SEARCH

SEARCH BY CITATION

Keywords:

  • Antiepileptic drugs;
  • Stroke risk;
  • Pharmacoepidemiology;
  • Patients with epilepsy

Summary

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. Disclosure
  9. References
  10. Supporting Information

Purpose:  Patients with epilepsy have higher stroke-related morbidity and mortality, leading to the suspicion that the increased stroke events may be associated with antiepileptic drug (AED) exposure. We evaluated the comparative risk of stroke in adult patients with epilepsy receiving phenytoin (PHT), valproic acid (VPA), or carbamazepine (CBZ) to help determine the stroke risk for Asian patients with specific AED exposure.

Methods:  We conducted a population-based, retrospective cohort study using the Taiwan National Health Insurance Research Database (NHIRD). The cohort consists of adult patients with epilepsy who were new to PHT, CBZ, or VPA monotherapy and without prior stroke history. Patients were followed for 5 years. The event of interest was a hospitalization or emergency visit due to stroke. Cox proportional hazard models were used to estimate the comparative risk of AEDs. Subanalyses included an evaluation of different subtypes of stroke, the propensity score matched technique, the intention-to-treat approach, and stratification analyses.

Key Findings:  Patients receiving PHT had a significantly higher stroke risk (adjusted hazard ratio [HR] 1.72; 95% confidence interval [CI] 1.20–2.47), followed by VPA (adjusted HR 1.27; 95% CI 0.78–2.07), when compared with CBZ. The results of all subanalyses showed a consistent trend of higher stroke risk with PHT use. In addition, there appeared to be a dose–response relationship between stroke risk and PHT prescriptions.

Significance:  The stroke risk was higher in PHT but not significantly different in VPA as compared to CBZ. Physicians should reconsider using PHT for patients with epilepsy who already have a higher risk of stroke.

Epilepsy is a common and serious neurologic disorder worldwide, with an estimated prevalence rate of 0.5–1% in the general population (Duncan et al., 2006). It is a chronic dynamic medical problem, and most patients require long-term therapy with antiepileptic drugs (AEDs) (Perucca & Tomson, 2011). Recently, concern has been raised that AEDs might increase the levels of several serologic markers associated with an increased risk of vascular diseases (Mintzer, 2010). For example, cytochrome P450 (CYP) enzyme–inducing AEDs, such as carbamazepine (CBZ), increase serum levels of total cholesterol, low-density lipoprotein cholesterol (LDL-C) (Nikolaos et al., 2004; Tomoum et al., 2008), and lipoprotein (a) (Bramswig et al., 2003). Several AEDs are also known to decrease serum levels of folic acid and vitamin B12, as well as to increase levels of homocysteine (Hcy) (Belcastro et al., 2010; Linnebank et al., 2011) and total oxidative stress (Hamed et al., 2007). The duration of AED exposure also predicts acceleration of atherosclerosis, independent of other risk factors such as age, gender, and oxidative stress (Tan et al., 2009). Based on these mechanisms, patients exposed to AEDs could have an increased risk of vascular diseases such as stroke.

Studies have found that patients with epilepsy have higher stroke-related morbidity (Gaitatzis et al., 2004) and mortality (Cockerell et al., 1994; Nilsson et al., 1997; Ding et al., 2006), leading to the suspicion that the increased stroke events may be associated with AED exposure (Olesen et al., 2011a). However, evidence concerning the comparative stroke risk of AEDs is limited. To our best knowledge, only one study, which is by Olesen et al., (2011a) and based on Caucasian populations, found that valproic acid (VPA) had a lower stroke risk than CBZ but found no significant difference between phenytoin (PHT) and CBZ. Indirect evidence in another study (Chuang et al., 2012) found that the carotid intima-media thickness (IMT) was increased more in Asian patients receiving PHT than CBZ, indicating that PHT may lead to higher stroke risk. It is well known that the varying prevalence of CYP family genotypes in different ethnic groups (Jaja et al., 2008) affects the metabolism of AEDs, so the risk profile of stroke in Asians may differ from Caucasians. Therefore, the purpose of this study was to determine the comparable stroke risk of specific AEDs in adult Asian patients with epilepsy.

