These authors contributed equally to the manuscript.
Address correspondence to Dr. Horng-Huei Liou, Department of Neurology, College of Medicine, National Taiwan University. No 7, Sec 1, Chung Shan S Rd., Taipei 100, Taiwan. E-mail: email@example.com or Dr. Hsiu-Hsi Chen, Division of Biostatistics, Institute of Epidemiology and Preventive Medicine, College of Public Health, National Taiwan University. No 17, Xuchow Rd., Taipei 100, Taiwan. E-mail: firstname.lastname@example.org
Purpose: We studied geographic variation in age- and gender-specific prevalence and incidence of epilepsy in four different areas of Taiwan.
Methods: By using large-scale, National Health Insurance (NHI)–based data from 2000–2003 in Taiwan, we identified 131,287 patients diagnosed with epilepsy (ICD code 345) receiving at least of one of 11 antiepileptic drugs (AEDs). Information on age, gender, and location were also collected. The multivariable Poisson regression analysis was used to assess the heterogeneity of the morbidity of epilepsy in different regions. External data validation was also performed to assess the accuracy of capturing epilepsy cases through our NHI data set.
Key Findings: The age-adjusted prevalence and incidence of epilepsy were 5.85 (per 1,000) between 2000 and 2003 and 97 (per 100,000 person-years) during the follow-up time from 2001 to 2003 in Taiwan. The sensitivity and specificity of ICD-9 coding for epilepsy in the NHI data set were 83.91% and 99.83%, respectively, resulting in a slight overestimation. Male patients had a higher probability of having epilepsy than did females. East Taiwan had significantly higher prevalence and incidence than did other areas. The age-specific incidence pattern in east Taiwan was atypical in that it revealed clustering in young and middle-aged groups.
Significance: Our study demonstrated geographic variation in epidemiologic patterns of epilepsy within Taiwan. The findings are informative and provide insight into the clinical management of epilepsy based on consideration of different target groups in different areas.
The prevalence of epilepsy varies worldwide (Lavados et al., 1992; Rwiza et al., 1992; Banerjee & Hauser, 1997; ILAE, 1997; Tekle-Haimanot et al., 1997; Forsgren et al., 2005), suggesting that risk in different geographic areas may be influenced by different factors. An improved understanding of the geographic variation of epilepsy would make a significant contribution to understanding the etiology of the condition. In countries with low economic development, the peak incidence is in young and middle-aged adults, and may be attributed to endemic infectious or parasitic diseases (Banerjee & Hauser, 1997; Carpio et al., 1998) arising due to poor medical facilities and general health. Despite these findings, very few studies have been conducted to simultaneously investigate geographic variation in the incidence and prevalence of epilepsy. Incidence is a fundamental measurement related to the etiology of the disease, whereas prevalence better reflects disease burden and is useful for the planning of the provision of health care services.
It is reasonable to assume that epilepsy patterns differ in different locations and may be affected by a variety of factors. However, opportunities to assess the heterogeneity of epidemiologic patterns of epilepsy are rare. Taiwan is divided into four major geographic areas (East, North, South, and Central), each with distinctive environmental and socioeconomic conditions (Fig. 1). In East Taiwan, the rift valley presents a unique geographic landscape not seen in other areas. East Taiwan also has the lowest economic development status and is home to a mixture of racial populations. North Taiwan has the highest population density and economic status. South Taiwan has large plains and a warmer climate, facilitating the development of agriculture. Conditions in the Central region of Taiwan are a blend of these areas. Therefore, the four distinct areas within Taiwan present a unique opportunity to compare the incidence and prevalence of epilepsy across four very different geographic regions.
Previously, there have been several nationwide studies in Western countries with emphasis on prescribing pattern of antiepileptic drugs (AEDs) (Savica et al., 2007; Hollingworth & Eadie, 2010) and the effect of epilepsy on health status in different age groups (Pugh et al., 2005). Nonetheless, because such population surveys are costly, few researchers have used large-scale, national data to examine epidemiologic patterns in the incidence and prevalence of epilepsy or elucidate prescribing patterns for AEDs in Asian countries. In Taiwan, the National Health Insurance (NHI) system was initiated in 1995 and covers >95% of the total population (Wen et al., 2008). This high coverage rate means that the NHI database is a comprehensive resource that allows access to large-scale epidemiologic information. In this report, we used International Classification of Disease (ICD) and prescription codes to identify epilepsy patients from the NHI database and thereby estimate the prevalence and incidence of epilepsy in Taiwan. Variations in the prevalence of epilepsy across the four geographic areas provided the reference for the administration and clinical management of patients diagnosed with epilepsy. Geographic variations in incidence were explored to identify vulnerable populations and identify risk factors for epilepsy in Taiwan.
