Age, Antipsychotics, and the Risk of Ischemic Stroke in the Veterans Health Administration
Background and Purpose—Time-dependent effects of antipsychotics on risk of stroke and potential effect modification by age have not been fully investigated. A case–case–time–control design uses within- and between-case comparisons to evaluate short-term effects at the same time as adjusting for unmeasured time-invariant confounders and exposure-time trends.
Methods—We conducted a case–case–time–control design study using data from the Veterans Health Administration. Veterans with inpatient hospitalizations for ischemic stroke between 2002 and 2007 were included. For every stroke case, the “current” exposure period was defined as 1 to 30 days before hospitalization and the “reference” period as 91 to 120 days before hospitalization. Exposure during the current period was compared with exposure during the reference period within cases. Exposure-time trend-adjusted estimates of the effect of antipsychotic exposure on risk of stroke were obtained by dividing exposure odds for antipsychotic exposure by average exposure odds for other medications over the same time period among the same cases.
Results—After adjusting for exposure-time trends, odds of stroke were 1.8 (95% CI, (1.7–1.9) times higher when exposed to antipsychotics than when unexposed. Age-stratified estimates suggest a greater triggering effect of antipsychotics among older patients.
Conclusions—Exposure to antipsychotics may be a proximal trigger for stroke. Elevation in risk is apparent after brief exposure to antipsychotics.
Warnings of increased risk of death and vascular adverse events associated with off-label use of antipsychotics among the elderly surfaced in 2002.1,2 In 2004, the UK Medicines and Healthcare Products Regulatory Agency warned clinicians that risperidone and olanzapine should not be used to treat behavioral and psychological symptoms of dementia due to increased risk of stroke; in 2005, in response to a meta-analysis of 17 placebo-controlled trials in elderly patients with dementia, the US Food and Drug Administration issued black box warnings to highlight the potential harms of atypical antipsychotic use among the elderly.1–3
The high prevalence of antipsychotic use among the elderly has persisted, due in part to continued controversy over the potential harmful effects as well as the lack of pharmacological alternatives for management of behavioral and psychological symptoms of dementia symptoms.4–6
Investigation of death or vascular events associated with antipsychotics among nonelderly adults has been limited. Previous studies of antipsychotics and stroke have compared “users” of antipsychotics with “nonusers” or compared atypical antipsychotics with conventional antipsychotics.5 These studies may have been affected by bias due to unmeasured differences between exposure groups.4,5,7 Classification of persons into exposure groups rather than person-time ignores the time-varying and transient nature of antipsychotic use in practice. It also disregards the biologically plausible window of effect after exposure to an antipsychotic. The majority of antipsychotics have half-lives of <24 hours, suggesting that no active drug is bioavailable within 1 week after discontinuation of medication.8 Additionally, evidence from clinical trials and observational studies has suggested that elevation in risk of stroke may begin soon after initiation of antipsychotics and dissipate with time.5,9–11
We use the case–case–time–control design to evaluate the role of antipsychotics as a proximal trigger for stroke among elderly and nonelderly adult veterans. Like other case-only designs, the case–case–time–control uses within-case comparisons to adjust for confounders that do not vary within the individual. The design adds an adjustment for exposure-time trends and bias that could occur if a prodrome of stroke resulted in increased use of antipsychotics in the time before the stroke event (protopathic bias).12
Materials and Methods
We used comprehensive, linked medical, pharmacy, and administrative data from the Veterans Health Administration (VHA) to analyze the risk of stroke related to antipsychotic use among elderly and nonelderly veterans. The patient population of the VHA is comprised primarily of older males with multiple physical and psychiatric comorbidities, a group at higher than average risk of stroke.13–15 In 2007, analyses of linked databases showed that 60% of antipsychotic prescriptions in the VHA were to patients without diagnoses of schizophrenia or bipolar disorder. Among patients with diagnoses of dementia, organic brain disorders or post-traumatic stress disorder, and no diagnoses indicative of a labeled indication for antipsychotic use, approximately 20% had exposure to antipsychotics.16
Deidentified data from the VHA Decision Support System including data from the medical SAS data sets (inpatient and outpatient encounters) as well as pharmacy and laboratory data sets for Fiscal Years 2002 to 2007 were extracted. Information from each data set was linked for all veterans who had a hospitalization for ischemic stroke (International Classification of Diseases, 9th Revision codes 433.X1, 434.X1) and no previously recorded inpatient or outpatient encounters associated with a diagnosis of stroke (N=17 422). We defined ischemic stroke using a modification of Reker et al's high specificity stroke diagnostic code algorithm, which was validated within the VHA system.17 Reker's definition, which included both ischemic and hemorrhagic stroke diagnoses, had a positive predictive value of 75%.17 If >1 stroke hospitalization occurred for a particular patient during the follow-up window, the first recorded admission was used for the analysis.