Methods

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. Disclosure
  9. References
  10. Supporting Information

Data source

Electronic datasets containing information on 2 million individuals randomly sampled from the National Health Insurance Research Database (NHIRD) in Taiwan were used for this retrospective study. The NHIRD is maintained by the National Health Research Institute and made accessible for research purposes. Taiwan launched a single-payer and compulsory National Health Insurance (NHI) program on 1 March 1995, and by 2007, nearly 99% of the population had enrolled in this program. The NHIRD compiles information on enrollees, health care professionals and facilities, as well as service claims from inpatient and ambulatory care and contracted pharmacies for reimbursement purposes. Personal identities have been encrypted for privacy protection, but all datasets can be cross-linked with the unique and anonymous identifiers created by the National Health Research Institute for this purpose. Use of the NHIRD for research purposes is therefore exempt from the institutional review board (IRB) in Taiwan (National Health Research Institute. National Health Insurance Research Database, Background. Accessed at http://www.nhri.org.tw/nhird/en/index.htm on 23 September 2011). There was no significant difference in distribution of the age, gender, annual births, and average premium of beneficiaries, using chi-square tests at alpha level 0.05, between the patients in the sampled databases and the original full population of the NHIRD.

Study design and cohort assembly

From the database population of 2 million, we identified a cohort of adult patients (18 years or older) who were diagnosed between 2001 and 2008 with epilepsy as defined by the International Classification of Diseases, Ninth Revision, Clinical Modification (ICD-9-CM) code 345.xx and who were newly treated with monotherapy of PHT, VPA, or CBZ. These three AEDs were selected as targeted agents because under the NHI program, AEDs including PHT, VPA, and CBZ were reimbursed as first-line monotherapy for epilepsy and these three drugs were AEDs with most prevalent use for patients with epilepsy in Taiwan (Hsieh & Huang, 2011). We retrieved all the AED prescriptions using the Anatomical Therapeutic Chemical (ATC) classification system (WHO Collaborating Centre for Drug Statistics Methodology. International Language for Drug Utilization Research ATC/DDD. Accessed at http://www.whocc.no/ on 23 September 2011). Because this study was exploratory in nature, prior knowledge of event rate of various drugs that was required for computing pre-planned sample size was not available. Instead, as a population-based study, we included all eligible patients in the study and computed event rate accordingly.

The index date was defined as the first date with prescription of the targeted AED, and the baseline period was defined as 1 year prior to the index date. AED users who had not received a prescription of any AEDs (ATC code N03A, not limited to the three targeted AEDs; Table S2) in the baseline period were considered new users. To ensure that the AED was prescribed for antiepileptic treatment, we selected patients who had been diagnosed with epilepsy 180 days before or after the index date and who had received the targeted AED for ≥30 days. Patients with prior stroke (ICD-9-CM code 430-438) at the baseline period were excluded. Also excluded were patients who incurred a stoke event within 30 days after the index date, considering that such a short-term exposure to AEDs was less likely to be the cause of a stroke. Also excluded were patients with any history of several rare conditions (Lin et al., 2008; Kang et al., 2009, 2010, 2011; Hu et al., 2010; Sheu et al., 2010a,b; Yang et al., 2011) that have been reported to be associated with higher risk for stroke.

Definition of events and covariates

All patients were followed until stroke occurrence, any switch from the original AED to another AED, the addition of another AED, the discontinuation of the AED regimen (defined as no AED prescription for >30 days after the end date of the previous prescription), disenrollment from the NHI for any reason, the end date of the database (December 31, 2008), or 5 years after the index event, whichever came first. The primary event for this study was any kind of stroke, which was determined by a diagnosis code of ICD-9-CM 430-438 listed on the claims of emergency services or hospitalizations. In addition, subtypes of stroke including hemorrhagic stroke (defined by ICD-9-CM codes 430-432), ischemic stroke (433, 434, and 436), and others (435, 437, and 438) were also evaluated respectively in the sub-analyses. The accuracy of the NHIRD in cases with diagnosis of ischemic stroke has been well validated (Cheng et al., 2011).