Computerized data from medical claims for patients with epilepsy who visited health service locations during the period from 2000 to 2003 were retrieved from the NHI database. The NHI database records individual diagnostic information, usually provided by neurologists, as in Taiwan most patients with epilepsy are diagnosed by these specialized physicians. The NHI database also includes details of pharmaceutical claims for prescriptions of the 11 AEDs approved for patients with epilepsy by the Taiwan National Health Insurance Bureau: specifically carbamazepine, phenytoin, gabapentin, levetiracetam, oxcarbazepine, primidone, tiagabine, topiramate, sodium valproate, vigabatrin, and phenobarbital. These active agents are available in 150 commercial forms.
Definition of epilepsy
Individuals with an episode of disease coded 345.xx (i.e., epilepsy, as per the ninth revision of International Classification of Diseases, ICD) and mention of at least one of AEDs listed above were classified as having epilepsy. Following Pugh et al. (2005), those who did not meet these criteria were classified as not having epilepsy. It should be noted that each individual record for a medical claim in the NHI data set may consist of up to three ICD codes (corresponding to the primary, secondary, and tertiary causes). Therefore, a patient with epilepsy and comorbid disease might have their epilepsy recorded in any one of the three places. The main reason for the claim is recorded by the physician as the primary cause and any comorbidities are recorded as the secondary and tertiary causes. Therefore, for epilepsy patients with comorbid conditions, clinical consultations relating to each condition would have been recorded in the NHI data set independently. To enhance the accuracy of identifying patients with epilepsy, the validation was further done by checking the presence of one of three ICD codes equal to ICD-9 code 345 at least twice in NHI data set.
To validate the accuracy of our procedure for identifying epilepsy cases through the NHI database, an external validation was carried out by linking data from a 2001 community-based survey on epilepsy prevalence in Keelung city, Taiwan (regarded as the gold standard) with that we obtained from the NHI data set relating to the period 2001–2003. The details of this study have been described in full elsewhere (Chen et al., 2006). In brief, a total of 13,663 patients aged 30 years or older were invited to participate in a community-based neuroepidemiologic survey to estimate the prevalence of epilepsy during the year 2001. Fifty-two epilepsy cases (including active epilepsy and epilepsy in remission) were detected by neurologists using a one-stage method. First, we traced these 13,663 patients in the population registry to determine migration and death status. Those still alive and living in Keelung comprise the data set for cross-checking to the NHI data set.
Study design for estimating prevalence and incidence of epilepsy
The NHI data provide a unique opportunity to estimate the national prevalence and incidence of epilepsy. The definition of prevalence and incidence of epilepsy we used is similar (but not identical) to those used in two previous studies (Pugh et al., 2005; Savica et al., 2007). Prevalent epileptic cases were individuals with chronic epilepsy identified from NHI data set meeting the diagnostic criteria and pharmacy data criteria as mentioned above similar to those defined in a prior study (Pugh et al., 2005). It should be noted that because our National Health Insurance program has such high coverage, up to 95% according to one previous study (Wen et al., 2008), and most patients with chronic epilepsy (95%) would visit their neurologists on an annual basis, it is possible to estimate the prevalence of epilepsy in a given year from these data. The NHI’s benefits are comprehensive and cover emergency, outpatient and inpatient care, laboratory tests, diagnostic imaging, and prescription charges, with patients paying an average of 1,500 New Taiwan dollars (NTD) (1 USD: 34 NTD) in insurance premiums every month. For this reason most patients, particularly those with chronic diseases, would have regular visits to clinics, as demonstrated by the consistent annual prevalence figures (see below). However, we report here not only annual prevalence estimates but estimates for the period 2000–2003. Fig. 2 shows the study design for determining the annual and period prevalence of chronic epilepsy and incidence of newly diagnosed cases.
To estimate the incidence of epilepsy, we identified a normal cohort (e.g., persons free of epilepsy) by excluding those with preexisting epilepsy in the year 2000. We then followed this normal cohort over a 3-year period (from 2001 until 2003) to identify newly diagnosed (incident) cases using a definition for new-onset epilepsy similar to that used in a prior study (Pugh et al., 2005).
