Association of the US Food and Drug Administration Antipsychotic Drug Boxed Warning With Medication Use and Health Outcomes in Elderly Patients With Dementia.

Importance: Atypical antipsychotics (AAPs) are often used off-label to manage dementia-associated neuropsychiatric symptoms. In 2005, the US Food and Drug Administration (FDA) issued a boxed warning for the use of AAPs in elderly patients. The long-term association of this warning with health outcomes is unknown to date. Objective: To assess the long-term association of the 2005 FDA boxed warning on AAPs with psychiatric medication and opioid use, health events, and quality of life among elderly individuals with dementia. Design, Setting, and Participants: For this cross-sectional study, data were analyzed from the household component of the Medical Expenditure Panel Survey (MEPS), the National Ambulatory Medical Care Survey (NAMCS), and the National Hospital Ambulatory Medical Care Survey (NHAMCS) fielded between January 1, 1996, and December 31, 2014. This interrupted time-series analysis applied to 3-year moving means derived from the 1996-2014 MEPS, NAMCS, and NHAMCS. All survey respondents included in this analysis were 65 years or older and had dementia. Data analysis was performed from December 1, 2017, to March 15, 2018. Exposures: The 2005 FDA boxed warning on AAPs. Main Outcomes and Measures: Use of psychiatric medications and opioids, prevalence of cerebrovascular and cardiovascular events, prevalence of falls and/or fractures, 2-year mortality, and health-related quality of life assessed by the Medical Outcomes Study 12-Item Short-Form Health Survey scores.
Results: A total of 2430 (MEPS) and 5490 (NAMCS and NHAMCS) respondents were identified, corresponding to weighted populations of 22 996 526 (MEPS) and 65 502 344 (NAMCS and NHAMCS) noninstitutionalized elderly individuals with dementia (mean [SD] age, 81.06 [1.13] years; 63.1% female). In the MEPS sample, compared with before 2005, AAP use (from an annual slope of 0.99 to -0.18 percentage points), cerebrovascular events (0.75 to -0.50 percentage points), and falls and/or fractures (-1.72 to -0.40 percentage points) decreased and opioid use (0.04 to 1.29 percentage points), antiepileptic use (-0.42 to 1.21 percentage points), cardiovascular events (-0.13 to 1.30 percentage points), and 2-year mortality risk (-0.68 to 0.18 percentage points) increased. Health-related quality of life remained relatively unchanged. The NAMCS and NHAMCS sample yielded similar findings. Conclusions and Relevance: These data suggest that the 2005 FDA boxed warning was associated with some unintended negative patient outcomes.

Introduction

Dementia in general and specifically Alzheimer disease (AD), which accounts for 60% to 80% of dementia cases,[1] is characterized by a rapid cognitive decline, deteriorating functional status, and worsening neuropsychiatric symptoms (NPSs).[2] Up to 90% of patients with dementia experience NPSs, a group of distressing and disruptive behavioral symptoms that include mood disorders, sleep disorders, psychotic symptoms, agitation, and excessive verbal or physical motor activity.[3,4,5] Symptoms of agitation and aggression affect 40% to 60% of patients with AD.[6] Neuropsychiatric symptoms are highly prevalent in all stages of AD[4] and contribute to poor patient and caregiver health-related quality of life (HRQoL),[4,7,8] caregiver burden and burnout,[3,9] institutionalization,[3,6,10,11,12] and increased medical costs.[12,13]

Atypical antipsychotics (AAPs) are often used off-label to manage dementia-associated NPSs.[14] The 2016 American Psychiatric Association practice guideline[15] on the treatment of agitation or psychosis in patients with dementia recommends a patient-centered comprehensive treatment plan incorporating nonpharmacologic and pharmacologic approaches. Antipsychotics are recommended as part of this approach when symptoms are serious, are dangerous, and/or cause significant patient distress.[15] Currently, no medication has been approved by the US Food and Drug Administration (FDA) for the treatment of dementia-associated NPSs, including agitation.

