Antipsychotic medications and stroke in schizophrenia: A case-crossover study

Background The association between antipsychotic use and the risk of stroke in schizophrenic patients is controversial. We sought to study the association in a nationwide cohort with schizophrenia. Methods Using a retrospective cohort of patients with schizophrenia (N = 31,976) derived from the Taiwan National Health Insurance Research Database, 802 new-onset cases of stroke were identified within 10 years of follow-up (from 2000 through 2010). We designed a case-crossover study using 14-day windows to explore the risk factors of stroke and the association between antipsychotic drugs and the risk of stroke. We analyzed the risks of individual antipsychotics on various subgroups of stroke including ischemic, hemorrhagic, and other strokes, and the risks based on the antipsychotic receptor-binding profile of each drug. Results Use of any second-generation antipsychotic was associated with an increased risk of stroke (adjusted risk ratio = 1.45, P = .009) within 14 days while the use of any first-generation antipsychotic was not. Intriguingly, the use of any second-generation antipsychotic was associated with ischemic stroke but not hemorrhagic stroke. The antipsychotic receptor-binding profile analysis showed that the antihistamine 1 receptor was significantly associated with ischemic stroke (adjusted risk ratio = 1.72, P = .037), and the sensitivity analysis based on the 7-day window of exposure validated the association (adjusted risk ratio = 1.87, P = .015). Conclusions Use of second-generation antipsychotic drugs appeared to be associated with an increased risk of ischemic stroke in the patients studied, possibly mediated by high affinity for histamine-1 receptor blockade. Further research regarding the underlying biological mechanism and drug safety is suggested.


Introduction
Schizophrenia is a debilitating mental illness, and the lifespan of patients with schizophrenia is estimated to be 12-15 years shorter than that of the general population [1]. This reduction of lifespan could be due to comorbid physical illnesses such as stroke. A recent meta-analysis showed that schizophrenia patients appeared to be at greater risk of stroke [2]. Clinical comorbidity, such as diabetes, hypertension, and hyperlipidemia may be linked to the development of atherosclerosis, which is a pathogenesis shared with stroke [3]. Other factors that could contribute to the stroke event include unhealthy lifestyles, poor healthcare, and drug exposure [4,5]. Antipsychotics are crucial for the treatment of schizophrenia, and drug safety is always of critical concern. The dose and duration of antipsychotic use for treating schizophrenia are even higher and longer than prescriptions for other psychiatric conditions. Therefore, the role of antipsychotic drugs in the causal relationship of stroke in patients with schizophrenia needs clarification.
Most studies on the effects of antipsychotics on stroke risk were conducted in elderly patients and were seldom specific to schizophrenia patients. The pooling of results of metaanalyses from several randomized, placebo-controlled clinical trials showed an increased risk of cerebrovascular events in patients with dementia receiving second-generation antipsychotics (SGAs) [6]. A self-controlled case-crossover study [5] demonstrated that both FGAs and SGAs were associated with the risk of stroke among elderly patients. In contrast, a populationbased case-control study [7] showed that only the use of FGAs was slightly associated with the risk of stroke but not with the use of SGAs. Thus, the above studies showed inconsistent findings. Furthermore, we hypothesized that the previous findings in elderly populations should not be generalized directly to schizophrenia patients, which is a topic that still deserves further study.
In previous studies with case-control designs, the limitations have concerned unrecognized confounding factors, which could bias the estimation of the risks [8]. Prior studies [9,10] have shown a temporal relationship between antipsychotic use and stroke as an acute effect (most significant within 7-14 days) instead of a long-term effect. Considering this knowledge gap, we applied a case-crossover design in our study such that the association of drug exposure and risk of stroke was identified an acute effect [9,10]. The case-crossover design has been widely used [11,12] when a brief exposure has caused a transient change in the risk of a rare acuteonset disease, in which a medication change was an exacerbating factor in acute stroke. The design only involved the cases, and intriguingly, self-matching of cases further eliminated the threat of control-selection bias and increased efficiency [13]. Therefore, this design has the advantage that the study subjects serve as their own controls and can remove unmeasured time-invariant confounders [11,12].
In addition, the mechanism of antipsychotic-associated stroke remains speculative and may involve different pathophysiological pathways, ranging from the facilitation of thrombosis, pre-existing cardiovascular factors, sedation, and a common diathesis for strokes in psychiatric disorders [14]. However, there is little evidence to support these hypothetical mechanisms. The receptor-binding profiles of antipsychotics are associated with specific side effects. Therefore, these profiles can provide an opportunity to understand the mechanism of associations between receptor-binding of antipsychotics and stroke [8].
In this study, we assess the association between antipsychotic exposure and the risk of stroke by employing a large schizophrenia cohort derived from a nationwide population-based insurance claim dataset. The effects of antipsychotics on different types of stroke were evaluated, and a possible mechanism of action is hypothesized based on the receptor-binding profiles of the antipsychotics.