More than 70 potentially significant confounding covariates (Table S1) were studied to assess the association between AED and stroke, which included patient demographics, comorbidities, and concomitant medications based on previously published literature (Mehta et al., 2010). We retrieved all the ICD-9-CM codes for comorbidities within the 1-year baseline period as well as any prescriptions within 180 days before the index date as concomitant medications. The defined daily dose (DDD) and prescribed daily dose (PDD) were computed for the targeted AEDs. The DDD assignment was based on dose information obtained from the World Health Organization Collaborating Center, and the PDD was calculated from prescription data of hospital visits. The PDD/DDD ratio of an AED thus indicates the relative dose of a drug as compared to what has been recommended, and the ratio may reflect the severity of epilepsy. The premium each patient paid for the NHI was used as a proxy of income, because the premium is determined by wage level under the NHI program.

Main analysis

Hazard ratio (HR) and 95% confidence intervals (CIs) obtained from Cox proportional regression models were computed to evaluate the relative stroke risk of AEDs. The final regression models were adjusted for patient demographics (age, gender, year of index date, relative dosage, and premium levels), and potential confounding covariates were determined by stepwise selection at an alpha level of 0.1. CBZ was chosen as the reference because it is the most frequently used AED worldwide (Mintzer & Mattson, 2009; Olesen et al., 2011a).

Sensitivity analyses

To test the robustness of this study, we performed additional analyses by matching the targeted AED treatment groups by propensity score (PS) to avoid selection bias. The PS technique calculates, using logistic regression models, the probability for a patient to receive one treatment versus other treatments. All of the covariates (excluding relative dosage) were included in the calculation of PS. Patients were then matched by PS with the Greedy 5[RIGHTWARDS ARROW]1 digit matching method, which has been shown to reduce bias caused by incomplete and inexact matching (Parson LS. Reducing bias in a propensity score matched-pair sample using greedy matching techniques. Accessed at http://www2.sas.com/proceedings/sugi26/p214-26.pdf on 8 November 2011). Using the PS from two models, patients receiving CBZ were matched with those using PHT or VPA, respectively; these two matched subcohorts, CBZ-PHT and CBZ-VPA, were further analyzed.

In the main analysis, to avoid potential misclassification bias, we censored patients whose AED regimen changed in the follow-up period; however, since the AED regimen often changed for patients with epilepsy that would underestimate the stroke risk. To evaluate the impact due to censoring at AED switching, two additional analyses were performed where patients were defined by the intention-to-treat (ITT) method. In ITT Analysis 1, First AED Model, patients were classified according to the initial AED they received, regardless of subsequent usage patterns. Patients in ITT Analysis 2, Major AED Model, were classified according to the “major” AED received. The major AED was defined as the AED used for >50% of all drug-supplied days in the followed-up period. Patients were excluded if the major AEDs were not the targeted agents.

Subgroup analysis by excluding those who used an injection formulation at the index date was performed to evaluate potential confounding by disease severity. The exclusion decision was based on the assumption that those who used an injection formulation as the first AED may have more severe epilepsy and therefore be at higher risk of stroke. The last subgroup analyses were carried out by stratifying patients by age (elderly, >65; young, 18–65), gender, and duration (<90; 90–360; 360–720; >720 days) and relative dosage (PDD/DDD ratio ≧ 0.5; <0.5) of AEDs. All statistical analyses were performed using SAS 9.2 version (SAS Institute, Cary, NC, U.S.A.) software.