Concerns regarding increased risk of cerebrovascular events in elderly patients with dementia associated with certain AAPs began in the 2000s,[16,17,18,19] leading to FDA warnings for risperidone in 2003, olanzapine in 2004, and aripiprazole in 2005.[16] A subsequent meta-analysis[16] of clinical trials suggested that use of AAPs in patients with dementia was associated with increased mortality (odds ratio, 1.54; risk difference, 1%). In response, the FDA issued a boxed warning in 2005 regarding increased mortality associated with the use of AAPs in elderly patients with dementia-related psychosis.[20] In 2008, the FDA extended the boxed warning to all antipsychotics[21] based on reports[22,23] of similar or higher mortality risk in elderly individuals with dementia taking conventional antipsychotics.

Several studies[24,25,26,27,28] reported a decreased prevalence of antipsychotic medication use by elderly patients with dementia after the 2005 warning. However, the warning’s long-term association with health outcomes in elderly patients remains unknown. To address this gap, we used data from 3 nationally representative health care surveys of the noninstitutionalized US population to compare (1) use of AAPs, other psychiatric medications, and opioids; (2) health events, including cerebrovascular events, cardiovascular events, falls and/or fractures, and mortality; and (3) HRQoL between the prewarning (1996-2004) and postwarning (2005-2014) periods.

Methods

Data Sources and Study Samples

For this cross-sectional study, we analyzed data from 3 nationally representative health care surveys, the household component of the Medical Expenditure Panel Survey (MEPS), the National Ambulatory Medical Care Survey (NAMCS), and the National Hospital Ambulatory Medical Care Survey (NHAMCS), fielded between January 1, 1996, and December 31, 2014. We used the MEPS as our primary data source; pooled data from the NAMCS and NHAMCS were used to confirm findings for measures common across all surveys. At the time of data analysis, the most recent data were from 2014 because all surveys typically release data after a 2- to 3-year lag.[29,30] The data used in this study did not involve the interaction or interview with any individuals, and the data do not include any individually identifiable data (eg, does not include names, addresses, Social Security or medical record numbers, or other obvious identifiers), and as such is not research involving humans. Furthermore, this study used existing, fully deidentified data, and the investigator(s) cannot be identified, directly or through identifiers linked to individuals, and as such institutional review board approval is not required. This study followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.

The health component of the MEPS is designed to be nationally representative of the civilian noninstitutionalized US population.[31] Data on sociodemographic characteristics, health characteristics, insurance coverage, use of health care services, and health care costs are collected from survey respondents during 5 interview rounds during 30-month periods and from their medical practitioners. The health component of the MEPS follows an overlapping panel survey design, with a new panel of households sampled yearly.

The NAMCS and NHAMCS are designed to be representative of visits to office-based physicians and to emergency departments, outpatient departments, and ambulatory surgery locations of noninstitutional general and short-stay hospitals.[30] Both the NAMCS and NHAMCS exclude federal facilities, share a common design, and collect data (abstracted from medical records) on similar measures (eg, patient demographic characteristics, reason for visit, diagnosis, and services and treatments provided). Each physician participating in the NAMCS is randomly assigned to a 1-week reporting period and each outpatient facility participating in the NHAMCS to a 4-week reporting period during which data from a systematic random sample of visits are recorded. Data from the 2 surveys can be combined to analyze ambulatory visits to US nonfederal facilities.[30]

We defined 2 study samples by data source. The first sample included all MEPS respondents 65 years or older who reported, either themselves or through a proxy, a diagnosis of dementia or AD. The second sample included all physician (NAMCS) and ambulatory care (NHAMCS) visits by patients 65 years or older who had a practitioner-assigned diagnosis of dementia or AD. Dementia and AD were identified by International Classification of Diseases, Ninth Revision, Clinical Modification diagnosis codes (eTable 1 in the Supplement).