To limit bias from patients receiving healthcare outside of the VHA, we restricted the study to active users of the VHA healthcare system. Active use was defined as at least 1 physician visit and 1 pharmacy dispensing in the 6 to 12 months before hospitalization for stroke (N=14 671).
The means test, a financial assessment tool that determines veterans' copayment responsibilities, indicated that among patients in our sample with transient exposure to antipsychotics before their hospitalization, 91% were exempted from copayments.18
The case–case–time–control design uses a within-case comparison identical to that used in a case–crossover design, comparing the exposure odds in the period during which the outcome event occurs with the exposure odds in a referent period comprised of person-time sampled from the case before the event occurrence.19 An adjustment for potential bias from exposure-time trends is added by calculating an exposure OR using person-time sampled from cases to estimate the average exposure OR for treatment with prescription medications over the same time window. This is an extension of the case–time–control design proposed by Suissa, which uses person-time sampled from a noncase–control group to estimate exposure trends in the population over time.20 The crossover analysis estimating the exposure OR for medications other than the medication of interest provides an estimate of the expected trend in exposure to treatment in the time leading up to a stroke event. Trends in exposure are likely to differ across patient populations; estimates for time trends calculated using person-time sampled from the cases themselves can more accurately represent the expected trends among that population than estimates derived from an external, noncase–control population.
Crossover analyses are used to evaluate transient exposures or exposure with transient effect. Because within-subject comparisons are used, only cases with transition in exposure contribute to estimation of the strength of the exposure–outcome relationship; information from cases that are always exposed or never exposed does not contribute. We used a logistic regression model conditioning on individual cases, in which exposure to antipsychotics was predicted by time (current versus reference) and a stratified logistic regression model conditioning on the individual, which calculated the average OR for exposure to nonpsychotropic classes of medication over the same time period. The model predicting exposure to the other classes of medication provides an estimate of the trend for increased exposure to treatment in the time before a stroke event. The estimated OR for exposure to antipsychotics in the current period is divided by the average OR for exposure to other classes of medication to obtain an effect estimate that is adjusted for the exposure-time trend. We obtained bootstrapped 95% CIs for the time-adjusted estimate.
Dates of dispensing and the days supplied for antipsychotics dispensed from a VHA pharmacy were identified for each stroke case. We classified days between the dispensing date and the end of the days supplied as exposed days. An additional 7 days was added to the days supply for outpatient prescriptions to allow for nonadherence. For each stroke case in the sample, the date of hospitalization was the index date for the current period. If there was a minimum of 3 exposed days within the 30 days before this index date, the patients were considered exposed during the current period. Similarly, cases were considered exposed during their referent period if there was a minimum of 3 exposed days within the 30 days before the referent index date. The referent index date for each case was 90 days before their hospitalization. The minimum exposure duration and the lag between index dates were selected based on the average half-life of the various antipsychotic medications and evidence that the effect of antipsychotics on risk of adverse events starts early and is time-limited. Exposure to other classes of medication during the current and referent periods was defined using the same criteria as for antipsychotics.
In sensitivity analyses, we varied the minimum exposure duration (3, 7, 14 days), the lag time between index dates (90, 180 days), and the number of days added to outpatient dispensations for nonadherence (0, 7, 14 days).
Our sample included 511 stroke cases with crossover in exposure over the current and referent periods (Table 1). Cases with transient exposure to antipsychotics were predominantly male and between the ages of 60 and 90 years. Examining claims generated during the 30-day current and referent periods, the number of days with healthcare encounters and number of medication dispensations was greater in the 30 days immediately before the stroke than in the 90 to 120 days before the stroke. Among cases with crossover in exposure, 85% were exposed to antipsychotics in the current period, whereas 15% were exposed in the referent period.
The results of the case–case–time–control analysis are displayed in table 2. We present estimates for the strength of the relationship between exposure to antipsychotics and stroke after adjusting for the effect of time-invariant within-person confounders as well as exposure-time trends. The odds of stroke were 1.8 (95% CI, 1.7–1.9) times greater when exposed to antipsychotics for at least 3 of the last 30 days than when there were <3 days of antipsychotic exposure within 30 days. Age-stratified estimates suggest a greater triggering effect of antipsychotics on stroke with increasing age.
This case–case–time–control of antipsychotic exposure and risk of ischemic stroke uses within-case comparisons to remove bias from measured or unmeasured between-person confounders and uses estimates of exposure-time trends over the same time period to adjust for potential exposure-time trends and protopathic bias.