Results

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. Disclosure
  9. References
  10. Supporting Information

Cohort assembly and matching process

Figure 1 shows the cohort assembly and study flowchart for the three targeted AEDs. From 2001 to 2008, there were 4,065 new AED users who were diagnosed with epilepsy and had received one of the targeted AED monotherapies for >30 days. After extracting for exclusion criteria, a total of 2,874 patients remained, of which 1,957 (68%) were PHT, 524 (18%) were CBZ, and 393 (14%) were VPA users. The baseline characteristics of the patients in the three groups are presented in Table 1.

image

Figure 1.   Study cohort assembly flowchart.

Download figure to PowerPoint

Table 1.   Baseline characteristics of antiepileptic drug users
 Carbamazepine (N = 524)Phenytoin (N = 1,957)Valproic acid (N = 393)p-Value*
  1. SD, standard deviation; NHI, national health insurance; NT, new Taiwan dollar; AED, antiepileptic drug; DDD, defined daily dose; PDD, prescribed daily dose; COPD, chronic obstructive pulmonary disease; NSAIDs, nonsteroidal antiinflammatory drugs; CIs, confidence intervals.

  2. *p-values were obtained from analyses of variance (ANOVAs) for continuous variables and Chi-square for categorized variables.

  3. a1 N.T. dollar is approximately equal to 0.034 U.S. dollar.

  4. bListing five major conditions respectively. More information on comorbid conditions and concomitant medications is presented in Tables S1 and S2.

  5. cOdds ratios were obtained from logistic regression.

  6. dHazard ratios were obtained from Cox proportional hazard regression.

Demographics    
 Age, mean (SD), in year47.75 (18.86)51.11 (19.78)47.03 (19.58)<0.01
 Female, n (%)258 (49.24)714 (36.48)181 (46.06)<0.01
 Patient distribution by index year, n (%), in year    
  2001103 (19.66)279 (14.26)36 (9.16)<0.01
  200291 (17.37)249 (12.72)41 (10.43)
  200377 (14.69)259 (13.23)44 (11.20)
  200463 (12.02)260 (13.29)59 (15.01)
  200549 (9.35)315 (16.10)52 (13.23)
  200651 (9.73)229 (11.70)42 (10.69)
  200761 (11.64)201 (10.27)56 (14.25)
  200829 (5.53)165 (8.43)63 (16.03)
 Patient distribution by NHI premium levels, n (%), NT$a    
  <15,000132 (25.19)545 (27.85)105 (26.72)0.35
  15,000–25,000314 (59.92)1182 (60.40)240 (61.07)
  >25,00078 (14.89)230 (11.75)48 (12.21)
 Relative dosages of AEDs, mean (SD), PDD/DDD ratio0.27 (±0.29)0.60 (±0.40)0.36 (±0.30)<0.01
Comorbid conditions, n (%)b    
 Psychosis147 (28.05)436 (22.28)103 (26.21)<0.01
 Hypertension111 (21.18)459 (23.45)79 (20.10)0.24
 Gastric disorders116 (22.14)396 (20.24)70 (17.81)0.27
 COPD57 (10.88)267 (13.64)38 (9.67)0.04
 Diabetes mellitus57 (10.88)252 (12.88)37 (9.41)0.10
Concomitant medications, n (%)b    
 NSAIDs365 (69.66)1140 (58.25)235 (59.80)<0.01
 Antibacterial agents272 (51.91)1040 (53.14)195 (49.62)0.43
 Benzodiazepines260 (49.62)916 (46.81)160 (40.71)0.25
 Antihistamines274 (52.29)857 (43.79)201 (51.15)<0.01
 Bowel medications165 (31.49)595 (30.40)115 (29.26)0.77
Risk of stroke    
 Number of stroke patients, n (incident rate, per 1,000 persons)37 (71)243 (124)29 (73)
 Time to events, mean (SD), in day365 (485)256 (388)268 (383)
 Univariate analysis    
  Odds ratio (95% CIs)c1.001.87 (1.30–2.68)1.05 (0.63–1.74)
  Hazard ratio (95% CIs)d1.002.36 (1.67–3.34)1.32 (0.81–2.14)