Study Measures

We obtained information on demographic characteristics (eTable 2 and eTable 3 in the Supplement); use of typical and atypical antipsychotics, hypnotics, antidepressants, antiepileptics, antidementia agents, and opioids (eTable 4 in the Supplement); cerebrovascular events (any stroke or transient ischemic attack); cardiovascular events (myocardial infarction, ischemic heart disease, or coronary ischemia); and falls and/or fractures (eTable 1 in the Supplement) from all 3 data sources. Data on 2-year mortality and HRQoL, as measured by the mental component summary (MCS) and physical component summary (PCS) scores of the Medical Outcomes Study 12-Item Short-Form Health Survey (SF-12), were obtained from the MEPS.[32]

Statistical Analysis

We used population-level interrupted time-series analysis (ITSA) methods to assess the association of the 2005 FDA warning with annual measures of medication use; prevalence of cerebrovascular, cardiovascular, and fall and/or fracture events; 2-year mortality; and HRQoL. We estimated annual weighted proportions for the drug use, health events, and mortality measures and weighted means for the HRQoL measures (eTable 5 and eTable 6 in the Supplement). The person- or visit-level weights provided in each data source were used to compute national estimates of these outcomes, accounting for the multistage stratified sampling design of each survey.[33,34] To reduce year-to-year variability and allow more prominent trends to be identified, we computed 3-year moving means of the weighted summary data for each outcome for 1997 through 2013 (eg, the moving mean for an outcome in 1997 is the mean of the weighted summary values from 1996 to 1998) (eTable 7 and eTable 8 in the Supplement). We then fitted a series of multiple regression equations in which the dependent variables were based on the 3-year moving means. The associations of the 2005 warning, in terms of the differences between the slopes before and after 2005 and the immediate changes (ie, level change) because of the warning, were determined from the regression coefficients (eAppendix in the Supplement). We chose 2005 as the interruption year a priori because that was the year of the FDA warning, and previous research on the association of the warning with AAP use also compared rates before and after 2005.[25,27,28,35] A negative slope indicates a decrease in prevalence within the time frame, and a positive slope indicates an increase in prevalence. Analyses were performed using SAS software, version 9.4 (SAS Institute Inc) from December 1, 2017, to March 15, 2018.

Results

We identified 2430 (of 620 246) observations for elderly MEPS respondents with dementia or AD between 1996 and 2014 that met our sample selection criteria, representing 22 996 526 US noninstitutionalized individuals. We also identified 5490 (of 1 715 005) visits by elderly patients with dementia or AD during the same period in the NAMCS and NHAMCS data, representing 65 502 344 US noninstitutionalized individuals. Among all individuals, the mean (SD) age was 81.06 (1.13) years, and 63.1% were female. The MEPS sample was predominantly white non-Hispanic (78%-94%) and living in urban areas (67%-83%) (eTable 2 in the Supplement). The demographic distribution of the NAMCS and NHAMCS sample was similar (eTable 3 in the Supplement). Unless otherwise noted, results reported herein are derived from the MEPS data. The ITSA results are given in the Table.

Table.