Data sources
The National Health Institutes Research Database (NHIRD) in Taiwan is available to researchers upon application approval [15]. The NHIRD contains the Psychiatric Inpatient Medical Claims database (PIMC), which includes patients whose admitting diagnosis, based on the International Classification of Diseases, Ninth Revision (ICD-9), included codes ranging from 290 to 319 from January 1, 1996, to December 31, 2008 (N = 187,117). The veracity of the database is ensured by the periodic review and recertification of each hospital that provides psychiatric hospitalization in Taiwan. The Taiwan Joint Commission on Hospital Accreditation, an independent, non-governmental organization, oversees hospital accreditation. The accreditation for qualified psychiatric services requires board-certified psychiatrists to diagnose inpatients. In addition, this database has been the source for numerous epidemiological and clinical studies published in peer-reviewed journals [8,10,15].
For defining the longitudinal retrospective cohort with possible incident schizophrenia instead of prevalent schizophrenia, we first selected patients from the database with at least one psychiatric hospital admission between 2000 and 2008 but no psychiatric admissions between 1996 and 1999 to recruit the patients who had had their first incident of mental illness requiring psychiatric hospitalizations (n = 125,225). Then, the inclusion criteria for the subjects were at least one discharge diagnosis of schizophrenia (ICD-9-CM code 295.xx) with age at the first psychiatric admission ranging from 18 to 65 years (n = 33,024). All of the prescription records of these patients were retrieved from 2000 through 2010. Each prescription record contained the medications, dosage, and duration of the drug regimen. Patients who were diagnosed with stroke (ICD-9-CM codes 430.x to 436.x) before the first psychiatric admission (baseline) were excluded (n = 1048).
We enrolled patients with an incident stroke after their schizophrenia diagnosis as cases (n = 802), defined as those with a primary diagnosis of hospitalization for cerebrovascular events encoded as ICD-9-CM 430, 431, and 432 for hemorrhagic stroke, 433, 434, and 435 for ischemic stroke, and 436 for other types of stroke. The onset of stroke (index date) was defined as the date of the first-time hospitalization or using an emergent service, with a diagnosis of stroke. Fig A in S1 File presents the flowchart of subject enrollment. Sociodemographic and clinical characteristics were recorded, including age at first psychiatric admission, age at stroke event, Charlson comorbidity score [16,17], the number of psychiatric admissions within one year before the index date and cumulative days of any antipsychotic use within one year before the index date.

Ethics statement
The Institutional Review Board of Taipei City Hospital approved this case-crossover study using national claims data from the NHIRD. A waiver of informed consent was granted since the patient information was de-identified before the analysis. All researchers signed an agreement guaranteeing patient confidentiality before using the database. The National Health Research Institute in Taiwan established the NHIRD to provide data for research purposes. Please find the website (http://nhird.nhri.org.tw/en/) to request the information this study used, such as a point of contact and data files. The database constitutes medical claim files representative of the entire population in Taiwan.

Case-crossover study
The case-crossover design is a standard method for assessing the effect of brief exposures on acute outcomes [13,18], and thus, is suitable for our purpose to reveal the changes of antipsychotic exposures that vary over time and produce a transient change in the risk of an acute stroke within a short period. With this study design, each case subject served as their own control. We assessed potential factors during a 14-day window as in a similar previous study [10] and compared the 14-day case period (1-14 days just before the stroke) with four individual 14-day control periods (every 13 weeks: 92-105 days, 183-196 days, 274-287, and 365-378 days before the onset of stroke).