Comparative stroke risk

A higher stroke incident rate (IR) was found in patients receiving PHT and VPA (124 and 73 per 1,000 persons, respectively), with odds ratios of 1.86 (95% CI 1.30–2.68) and 1.05 (95% CI 0.63–1.74), respectively, compared with CBZ (IR of 71 per 1,000 persons). Table 2 presents a multivariate Cox proportional hazards regression model for comparative risk of stroke in AED users. Patients receiving PHT had a significantly greater stroke risk than CBZ (adjusted HR 1.72; 95% CI 1.20–2.47), whereas there was no significant difference between VPA and CBZ (adjusted HR 1.27; 95% CI 0.78–2.07). The stroke risk was increased by 3% per year of increased age and by 70% per unit increase in the PDD/DDD ratio of AEDs. Greater risks were also found in male patients and those with diabetes mellitus or who had used renin-angiotensin system inhibitors or lipid-lowing agents.

Table 2.   Multivariate Cox proportional hazards regression model for risk of stroke in antiepileptic drug users
 Hazard ratios (95% confidence intervals)p-Value
  1. AED, antiepileptic drugs; NHI, national health insurance; NT, new Taiwan dollar; DDD, defined daily dose; PDD, prescribed daily dose.

  2. aSignificant at alpha level of 0.05.

  3. bVariables determined by stepwise selection at alpha level of 0.1.

Targeted AEDs  
 Carbamazepine1.00
 Phenytoin1.72 (1.20–2.47)a<0.01
 Valproic acid1.27 (0.78–2.07)0.32
Demographics  
 Age, per year1.03 (1.03–1.04)a<0.01
 Relative dosage, per PDD/DDD ratio1.70 (1.26–2.28)a<0.01
 Gender  
  Female1.00
  Male1.20 (1.01–1.51)a0.05
 NHI premium levels, NT$  
  <15,0001.00
  15,000–25,0000.95 (0.74–1.21)0.68
  >25,0000.93 (0.58–1.50)0.77
 Index year  
  20011.00
  20021.12 (0.75–1.68)0.50
  20031.15 (0.76–1.75)0.67
  20041.09 (0.72–1.67)0.99
  20050.99 (0.66–1.51)0.84
  20061.05 (0.66–1.65)0.64
  20071.12 (0.69–1.83)0.74
  20080.90 (0.48–1.68)0.58
Selected comorbid conditions or concomitant medicationsb  
 Diabetes mellitus1.41 (1.06–1.87)a0.02
 Coronary heart disease0.72 (0.48–1.04)0.07
 Bowel medications1.26 (0.99–1.60)0.07
 Renin-angiotensin system inhibitors1.53 (1.13–2.07)a<0.01
 Lipid-lowing agents1.75 (1.15–2.67)a0.01
 Antihemorrhagic agents1.35 (0.96–1.90)0.07

Subanalyses and sensitivity analyses

Table 3 presents several analyses for specific subcohorts. The use of PHT was associated with significantly higher hemorrhagic (adjusted HR 2.00; 95% CI 1.25–3.18) and ischemic (adjusted HR, 6.51; 95% CI, 2.03–20.88) stroke compared to CBZ. There was no significant difference between VPA users and CBZ users in any subtypes of stroke. After PS matching, there were 1,024 (PS 0.66 ± 0.26 standard deviation [SD] in each AED) and 644 (PS 0.44 ± 0.14 SD in each AED) matched patients in the CBZ-PHT and CBZ-VPA subcohorts, respectively. A similar trend was found in PS-matched cohorts, with a higher stroke risk for those who received PHT (adjusted HR 2.05; 95% CI 1.36–3.09), as well as in the first-used models (adjusted HR 1.63; 95% CI 1.18–2.23) and major-used models (adjusted HR 2.23; 95% CI 1.24–4.03) of ITT analysis. The trends of stratification analyses were similar to the main analysis. The risk was higher in patients with longer durations and higher relative dosage of PHT.