Estimates of Medication Use and Health Outcomes Among Individuals With Dementia

Outcome	MEPS				NAMCS and NHAMCS
	Before warning		After warning		Before warning		After warning
	Slope, % (95% CI)	Level change in 2005, % (95% CI)	Slope, % (95% CI)	Change in slopes, % (95% CI)	Slope, % (95% CI)	Level change in 2005, % (95% CI)	Slope, % (95% CI)	Change in slope, % (95% CI)
Antipsychotic use
Any	0.87 (0.27 to 1.48)	0.97 (−2.92 to 4.86)	−0.16 (−0.66 to 0.34)	−1.04 (−1.90 to −0.17)	−0.03 (−0.86 to 0.79)	2.05 (−1.40 to 5.49)	−0.82 (−1.22 to −0.41)	−0.78 (−1.71 to 0.15)
Atypical	0.99 (0.31 to 1.67)	−0.15 (−4.32 to 4.02)	−0.18 (−0.68 to 0.32)	−1.17 (−2.09 to −0.26)	1.57 (1.12 to 2.01)	−0.54 (−3.18 to 2.10)	−0.83 (−1.27 to −0.39)	−2.40 (−3.01 to −1.78)
Typical	−0.12 (−0.21 to −0.03)	1.46 (0.71 to 2.20)	0.00 (−0.14 to 0.14)	0.12 (−0.03 to 0.28)	−1.58 (−2.03 to −1.13)	2.66 (0.68 to 4.65)	0.00 (−0.08 to 0.07)	1.58 (1.13 to 2.02)
Psychiatric medication or opioid use
Any	3.92 (1.91 to 5.93)	0.48 (−8.45 to 9.40)	0.67 (0.34 to 1.01)	−3.24 (−5.27 to −1.21)	1.88 (1.28 to 2.47)	3.46 (−2.53 to 9.44)	0.14 (−1.01 to 1.30)	−1.73 (−3.03 to −0.44)
Hypnotic	−0.22 (−1.16 to 0.73)	−2.51 (−6.68 to 1.67)	−0.47 (−0.53 to −0.40)	−0.25 (−1.20 to 0.70)	0.02 (−0.37 to 0.41)	0.73 (−2.03 to 3.48)	0.16 (−0.49 to 0.82)	0.14 (−0.63 to 0.92)
Antidepressant	1.87 (0.84 to 2.89)	−6.37 (−12.89 to 0.17)	0.99 (0.52 to 1.46)	−0.88 (−1.91 to 0.15)	0.57 (−0.33 to 1.47)	−0.02 (−5.10 to 5.05)	0.49 (0.17 to 0.81)	−0.08 (−1.06 to 0.91)
Antiepileptic	−0.42 (−1.08 to 0.24)	2.23 (−0.98 to 5.44)	1.21 (0.66 to 1.76)	1.63 (0.76 to 2.49)	−0.03 (−0.17 to 0.11)	2.42 (−0.90 to 5.74)	0.60 (−0.16 to 1.37)	0.63 (−0.17 to 1.43)
Antidementia	14.03 (13.28 to 14.79)	−9.76 (−12.80 to −6.72)	0.63 (0.17 to 1.08)	−13.41 (−14.30 to −12.51)	3.28 (2.64 to 3.93)	2.44 (−2.20 to 7.08)	0.09 (−0.52 to 0.70)	−3.19 (−4.12 to −2.26)
Opioid	0.04 (−0.98 to 1.06)	−1.14 (−4.73 to 2.45)	1.29 (1.01 to 1.58)	1.25 (0.19 to 2.31)	0.28 (−0.03 to 0.58)	−0.29 (−2.74 to 2.16)	0.63 (0.13 to 1.12)	0.35 (−0.24 to 0.94)
Cardiovascular events	−0.13 (−0.69 to 0.42)	1.42 (−3.48 to 6.32)	1.30 (0.30 to 2.30)	1.44 (0.26 to 2.61)	−0.04 (−0.31 to 0.22)	−0.32 (−3.51 to 2.87)	0.87 (0.37 to 1.37)	0.91 (0.39 to 1.44)
Cerebrovascular events	0.75 (0.46 to 1.04)	−0.74 (−2.34 to 0.86)	−0.50 (−0.71 to −0.28)	−1.25 (−1.63 to −0.86)	−0.23 (−0.52 to 0.07)	1.06 (−0.28 to 2.39)	0.16 (−0.10 to 0.42)	0.39 (0.00 to 0.77)
Falls and/or fractures	−1.72 (−1.93 to −1.51)	1.88 (0.87 to 2.88)	−0.40 (−0.56 to −0.24)	1.32 (1.06 to 1.59)	−0.23 (−0.43 to −0.03)	−0.21 (−1.12 to 0.70)	0.29 (0.16 to 0.42)	0.52 (0.27 to 0.76)
Mortality
2-y	−0.68 (−1.33 to −0.03)	2.04 (−0.75 to 4.83)	0.18 (−0.07 to 0.42)	0.85 (0.13 to 1.57)	NA	NA	NA	NA
2-y excluding 1997	−0.26 (−0.45 to −0.06)	0.78 (−0.73 to 2.29)	0.18 (−0.07 to 0.42)	0.43 (0.09 to 0.77)	NA	NA	NA	NA
SF-12
MCS	0.41 (0.21 to 0.60)	−1.63 (−2.46 to −0.80)	0.24 (0.12 to 0.36)	−0.16 (−0.38 to 0.06)	NA	NA	NA	NA
PCS	0.55 (0.50 to 0.60)	0.16 (−0.53 to 0.84)	−0.29 (−0.49 to −0.09)	−0.84 (−1.04 to −0.64)	NA	NA	NA	NA