Time-variant factors
According to the defined time windows, we assessed the association of stroke with the potential factors in schizophrenia. The potential factors comprised several physical illnesses including hypertension, non-hypertensive cardiovascular disease, diabetes mellitus, cancer, chronic hepatic disease, asthma, upper respiratory tract infection, and delirium. Additionally, concomitant medications included cardiovascular drugs, lipid-modifying agents, drugs used in diabetes, antithrombotic agents, corticosteroids, anti-Parkinson drugs, respiratory drugs, and benzodiazepines. The potential factors were selected as candidate risk factors due to their potential associations with the risk of stroke [19]. The identified risk factors also served as covariates for assessing the associations of antipsychotics, mood stabilizers, and antidepressants with the risk of stroke.

Exposure to antipsychotics, mood stabilizers, and antidepressants
We retrieved prescription drug data on the use of antipsychotics, mood stabilizers, and antidepressants. We defined antipsychotics (N05A) according to the Anatomical Therapeutic Chemical Classification System (ATC/DDD Index 2015. http://www.whocc.no/atc_ddd_index/ [accessed June 6, 2015]) and classified them as either FGAs or SGAs.
We obtained the binding affinities for a variety of neurotransmitter receptors, including serotonin (5HT [hydroxytryptamine] 1A, 5HT2A, 5HT6, and 5HT7), dopamine (D2 and D4), histamine (H1), muscarinic (M1), and adrenergic receptors (A1 and A2) from the National Institute of Mental Health Psychoactive Drug Screening Program Ki Database [20]. Patients with antipsychotics for which the Ki value for binding affinity was not available in the database were categorized as others in the analyses of the neurotransmitter receptor-binding profiles and stroke risk. Frequently prescribed antipsychotics (more than 2.0% for each in the case period) including chlorpromazine, haloperidol, flupentixol, sulpiride, clozapine, olanzapine, quetiapine, zotepine, risperidone, amisulpride, and aripiprazole were assessed based on the antipsychotics' dissociation constants for the affinities to a variety of receptors (Table A in S1 File). Each antipsychotic drug was categorized into either a high or a low group, based on the median values of the binding affinity for each neurotransmitter receptor.

Statistical analysis
We conducted a two-step multivariate, conditional logistic regression analysis. First, we explored physical illnesses and concomitant medications as potential risk factors for stroke. We conducted the multivariate regression based on the strategy of backward variable selection. The variables with a modest association (P < .20) were retained in the final model. Second, we investigated the effects of individual antipsychotic drugs, mood stabilizers, and antidepressants on the risk of stroke. Because the prescription of psychiatric drugs in the different time windows could be confounding by indications with the presence of comorbid physical illnesses and concomitant medications [21], we conducted a propensity score adjusted regression analysis [22] to estimate the effect of the individual psychiatric drug on the risk of stroke.
Subgroup analyses were performed to estimate the effects of the antipsychotics on the various types of stroke by restricting the outcomes to ischemic, hemorrhagic, and other types of stroke due to their different underlying mechanisms. The receptor affinity of the antipsychotics on the risk of stroke was analyzed based on four categories (no use, high affinity, low affinity, and others), and we used the physical illnesses and concomitant drugs as covariates for the multiple regression analysis. All analyses were conducted using SAS software, version 9.2 (SAS Institute Inc., Cary, NC, USA), and a P-value of less than .05 was considered significant.
We further computed the odds ratios using two different time windows of 7 days and 28 days before the incident stroke for the sensitivity analysis.

Characteristics of the study subjects
The study cohort included 31,976 schizophrenia patients, and 802 of them had incident strokes. The annual incidence of stroke was 34.6 cases per 10,000 person-years. Among the 802 cases, their age (SD) at first psychiatry admission was 50.0 (12.2) years and the mean (SD) time to the incident stroke was 3.9 (2.7) years during the 10-year follow-up period. Table 1 provides the demographic features and reveals that more than two-thirds of patients had the first stroke after their 46th year, and more than half (56.6%) of the patients with strokes were men. The types of stroke included ischemic (55.6%), hemorrhagic (29.3%), and others (15.1%). Most of the patients (93.7%) had Charlson comorbidity scores of less than three at their first psychiatry admission, and only 12.8% of patients showed no cumulative days of any antipsychotic use within one year before the incident stroke.

