third set of time-varying control variables captures resident case-mix. Antipsychotics are
indicated for residents with schizophrenia and bipolar disorder, but are also used ‘off-label’ to manage
behavioral symptoms of dementia. It is important to account for the increase in the number of residents
that may have these and other relevant medical conditions, but a limitation of OSCAR is it only provides
broad measures of mental health case-mix: the proportion of residents in the facility with dementia,
depression, developmental disability, and psychiatric illness other than dementia or depression. The level
of dependence and use of special medical procedures of the residents in the facility, or facility acuity
level, is measured using the ACUINDEX (Cowles, 2002).
Nurse staffing is the largest input cost of nursing homes and is included as the final category of
control variables. Nurse staffing variables are measured in terms of the level and composition of
staffing by each type of nurse and an indicator variable reflecting whether the facility had any staff
specializing in mental health services. Nurse staffing categories include registered nurses (RNs),
licensed practical nurses (LPNs), and certified nurse aides (CNAs). Each type of nursing staff is
included in the regression and measured in terms of staffing hours per resident day (HPRD) to
standardize across facilities of various sizes. In order to identify and correct for occasional improbable
values in the HPRD that may be recording errors, we identified observations for each nurse staff type that
were unreliable based on the following criteria: (A) more than twenty-four hours of staffing; (B) zero
staffing; and (C) among facilities that do not fall into first two categories, those that are outside three
standard deviations of the mean. Unlike other studies that have dropped these observations (Banaszak-
Holl et al., 2002; Harrington et al., 2006; Kash et al., 2007), no observations are dropped, but instead
indicator variables are created to identify which observations have unreliable staffing records.
Since staffing levels are an input that nursing homes can change in response increase to regulatory
factors, staffing changes and the antipsychotic use rate could be jointly determined due to substitution of
inputs (Cawley et al., 2006). This could lead to endogeneity bias. To assess the size of this potential bias,
we estimate the model with contemporaneous, lagged, and without staffing variables. The coefficient
estimates of the state-level regulatory variables are the same, but any causal inference of the staffing
variables should be interpreted with caution because of the potential endogeneity of staffing levels.
The final set of time varying control variables is the average state Medicaid reimbursement rate
adjusted for inflation using the CPI-U and the state minimum nursing staff requirement. The impact of
Medicaid reimbursement on quality is mixed and depends on the level of excess demand (Nyman, 1985;
Gertler, 1989; 1992; Grabowski, 2001). The average state Medicaid reimbursement rate is the average
per diem reimbursement for Medicaid nursing home residents in the state. Since this variable is only
available by calendar year, the reimbursement rate for the prior calendar year is used in the regression.
The minimum nurse staffing rate is defined as the minimum number of direct care nursing hours per
resident day (HPRD) required by state regulation. To keep this variable consistent across all states and
years, data for all states that reported requirements in terms of nurses-per-resident or nurses-per-bed were
converted to hours per resident day. In addition, some states have different staffing requirements based
on the number of residents; therefore, we standardized the minimum direct care staffing ratio used in the
analysis based on 100 residents (Harrington, 2001; 2008). Since the effective date of the state minimum
nurse staffing level is known for all states, the minimum staffing level variable is based on the minimum
staffing requirement in effect 365 days before the current regulatory review. Robustness checks used
contemporaneous Medicaid reimbursement rates and state minimum staffing rates, and a squared term for
Medicaid reimbursement rates. The parameter estimates for the regulatory variables were similar for all
4.3 General Model Specification
The empirical model uses a reduced form relationship between antipsychotic use and regulatory
deficiency variables to determine how state regulatory stringency impacts the use of antipsychotics in
nursing homes. To simplify notation, assume that each nursing facility i in state s is observed only once
14 Download full-text
per year t. The empirical model to be estimated is the proportion of residents using antipsychotics (??????
regressed on regulatory deficiency variables that impact all facilities in the state ?????
, the facility use of
physical restraints ?????????, other time-varying facility control variables and time dummies (????), and
facility-specific heterogeneity (???. The facility-specific heterogeneity is treated as a fixed effect and
captures both observed and unobserved differences across facilities that are constant over time. Since
data from 1999 are used to construct the state-level deficiency variable, the following reduced form model
is estimated for years 2000 to 2005:
? ?????????? ??????? ??? ????
where ????? ???????? ????. By assuming that ???? is independent and identically distributed, the
preceding equation can be estimated by the technique described by Baltagi and Wu (1999). Hausman
tests find that a fixed effect and serial correlation are consistent with the data.
The percentage of nursing home residents receiving antipsychotic medication by state is reported in
Table 1 for the years 2000 and 2005. Across the forty-eight states in the sample, the average increase in
the proportion of nursing home residents using antipsychotic medications is 6.14 percentage points over
the 5-year period, from 20.71% of residents in 2000 to 26.86% of residents in 2005. We found wide
variation across states in the change rates for use of antipsychotics, although increases were noted in
every state. Michigan had the smallest increase in the proportion of residents using antipsychotics with a
1.71 percentage point increase while Alabama had the largest increase with 11.51 percentage points. In
order to determine whether these rates of increase are correlated with selected regulatory variables, the
remaining columns of Table 1 report the proportion of facilities that received specific deficiencies and the
average number of deficiencies per regulatory review in 1999.
<Insert Table 1>