Abbreviations: MCS, mental component summary; MEPS, Medical Expenditure Panel Survey; NA, not applicable; NAMCS, National Ambulatory Medical Care Survey; NHAMCS, National Hospital Ambulatory Medical Care Survey; PCS, physical component summary; SF-12, Medical Outcomes Study 12-Item Short-Form Health Survey.

Medication Use

The prevalence of AAP use decreased −0.15 percentage points (95% CI, −4.32 to 4.02 percentage points) in 2005 and continued to decrease −0.18 percentage points annually after 2005 (95% CI, −0.68 to 0.32 percentage points) (Figure 1). Results based on the NAMCS and NHAMCS data were more pronounced, with a level change of −0.54 percentage points (95% CI, −3.18 to 2.10 percentage points) and a postwarning slope of −0.83 percentage points (95% CI, −1.27 to −0.39 percentage points). The decrease in AAP use after 2005 was accompanied by changes in use of other psychiatric medications (which included typical antipsychotics, hypnotics, antidepressants, antiepileptics, and antidementia agents) and opioids (Figure 2). The prevalence of use of any psychiatric medication or opioid increased by 0.48 percentage points (95% CI, −8.45 to 9.40 percentage points) in 2005 and continued to increase by 0.67 percentage points (95% CI, 0.34-1.01 percentage points) annually. We observed similar trends using data from the NAMCS and NHAMCS sample. When data were analyzed by drug class, we observed that the prevalence of antiepileptic use increased by 2.23 percentage points (95% CI, −0.98 to 5.44 percentage points) in 2005 and continued to increase at a rate of 1.21 percentage points (95% CI, 0.66-1.76 percentage points) per year. Although stable before 2005, the prevalence of opioid use decreased by −1.14 percentage points (95% CI, −4.73 to 2.45 percentage points) in 2005 but increased by 1.29 percentage points (95% CI, 1.01-1.58 percentage points) annually. However, in the NAMCS and NHAMCS, the prevalence of opioid use was lower (Figure 2). The prevalence of opioid use increased by 0.28 percentage points (95% CI, −0.03 to 0.58 percentage points) before 2005 and by 0.63 percentage points (95% CI, 0.13-1.12 percentage points) after 2005, but the slope change was not statistically significant. The prevalence of antidementia medication use continued to increase after 2005 but at a slower pace (eFigure in the Supplement).

Figure 1.

Proportion of Elderly Individuals With Dementia Who Have Used Psychiatric Medications and Proportion of Visits for Alzheimer Disease With Recorded Use of Psychotropic Medications

Data are from the Medical Expenditure Panel Survey (MEPS) (blue) and the National Ambulatory Medical Care Survey (NAMCS) and National Hospital Ambulatory Medical Care Survey (NHAMCS) (orange). MEPS obtains data on health conditions at multiple points during a 30-month period, whereas NAMCS and NHAMCS obtain data from medical records corresponding to a sampled visit. Data represent 3-year moving means. The vertical line at 2005 indicates the time that the boxed warning was implemented.

Figure 2.

Proportion of Elderly Individuals With Dementia Who Are Taking Selected Medications and Proportion of Visits for Alzheimer Disease With Recorded Use of Selected Medications

Data are from the Medical Expenditure Panel Survey (MEPS) (blue) and the National Ambulatory Medical Care Survey (NAMCS) and National Hospital Ambulatory Medical Care Survey (NHAMCS) (orange). MEPS obtains data on health conditions at multiple points during a 30-month period, whereas NAMCS and NHAMCS obtain data from medical records corresponding to a sampled visit. Data represent 3-year moving means. The vertical line at 2005 indicates the time that the boxed warning was implemented.