Table 3.   Subanalyses and sensitivity analyses of stroke risk in PHT and VPA groups, with CBZ group as reference
Targeted AEDsCarbamazepinePhenytoinValproic acid
Ns/NtNs/NtAHR (95% CIs)Ns/NtAHR (95% CIs)
  1. AED, antiepileptic drugs; AHR, adjusted hazard ratio; CBZ, carbamazepine; PHT, phenytoin; VPA, valproic acid; Ns/Nt, Number of patients who had stroke/Number of all patients in the subcohort in each AED; DDD, defined daily dose; PDD, prescribed daily dose.

  2. aAdjusted by patient demographics (age, gender, year of index date, and NHI premium levels), and variable determined by stepwise selection at alpha level of 0.1.

  3. bPropensity score was derived from patient demographics (age, gender, year of index date, and premium of NHI), comorbid conditions, and concomitant medications by using logistic regression.

  4. cAdjusted for age, gender, relative dosage, index year, NHI premium levels, diabetes mellitus, coronary heart disease, bowel medications, renin angiotensin system inhibitors, lipid-lowing agents, and antihemorrhagic agents.

  5. dIn first AED used model, patients were classified according to the AED they were exposed to initially, regardless of subsequent usage patterns; in major AED used model, patients were classified according to the major AED based on drug supplied days.

Subtype of strokea     
 Hemorrhage type21/508131/18452.00 (1.25–3.18)14/3781.27 (0.64–2.50)
 Ischemia type3/49052/17666.51 (2.03–20.88)3/3671.77 (0.36–8.79)
 Other type13/50060/17741.34 (0.73–2.46)12/3761.59 (0.72–3.49)
Propensity score matching analysesb     
 CBZ-PHT matched35/51270/5122.13 (1.41–3.20)–/–
 CBZ-VPA matched18/322–/–25/3221.54 (0.84–2.83)
Intention to treat analysesc,d     
 First AED used48/524198/19571.63 (1.18–2.23)30/3931.20 (0.76–1.89)
 Major AED used27/350147/14502.23 (1.24–4.03)21/2691.89 (0.96–3.68)
Stratification analysesc     
 Duration of AEDs used, days     
  <9010/204100/10151.48 (0.88–2.47)13/1910.85 (0.39–1.83)
  90–36019/17187/5352.08 (1.35–4.97)10/1181.68 (0.73–3.83)
  360–7206/7124/2003.41 (1.03–6.07)3/431.76 (0.42–7.36)
  >7202/7832/2076.68 (1.58–28.27)3/413.98 (0.66–24.05)
 Relative dosage of AEDs used, PDD/DDD ratio     
  <0.532/41393/6881.72 (1.14–2.58)21/2581.58 (0.91–2.75)
  ≧0.55/111150/12692.74 (1.11–6.77)8/1381.45 (0.47–4.46)
 Age of AED users, year     
  18–6517/391127/13832.51 (1.51–4.18)18/3081.76 (0.90–3.42)
  >6520/133114/5741.63 (1.00–2.65)11/851.19 (0.56–2.50)
 Gender     
  Male16/266162/12432.60 (1.55–4.37)20/2122.19 (1.13–4.23)
  Female21/25881/7141.36 (0.83–2.21)9/1810.82 (0.38–1.80)
Exclude patient receiving injection formulation of AED at index date37/524123/9802.08 (1.44–3.01)28/3831.29 (0.79–2.11)

Discussion

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. Disclosure
  9. References
  10. Supporting Information

The incidence of stroke was 3.5 per 1,000 persons among the general population in Taiwan (Hsieh et al., 2010). A much higher incidence of stroke was found in the present study as compared to that among the general population, which suggested that patients with epilepsy are at higher risk of stroke events, especially those who are receiving PHT. After adjusting for potential confounders by regression models, we found that exposure to PHT was associated with a higher stroke risk as compared to CBZ, whereas there was no significant difference in the stroke risk between VPA and CBZ users. Similar to the previous report (Pinto et al., 2004), our findings indicated that patients who were male, of advanced age, or with a comorbidity of diabetes had a greater risk of stroke. In addition, a longer duration and higher dosage of PHT demonstrated a dose–response relationship to increased stroke risk. The results of several subanalyses were consistent with the main analysis.