We also collected data on age, gender, location, date of clinic visit, and date of diagnosis. Age was grouped into 13 bands (<20 years, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70–74, and >75 years). Estimates of average annual family income, by county, were obtained from the Department of Statistics, Ministry of the Interior, Taiwan, and used as a surrogate for socioeconomic status (SES) in the 25 counties of Taiwan. The counties were grouped into four geographic areas, each with its own distinctive profile with respect to population density, proportion of indigenous people, size of mountainous areas, and number of medical personnel (as shown in Fig. 1). SES was measured on a continuous scale.
The prevalence of epilepsy during the period 2000–2003 for each gender, geographic area, and age group was calculated as the number of cases of epilepsy divided by the total number of people. We also estimated the annual prevalence, as defined above. As illustrated in Fig. 2, to estimate the incidence of epilepsy we began with a cohort that was epilepsy-free at entry to the study (2000) and followed them up between 2001 and the end of 2003 in order to identify newly diagnosed cases and compute person-years for the underlying population at risk. The independent effect of age and geographic factors on the incidence of epilepsy was assessed using a multivariable Poisson regression analysis, that is, treating epilepsy as a rare event. p-values <0.05 were considered statistically significant. To assess how sensitivity and specificity effected our estimates we used the following formula, with the true prevalence taken to be that observed in the Keelung community-based survey (i.e., the gold standard, P) and our estimate from the NHI database denoted .
Data processing and statistical analysis were performed using SAS 9.1 software (SAS Institute, Cary, NC, U.S.A.).
Prevalence and incidence
Table 1 shows our estimates of the prevalence and incidence of epilepsy, by gender and age groups. In total, 131,287 people diagnosed with epilepsy were identified during the study period, giving an overall age-adjusted prevalence of 5.85/1,000 [95% confidence interval (CI) 5.82/1,000–5.88/1,000) for the period 2000 to 2003. The estimates of annual prevalence were stable, being 3.90/1,000 in 2000, 3.99/1,000 in 2001, 4.00/1,000 in 2002, and 3.99/1,000 in 2003. Prevalence was higher in male patients (6.77/1,000; 95% CI 6.72/1,000–6.82/1,000) than in female patients (5.05/1,000; 95% CI 5.01/1,000–5.09/1,000). The incidence between 2001 and 2003 was 97 per 100,000 person-years (95% CI 96/100,000–98/100,000), and again higher in male patients (112/100,000; 95% CI 111/100,000–113/100,000) than in female patients (82/100,000; 95% CI 81/100,000–83/100,000). The patterns of overall age-specific incidence and prevalence followed U-shaped trends, with the incidence rate (per 100,000 person years) being 100 in the youngest age group (0–19 years), decreasing gradually to 54 in the 30–34-year-old group, and then rising rapidly to 330 in the oldest group (>75 years).
Table 1. Gender- and age-specific prevalence and incidence (per 100,000) of epilepsy during the period 2001–2003 in Taiwan
Period prevalence (2000–2003) Number of cases (%)
Person years (2001–2003)
Age group (years)
To compare the incidence rates in the four geographic areas of Taiwan, we used the multivariable Poisson regression model (Table 2). We found East Taiwan to have a higher incidence rate than other areas (relative rate 1.43; 95% CI 1.10–1.77). The three other areas had comparable rates, with no significant difference among them. As regards SES, the average annual family income was the highest in the Northern region, followed (in order of size) by the Central, South, and East regions (Fig. 1). Low SES tended to be associated with a higher incidence rate, but this was not statistically significant. The pattern of age-specific incidence was different in East Taiwan compared to that in the three other areas (Fig. 3), for it did not follow the U-shape trend seen overall, and there were higher incidence rates among the young and middle-aged.
Table 2. Risk of epilepsy from the multivariable Poisson regression model
East Taiwan had a statistically significant higher risk of epilepsy after adjustment for age, gender, and social economic status (SES). Low SES showed a nonsignificant trend toward higher incidence.
Gender (male vs. female)
SES (average annual family income)
To validate our findings we compared them to the results of the Keelung survey for the same period. To ensure comparable population figures, we excluded 950 patients who had either died or migrated. Table 3 shows the cross-tabulation of the number of epilepsy cases we identified from the NHI data set versus the equivalent figures from the 2001 survey (Chen et al., 2006). The sensitivity and specificity of our method for identifying epilepsy cases through the NHI data set was 81.39% and 99.83%, respectively. Applying these to the formula in the statistics section gave a true prevalence for the period of 5.11/1,000, which is slightly lower than the 5.85/1,000 we estimated using the NHI data set.