Health Events

The prevalence of cerebrovascular events increased by 0.75 percentage points (95% CI, 0.46-1.04 percentage points) per year before 2005, decreased by −0.74 percentage points (95% CI, −2.34 to 0.86 percentage points) in 2005, and decreased by −0.50 percentage points (95% CI, −0.71 to −0.28 percentage points) annually after 2005 (Figure 3). The patterns for cerebrovascular events estimated from the NAMCS and NHAMCS data showed stable to slightly increasing prevalence but with more year-to-year variability than the MEPS data. The prevalence of cardiovascular events increased by 1.42 percentage points (95% CI, −3.48 to 6.32 percentage points) in 2005 and continued to increase annually at 1.30 percentage points (95% CI, 0.30-2.30 percentage points) (Figure 3), with the NAMCS and NHAMCS data yielding similar findings. The prevalence of falls and/or fractures decreased before the 2005 warning and continued to decrease by −0.40 percentage points (95% CI, −0.56 to −0.24 percentage points) per year despite an immediate increase of 1.88 percentage points (95% CI, 0.87-2.88 percentage points) in 2005 (Figure 3). Two-year mortality risk varied between 6.13% and 13.73% (mean, 8.26%) from 1997 to 2013 and decreased by −0.68 percentage points (95% CI, −1.33 to −0.03 percentage points) annually before 2005. It increased by 2.04 percentage points (95% CI, −0.75 to 4.83 percentage points) in 2005 and continued to increase by 0.18 percentage points (95% CI, −0.07 to 0.42 percentage points) per year thereafter (Figure 4). After removing the outlier observation for 1997 (13.7%, corresponding to the means of 25.1% for 1996, 7.7% for 1997, and 8.4% for 1998 as indicated in eTable 5 in the Supplement), the prewarning slope decreased slightly to –0.26% and the level changed to 0.78%, resulting in a smaller overall change in 2-year mortality risk observed after 2005.

Figure 3.

Proportion of Elderly Individuals With Dementia Who Have Experienced Cerebrovascular Events, Cardiovascular Events, and Falls and/or Fractures and Proportion of Visits for Alzheimer Disease With Recording of Such Events

Data are from the Medical Expenditure Panel Survey (MEPS) (blue) and the National Ambulatory Medical Care Survey (NAMCS) and National Hospital Ambulatory Medical Care Survey (NHAMCS) (orange). MEPS obtains data on health conditions at multiple points during a 30-month period, whereas NAMCS and NHAMCS obtain data from medical records corresponding to a sampled visit. Data represent 3-year moving means. The vertical line at 2005 indicates the time that the boxed warning was implemented.

Figure 4.

Patient-Reported Outcomes and Mortality (Medical Expenditure Panel Survey)

Data represent 3-year moving means. SF-12 indicates Medical Outcomes Study 12-Item Short-Form Health Survey. The vertical line at 2005 indicates the time that the boxed warning was implemented.

Health-Related Quality of Life

HRQoL remained relatively unchanged (Figure 4). The range of scores observed for the MCS (mean, 2.07) and PCS (mean, 2.68) barely exceeded the lower bound of the range of minimally clinically important differences (range, 2-7) established for SF-12 summary scores, suggesting that changes in HRQoL were minimal.[36,37,38] The SF-12 MCS score before 2005 varied between 42.42 and 44.49 (mean, 43.51) and decreased by –1.63 percentage points (95% CI, –2.46 to –0.80 percentage points) in 2005 but increased by 0.24 percentage points (95% CI, 0.12-0.36 percentage points) per year after 2005. The PCS increased by 0.16 percentage points (95% CI, –0.53 to 0.84 percentage points) in 2005 but decreased by –0.29 percentage points (95% CI, –0.49 to –0.09 percentage points) annually after the FDA warning.