Several mechanisms had been proposed to explain the association between AED exposure and cardiovascular risk. Exposure to PHT or CBZ was associated with decreased serum folate and increased serum homocysteine, thereby increasing the risk of stroke in patients with epilepsy (Mintzer et al., 2009; Linnebank et al., 2011). Patients taking CYP enzyme–inducing AEDs, such as PHT or CBZ, had increased blood concentration of cholesterol and the atherogenic lipid fraction LDL-C (Nikolaos et al., 2004; Tomoum et al., 2008; Mintzer et al., 2009), which may further increase the risk of stroke. Aside from the metabolism of xenobiotics, families of human CYP enzyme were also important in the synthesis of endogenous cholesterol (Mintzer & Mattson, 2009; Mintzer et al., 2009; Mintzer, 2010). The induction of CYP51A1 enzyme would increase the synthesis of cholesterol through metabolism of the oxysterol intermediate and reduce the feedback inhibition of hydrox-ymethylglutaryl-coenzyme A (HMG-CoA) reductase (Mintzer & Mattson, 2009). However, one study found different effects of PHT and CBZ on serum lipid profiles of patients with epilepsy, in that the former increased only the LDL-C level, whereas the latter increased levels of both LDL-C and high-density lipoprotein cholesterol (HDL-C), a component of nonatherogenic lipoprotein cholesterol (Nikolaos et al., 2004). The different effects on HDL-C profiles may explain why PHT had a higher stroke risk than CBZ in our study.

The duration of exposure to AED had been reported to predict the increase of carotid IMT, an effect that was independent of age, gender, and total oxidative stress (Tan et al., 2009). Chuang et al. had reported the different levels of increase in carotid IMT after PHT and CBZ monotherapy when compared with a healthy control (Chuang et al., 2012). We found PHT users had an increased risk of both ischemic and hemorrhagic subtypes of stroke, which corresponds to the results by Chuang et al. because carotid IMT is known to be associated with both ischemic and hemorrhagic stroke (Ohira et al., 2011).

The comparative risk of stroke after exposure to VPA remained controversial. Using time-dependent exposure covariate analysis, Olesen et al. had demonstrated that, compared with CBZ, VPA may attenuate the risk of myocardial infarction (Olesen et al., 2011b) as well as the risk of stroke (Olesen et al., 2011a) in patients with epilepsy. As opposed to enzyme inducers CBZ and PHT, blood levels of total cholesterol and LDL-C may not increase with VPA use (Nikolaos et al., 2004). Moreover, it had been reported that VPA may have a protective effect on the vascular system through mechanisms of ventricular remodeling (Lee et al., 2007) and increases in resistance to endoplasmic reticulum stress, which could inhibit atherosclerosis (Bowes et al., 2009; Khan et al., 2009). But the use of an enzyme-inhibiting AED like VPA possibly could also accelerate atherosclerosis (Tan et al., 2009) by mechanisms such as insulin resistance, microcapillary effect, and metabolic syndrome (Chengappa et al., 2002; Martin et al., 2009). We did not observe a significant difference in stroke risk between VPA and CBZ, although most of the point estimates of the analyses indicated VPA had a higher risk. The inconsistency of the comparative risk of stroke in VPA among studies may be due to different study designs and/or the different races of the study populations. Further exploration for other mechanisms is needed, which could provide important information in preventing stroke.

Consistent with the main analyses (Table 2), stratification analyses by AED duration or relative dosage implied that AED-associated stroke may be a slow, chronic, and dose-cumulative process. This could also be explained by the aforementioned mechanisms such as effects on the CYP system or serum lipid levels. However, it is worth noting that the stratification analyses were designed to eliminate the potential confounders. All the results were generated with reference to CBZ; therefore, overinterpretation that there is any relationship among different subgroups should be avoided. Further investigation will be needed to confirm the relationship between stroke risk and PHT in the female population.