Protopathic bias is a potential noncausal explanation for the association between antipsychotic use and stroke. This bias can occur when the exposure of interest is a medication that is “prescribed for early manifestations of a disease that has not yet been detected.”21 Cognitive deterioration could be a prodrome of stroke related to early, undetected ischemia. It is plausible that an association could be observed due to the link from cognitive deterioration to worsening behavioral and psychological symptoms of dementia, which in turn might lead to antipsychotic exposure.22 More generally, any patient deterioration related to an imminent stroke, which is also associated with antipsychotic treatment, can confound the analysis.
Our exploration of exposure-time trends indicates that there is an increased propensity for healthcare contact and receipt of treatments in the time immediately preceding a stroke; this trend may be reflective of the presence of protopathic bias. Although it has been well documented that prescribing of antipsychotics in the United States rose sharply between 2002 and 2007, each within-case comparison spanned only a 4-month period. Although it is unlikely that population-level trends affected individual prescribing in such a brief span of time, the person-time sampled from future cases was matched to the current and referent periods of each case on calendar time to adjust for potential biasing effects of population trends in prescribing.
The case–case–time–control analysis suggests that risk of stroke is elevated during periods of antipsychotic exposure relative to baseline risk and that this risk may increase with age. Sensitivity analyses using alternative exposure metrics, which varied the minimum exposure duration, the lag time between case and referent index dates, and the number of days added to outpatient dispensations to allow for nonadherence, yielded similar results (Appendix A). Our sample size was not large enough to allow us to examine specific agent or dose-related effects.
Our sample included only veterans who were active users of the VHA healthcare system and were hospitalized at a VHA hospital for their stroke. Nearly all of our sample was assessed to be in financial need and did not have copayments in the VHA system. Veterans in financial need are unlikely to use other sources of health care in which copayments are required. Whereas our restricted sample lowers the risk of incomplete capture of treatments received from other sources, it also reduces the generalizability of our results. Poverty-stricken elderly patients may be sicker and have more risk factors for stroke than their wealthier elderly counterparts. The presence of these risk factors may be a prerequisite for antipsychotics to act as a trigger for stroke.
Indeed, there are multiple contributing causes to a stroke event. For an exposure to trigger stroke, other contributing causes must be present. Among older patients already at high risk, exposure to antipsychotics may be a strong proximal trigger for stroke.
Further investigation of antipsychotic safety should be conducted in a larger and more diverse sample of patients in which finer stratifications of age, gender, race, and potential modification of risk by dementia/Alzheimer status can be examined.
Our findings provide evidence that even brief exposure to antipsychotics can result in adverse health outcomes. We recommend that antipsychotic therapy among the elderly be initiated only after alternative strategies of dealing with symptoms of behavioral and psychological symptoms of dementia have been fully investigated. Weighing the risks and benefits of antipsychotic treatment is particularly important for providers treating patients with multiple comorbid risk factors for stroke.
Sources of Funding
This material is the result of work supported with resources and the use of facilities at the Center on Systems, Outcomes, and Quality in Chronic Disease and Rehabilitation, a Research Enhancement Award Program of the Health Service Research and Development Service (Grant no. REA08-263) at the Providence Veterans Affairs Medical Center, Providence, RI. Predoctoral fellowship support provided by a National Research Service Award, Grant No. 2T32 HS000011 from the Agency for Healthcare Research and Quality. Graduate Student Support for Research Assistants was provided by Pfizer Epidemiology Award A 29843-001, Grant 140-N-502134-V2.
The online-only Data Supplement is available at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.111.617191/-/DC1.
The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the US Department of Veterans Affairs or the US government.
This manuscript will be presented as poster and podium presentations at the 2011 Epidemiology Congress in Montreal and the 2011 International Conference on Pharmacoepidemiology and Therapeutic Risk Management in Chicago.
- Received February 15, 2011.
- Revision received June 9, 2011.
- Accepted June 10, 2011.
- © 2012 American Heart Association, Inc.
- Salzman C,
- Jeste DV,
- Meyer RE,
- Cohen-Mansfield J,
- Cummings J,
- Grossberg GT,
- et al
- Curtis D
- Kizer KW
- Willich SN,
- Maclure M,
- Mittleman M,
- Arntz HR,
- Muller JE
VA Information Resource Center; VIReC Research User Guide: VHA Decision Support System Clinical National Data Extracts. 2nd ed. Hines, IL: US Department of Veterans Affairs, Health Services Research and Development Service, VA Information Resource Center; 2009. http://www.virec.research.va.gov/References/RUG/RUG-DSS-2nd-Ed-er.pdf.
- Maclure M