Effect of antipsychotics, mood stabilizers, and antidepressants on the risk of stroke
Based on the propensity score adjusted analysis, Table 3 reveals that FGAs are not associated with the risk of stroke while SGAs are significantly associated with an elevated risk of stroke (adjusted RR = 1.45, P = .009). Among the individual SGAs, quetiapine and zotepine were significantly associated with the risk of stroke. Among the mood stabilizers, valproic acid was the only one with an increased likelihood of stroke (adjusted RR = 2.68, P < .001). Use of antidepressants was not significantly associated with the risk of stroke. Regarding possible differences due to sex, further analyses show equivalent effect sizes of the adjusted risk ratios for the drugs studied in both men and women. Although some analyses showed adjusted risk ratios in men were significant, they were insignificant in women. For example, the adjusted risk ratios of any SGA on the stroke risk in men was 1.48 (p<0.05) and that of women was 1.43 (p = 0.089).

Subgroup analysis and effect of antipsychotic receptor-affinity profiles
The subgroup analyses revealed SGAs were significantly associated with ischemic stroke but not hemorrhagic stroke or other types of strokes (Fig 1). We further analyzed the risk of ischemic stroke based on the antipsychotic drug receptor-binding affinities in Table 4. After adjusting for the covariates, only a high affinity for histamine-1 receptor was associated with increased ischemic stroke risk. There were no significant associations for hemorrhagic and other types of strokes (Tables B and C in S1 File).
The results for the 7-and 14-day time windows as the case and control periods are grossly consistent in the sensitivity analysis, which indicate the robustness of our findings (Table D in S1 File). Compared with patients who did not use antipsychotics, the adjusted RRs for ischemic stroke for the 7-day and 14-day windows for high-affinity histamine-1 receptor were 1.87 (P = .015) and 1.72 (P = .037), respectively. The results using the 28-day time window as the case and control periods showed no significant association with any receptors (Table E in S1 File). The findings imply an acute or subacute effect (limited to 14 days) for antipsychotic drug use on the risk of ischemic stroke.

Main findings
In our study, the annual incidence of stroke in the patients with schizophrenia aged 18-65 years was 34.6 cases per 10,000 person-years. According to a prior study in Taiwan [23], the annual incidence of first-ever stroke is 21.3 per 10,000 person-years for people aged 36-64 years in the general population. Evidence indicates that the incidence of stroke increases with age, therefore, after considering the age stratification, our findings are consistent with the previously published literature [2] and reveal an increased risk of stroke in schizophrenia patients compared with the general population.
Our study is the first to apply a case-crossover design in the study of schizophrenia patients to investigate the association between antipsychotics and the risk of stroke. We found that any acute use (within two weeks before the onset of a stroke) of SGAs was associated with a 1.45-fold increased risk of stroke after adjusting for confounding factors. Higher stroke risk was observed in schizophrenia patients treated with quetiapine, zotepine, or valproic acid. Intriguingly, any use of an SGA was associated with ischemic stroke but not hemorrhagic stroke. Antipsychotics with higher binding affinity to H1 receptors could be the mechanism associated with the risk of ischemic stroke. Additionally, our results revealed that exposure to antihistamines and benzodiazepines increased the risk of stroke in schizophrenic patients.

Drug exposure and the risk of stroke
We found no previous research directly linking the systemic use of antihistamines to the likelihood of stroke in human beings. An animal study [24] revealed that histamine interacted simultaneously with histamine-1 receptors on the endothelium and histamine-2 receptors on  Table 4. Risk of ischemic stroke with antipsychotic use within the 14-day risk period according to the receptor-binding profile of the antipsychotics (n = 446).

Control period 1 N (%)
Control period 2 N (%) vascular muscle to mediate relaxation of the vessel. An in vitro study revealed a modulating interaction for histamine and second messengers in endothelium-dependent responses of microvascular compartments in the human brain [25]. Therefore, the systemic use of an antihistamine could interfere with histamine-mediated vasoconstriction or relaxation, which could lead to an increased risk of stroke. The identified association between antihistamine and risk of stroke could be related to one of the significant findings of this study, which showed higher antipsychotic drug receptor-binding affinity to histamine-1 receptors was associated with the risk of ischemic stroke. Our results also suggest a possible association between benzodiazepine use and the risk of stroke. Several explanations for this association are that brain blood flow and glucose consumption decrease following administration of a GABA agonist [26], and sedation with benzodiazepines transiently allows the re-emergence of focal motor deficits in patients with prior stroke [27]. In addition, our findings show that valproic acid significantly increases the risk of stroke in schizophrenia patients. Valproic acid can affect both pro-coagulatory and anticoagulatory factors [28]. Time-dependent thickening of the carotid artery intima-media was observed in epileptic adults treated with valproic acid monotherapy, which could contribute to the acceleration of atherosclerosis [29], leading to stroke. Since valproic acid is widely used in patients with bipolar disorder [30], the risk deserves future studies in bipolar populations.