Discussion

To date, most studies[24,25,26,27,28] have evaluated the association of the 2005 FDA boxed warning with the use of antipsychotic medications. We used an ITSA to assess the long-term association of the 2005 FDA boxed warning with the prevalence of the use of AAPs, other psychiatric medications, and opioids; the risk of cerebrovascular and cardiovascular events and falls and/or fractures; mortality risk; and HRQoL. We found that the prevalence of AAP use decreased after the 2005 FDA boxed warning, which is consistent with previous reports[24,25,26,27,28] of decreases in AAP prescriptions in the years immediately after the warning. Although ITSA results are considered to be hypothesis generating, the current study builds on previous research by extending the study period to nearly a decade before and after the warning to adequately adjust for existing health care trends and to assess the long-term consequences of the boxed warning.

We found that the prevalence of antiepileptic and opioid use significantly increased contemporaneously with the decreased prevalence of AAP use. Physicians may have responded to the FDA’s warning by prescribing AAPs less often and adopting, instead, alternative therapeutic strategies for managing NPSs in elderly individuals with dementia.

The steady increase in opioid use by elderly patients with dementia observed after the 2005 FDA boxed warning in the MEPS is an important finding, especially in the context of the ongoing opioid epidemic. Although we did not observe a significant increase in opioid use after 2005 in the NAMCS and NHAMCS, the direction of the opioid use trend was in line with the MEPS data. The overall lower prevalence of opioid use detected from the NAMCS and NHAMCS may have contributed to the discrepancy. A study[39] of Medicare beneficiaries found that opioid use remained stable between 2007 and 2017, suggesting that the increase observed in the MEPS may be specific to the population of elderly patients with dementia. The reasons for the increased rate of opioid use are unclear because the evidence base for their use in treatment of dementia-related agitation is limited.[40] Although it is possible that opioids could have been used to treat pain in some cases because pain is increasingly recognized as a contributor to agitation,[40,41,42] further research is needed to understand the patterns and reasons for opioid use. Furthermore, regardless of the reasons for opioid use, practitioners should be reminded that in all patients with agitation, pain should be assessed and managed as clinically indicated, including the use of nonopioid analgesics.[15] As for the increase in antiepileptics, their mood-stabilizing effects may ameliorate some of the symptoms of agitation and aggression in patients with dementia, although evidence for their use is mixed.[43,44] Additional studies that compare the risks of antipsychotics with those of opioids or mood stabilizers in elderly patients with dementia are necessary.

Our findings of a lack of change in the trends of antidepressant and hypnotic use are in line with previous ITSA analyses[25,26] of the 2005 warning. Reports by Gallinin et al[26] and Kales et al[25] indicated no change in the trends of antidepressant and anxiolytic use before and after the 2005 warning.

We found, based on the MEPS data, a concomitant decrease in the use of AAPs and a decrease in cerebrovascular events after 2005, which are consistent with the hypothesized association between the use of antipsychotics and stroke.[17,18,45,46] We did not, however, find a similar decrease in cerebrovascular events using the NAMCS and NHAMCS data. This discrepancy may be attributable to the higher year-to-year variability in the prevalence of cerebrovascular events in the NAMCS and NHAMCS data and how those data were collected. The MEPS obtains data from individuals and their proxies on health conditions and health care services that they have received in any setting from physician and nonphysician practitioners at multiple points during a 30-month period. The NAMCS and NHAMCS are more restricted and obtain data from the medical records corresponding to sampled physician visits in the ambulatory setting or nonfederal short-stay general hospitals. In addition, only health conditions mentioned or pertaining to the sampled visit are recorded in the NAMCS and NHAMCS, thus potentially underestimating the prevalence of health conditions, as has been shown in a previous report.[47]

We also found that the prevalence of cardiovascular events increased after the 2005 FDA warning. The reason for the observed increase is unclear in light of evidence that hospitalizations for acute cardiovascular events have been decreasing based on a study[48] of Medicare claims between 1999 and 2011. Nevertheless, increased rates of cardiovascular adverse events may signal a negative unintended consequence of the 2005 warning.

The lack of change in mortality risk and HRQoL after decreases in the prevalence of AAP use and cerebrovascular events may reflect the countervailing effects of increases in the prevalence of opioid use and cardiovascular events. In addition, 2 years for each MEPS panel may not be long enough to observe the associations of changing medication use with mortality. Given these limitations, our findings with respect to mortality risk should be viewed only as hypothesis generating.