Strengths

This study employed a large nationwide random sample that represented well the entire population of Taiwan. AEDs were reimbursed under the Taiwan NHI system, and thus all use for antiepileptic purposes was recorded in NHIRD. The diagnosis records of emergency-service and hospitalization claims in NHIRD are valid; ischemic stroke also has been validated in by Cheng et al., (2011). To generate a more homogenous group of patients for comparisons, we performed a new user-design and selected specific patients. In addition, many potential confounders, such as patient demographics, comorbid conditions, and concomitant medications, were adjusted in our study, and the results remained consistent throughout the series of adjusted processes and sensitivity analyses.

Limitations

The unavoidable limitation in this study was an inability to confirm whether the patients actually took their dispensed medicines; thus we possibly overestimated the stroke risk induced by AED use. This study was also limited by unavailable information including patients’ body weight and lifestyle, which are well-known risk factors for stroke. Possible confounding by indication may exist, since physicians may avoid VPA for patients with epilepsy and obesity. Because epilepsy without AED treatment has also been reported to be associated with a higher risk of stroke (Olesen et al., 2011a), the present study could not estimate whether the increase of stroke risk is caused by epilepsy or AEDs. However, using CBZ-treated patients with epilepsy as an active control, we did demonstrate that PHT users had a significantly higher stroke risk than CBZ users. Although many potential confounders were adjusted or stratified in our study and the propensity score method has been employed, we noted that those on PHT were older, more likely to be male, and had more comorbidities, which could potentially increase the risk of stroke for this group of patients. Finally, it has been reported that up to 30% of patients could be misdiagnosed with epilepsy (Zaidi et al., 2000). Epilepsy defined by one encounter coded with 345.xx may result in higher sensitivity of case ascertainment but with lower specificity. Although we have selected patients based both on ICD-9 code and a ≥30 days use of a targeted AED to increase the specificity of epilepsy, it is still possible that some patients selected might have only used the AEDs for a short-term treatment, such as those with a single seizure. On the other hand, the exclusion of individuals who used an AED for nonepileptic purposes, as well as those who had discontinued, switched, or added another type of AED within 30 days after index date, may result in a potential immortal time bias. We believe the bias was limited, however, because the proportions of switched or added-on therapies were similar among the three groups. At the same time, we identified a more homogenous population with longer exposure durations to assess the comparative stroke risk of each targeted AED.

Conclusion

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. Disclosure
  9. References
  10. Supporting Information

In conclusion, benefits of seizure control should be weighed against the risk of stroke in the long run. It is important for physicians to know about the comparative risk related to the specific AEDs in patients with epilepsy. The study result provides information on the comparative stroke risk of CBZ, VPA, and PHT and suggests avoiding the use of PHT for patients who already have a higher risk of stroke.

Acknowledgments

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. Disclosure
  9. References
  10. Supporting Information

This study was supported in part by grants from the Multidisciplinary Center of Excellence for Clinical Trial and Research (DOH100-TD-B-111-002), Department of Health, Executive Yuan, Taiwan. The authors thank Dr. Ching-Lan Cheng and Dr. Chih-Hung Chen for their helpful review and suggestions for the manuscript.

Disclosure

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. Disclosure
  9. References
  10. Supporting Information

None of the authors has any conflict of interest to disclose. We confirm that we have read the Journal’s position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

References

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. Disclosure
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Summary
  3. Methods
  4. Results
  5. Discussion
  6. Conclusion
  7. Acknowledgments
  8. Disclosure
  9. References
  10. Supporting Information

Table S1. Baseline characteristics of antiepileptic drug users.

Table S2. Antiepileptic drugs available in Taiwan.

FilenameFormatSizeDescription
epi3693_sm_SuppInfoS1-S2.doc180KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.