Adjusted risk ratio a P-value
Among the individual SGAs, we confirmed that quetiapine and zotepine were associated with an increased risk of stroke. Quetiapine can cause weight gain as it blocks histamine-1 receptors, which also contributes to sedation, especially when combined with its M1-antimuscarinic and alpha-1 adrenergic antagonist properties [31]. Histamine 1 and alpha-1 adrenergic antagonist properties are also prominent in zotepine [31]. Sedation-related adverse effects may be the mediating factors of the association.

Receptor-binding profile of antipsychotics and risk of ischemic stroke
Most previous studies have pooled outcomes of various types of stroke together. However, ischemic stroke and hemorrhagic stroke have distinctly different mechanisms [32]. Subgroup analysis in this study revealed that any use of an SGA was significantly associated with ischemic stroke but not hemorrhagic stroke or other types of stroke. We further conducted an analysis to identify a biological basis for the association between antipsychotics and the risk of ischemic stroke based on drug receptor-binding profiles.
We demonstrated that antipsychotics inducing histamine type 1 blockade were associated with a transient increase in the risk of ischemic stroke. This finding is reasonable in the case- crossover design, which is appropriate for short-term exposures of effect. The risk decreased from the 7-day time window to the 14-day window and diminished by the 28-day window, which implies that the mechanism of antihistamine action on ischemic stroke is an immediate or subacute, rather than a long-term process. Somnolence appeared to be a common side effect of antipsychotics with a high affinity to histamine-1 receptor blockade [31]. Excessive sedation could lead to venous stasis or dehydration and hemoconcentration, which might predispose a patient to ischemic stroke. Both quetiapine and zotepine are high-affinity histamine 1-receptor blockers. These findings corroborate our discussion about systemic antihistamine use increasing stroke risk in schizophrenic patients. Further physiological studies are needed to confirm these suggested mechanisms.

Limitations
Several limitations should be considered in our study. First, despite most stroke patients undergoing image testing for diagnosis, the actual reports of the imaging examinations were unavailable in the NHIRD [8]. Therefore, we only categorized stroke into ischemic, hemorrhagic, and other types; we were unable to explore the associations between antipsychotic use and specific stroke subtypes [32]. Second, the receptor binding affinities of the antipsychotics were measured in in vitro studies; therefore, the actual activity of the antipsychotics on the receptors in the human brain deserves further clarification. Third, due to the use of claims databases in this study, there was no information regarding the potential confounding factors such as obesity, smoking, other unhealthy lifestyles or drug compliance. However, we used a self-controlled study design to adjust for such unmeasured variables.

Conclusions
Higher stroke risk was observed in schizophrenia patients acutely treated with quetiapine, zotepine, and valproic acid. Furthermore, use of any SGA was associated with an acutely increased risk of ischemic strokes but not hemorrhagic strokes in patients with schizophrenia. Mediation by high affinity for histamine-1 receptor blockade of antipsychotics is a possible mechanism. Further research is needed regarding the underlying biological mechanisms and drug safety.
Supporting information S1 File. Fig A. Study flow diagram. Table A. Receptor binding affinity (Pki) for antipsychotic drugs. Table B. Risk of hemorrhagic stroke with antipsychotic use within 14-day risk period by the receptor-binding profiles of the antipsychotic drugs (n = 235). Table C. Risk of other stroke with antipsychotic use within 14-day risk period by the receptor-binding profiles of the antipsychotic drugs (n = 121). Table D. Risk of ischemic stroke with antipsychotic use within 7-Day risk period by the receptor-binding profiles of the antipsychotic drugs (n = 446). Table E. Risk of ischemic stroke with antipsychotic use within 28-day risk period by the receptor-binding profiles of antipsychotic drugs (n = 446). (DOC)