Findings related to falls and/or fractures were mixed because we observed a slower rate of decrease in the MEPS but an increased prevalence in the NAMCS and NHAMCS after 2005. This discrepancy may be attributable to how data were collected in these different data sets as described above. In addition, we cannot rule out that changes in use of centrally active medications, including hypnotics, antipsychotics, benzodiazepines, and antidepressants, may have been associated with the increase in falls observed in the NAMCS and NHAMCS data because all 4 medication classes have been associated with falls in a previous meta-analysis.[49] In addition, our findings of increased falls and/or fractures in the NAMCS and NHAMCS may reflect a population-related increase in fall-related mortality in US individuals 65 years or older during a similar period.[50]

Limitations

This study has limitations. Although the ITSA approach is regarded as a valuable study design for evaluating the effectiveness of population-level health interventions with a clearly defined point in time for implementation (eg, the introduction of a regulatory warning[51] and when randomization is not feasible),[52] our study was observational and could not account for all factors that may be associated with the studied outcomes. Although the annual demographic characteristics of our samples were stable during the study period, we were not able to assess unobserved patient, practitioner, or systemwide characteristics that may have been associated with the temporal patterns of the study outcomes. Thus, findings from this study should be considered as hypothesis generating and should not be used to suggest that the 2005 FDA warning was the only factor responsible for the changes that we observed.

A qualitative examination of the data revealed inflection points in the trends that did not always occur at 2005. Safety concerns with AAPs began in the 2000s,[16,17,18,19] and the 2005 FDA boxed warning on AAPs[16] was further expanded to conventional antipsychotics in 2008.[21] Thus, changes in practice patterns may have begun before 2005 and may have continued beyond the warning year. Before data analyses, we selected 2005 as the interruption year because this was the year of the FDA warning. Our choice is in line with previous studies that also used 2005 as the division point[25,27,28,35] and reported a reduction in AAP use after the 2005 warning.[24,25,26,27,28]

Although directionally similar trends were observed in the MEPS and the NAMCS and NHAMCS for the use of antipsychotics, antidepressants, psychiatric medications, opioids, antiepileptics, antidementia treatment, and cardiovascular events, different trends were observed between the data sources for the use of hypnotics, cerebrovascular events, and fall and/or fracture rates. These discrepancies may be attributable to the different sampling frames between the data sources, as discussed above. The MEPS captures a broader range of health events in more settings than the NAMCS and NHAMCS, and underestimation of health service visits has been reported in a previous article.[47]

There are also limitations arising from the data sources used in this study. Although the MEPS and the NAMCS and NHAMCS recruit large and nationally representative samples and are among the most well-known and cited sources of information about health care in the United States, certain aspects of their design may limit the interpretation of our findings. For example, the quality of the data on the presence of health conditions is limited by the ability of respondents and proxy respondents, in the case of the MEPS, to accurately recall the relevant information. Thus, diagnosis codes assigned to self-reported conditions without objective and independent verifications may be misclassified. Furthermore, these data sources represent cross-sectional observations rather than patient-level longitudinal data. Despite these limitations, we selected these data sources for this study because they provided the opportunity to assess a wide range of measures, such as medication use, diagnoses, mortality, and HRQoL, collected in a consistent manner during almost 2 decades. Furthermore, the study samples only included noninstitutionalized individuals. Thus, these findings are not generalizable to institutionalized and nursing home residents with dementia.

Conclusions

The prevalence of AAP use and cerebrovascular events continuously decreased during the decade after the 2005 FDA boxed warning on AAPs. The reduction in use of antipsychotics was associated with increases in the use of antiepileptics and opioids and with increased rates of cardiovascular events. Thus, the boxed warning on the use of antipsychotics may have been associated with unanticipated, potentially harmful risks. We believe that longitudinal studies would be helpful to support or refute these findings. Overall, our data suggest that new, effective, and safe options for treatment of elderly patients with dementia-associated NPSs are needed.