Table 3. External validation: comparison of findings from the National Health Insurance database with results from a previous community-based survey on epilepsy in adults aged 30 years or older
Community-based survey (gold standard)
aWe excluded 9 epilepsy cases and 941 nonepilepsy patients due to migration or death. The original numbers of identified epilepsy cases and patients free of epilepsy from the community-based survey were 52 and 13,611, respectively (see Chen et al., 2006).
National Health Insurance (NHI)
In this national study, we have demonstrated geographic variation in the incidence and prevalence of epilepsy within Taiwan. East Taiwan had a significantly higher incidence and prevalence than the other areas under study, and this appears to be due to a greater number of cases among the young and middle-aged. This disparity persisted after adjustment for age, gender, and SES. Because the inclusion criteria and study duration were the same across all areas, it seems that the geographic variation in epilepsy observed in Taiwan does not stem from methodologic factors. East Taiwan can be viewed as a unique region, where the risk of developing epilepsy is higher than in the other parts of Taiwan.
Possible explanation for the higher risk of epilepsy in East Taiwan
The excess of epilepsy cases in East Taiwan is predominantly in the young and middle-aged, and it is known that cases in these age groups may have an etiology different from that of symptomatic epilepsy arising at other ages. Head injury (Li et al., 1985; Haerer et al., 1986; Tekle-Haimanot et al., 1997; Oun et al., 2003; Chen et al., 2006) and central nervous system (CNS) infections are well-established risk factors in these age groups (Hesdorffer, 1997). In Taiwan, traffic accidents are the leading cause of traumatic brain injury (Chiu et al., 1997). According to a 1999 government report from the Department of the Interior, the two counties with the highest rates of traffic accident are in East Taiwan: 4.13 per 100,000 mobile vehicles in Taitung County and 3.16 per 100,000 mobile vehicles in Hualian County (Directorate General of Budget, Accounting and Statistics, 1999). Overall, the traffic accident rate in East Taiwan was more than twice that in the other three areas (Fig. 4). Another important factor is CNS infections (Annegers et al., 1988). In Taiwan, Japanese encephalitis (JE) was once a prevalent disease. Cases of JE have been reported in all areas of Taiwan, but East Taiwan has had the highest incidence since 1966. According to one report, the incidence of JE in Taiwan peaked in 1967, and a large proportion of cases were children (Wu et al., 1999). Survivors of JE have a higher risk of developing epilepsy later in life and would now fall into the middle-aged group in our study. Further research is needed to confirm the role of these two factors.
Low SES is associated with many factors that are known to increase the risk of epilepsy, including cerebrovascular disease, head trauma, congenital malformations, and CNS infection (Hesdorffer, 1997). However, the relationship between SES itself and the development of epilepsy is not straightforward. Some published studies have established that SES is itself a risk factor for epilepsy (Heaney et al., 2002; Hesdorffer et al., 2005). However, others have failed to find an association (Forsgren & Nystrom, 1990). In our multivariable Poisson regression model, low SES was not significant (Table 2); however, other environmental factors may still play a role. In East Taiwan, the major geographic feature is the rift valley, which is a major barrier for transportation. Although the absolute number of medical personnel in East Taiwan was similar to that in other areas (Fig. 1), the relative number is low considering the large territory and various services that need to be provided. According to the 2002 Annual Statistics Report, the two largest areas served by a single medical institute were two counties in East Taiwan: 23.44 km2 in Taitung county and 17.94 km2 in Hualian county (National Statistics, 2002, Available at: http://22.214.171.124/pxweb/Dialog/statfile9.asp). Therefore, the natural geography of East Taiwan may also be a critical factor leading to the poorer management of conditions associated with increased risk of epilepsy and subsequently to a greater number of cases.
East Taiwan has the highest concentration of indigenous people (Fig. 1); therefore, genetic susceptibility may be an additional factor. However, indigenous lifestyles may also play a role, as they traditionally have a high percentage of heavy drinking and high alcohol intake is a well-known cause of traffic accidents, which may in turn lead to epilepsy. In addition, due to the direct toxicity of alcohol, some reports have shown that alcohol itself is a risk factor for seizure disorders or epilepsy (Ng et al., 1988; Leone et al., 1997). Further study in this special population is needed to clarify this issue.
Comparison with previous studies
In this study, we found the incidence rate of epilepsy in Taiwan during 2001–2003 to be 97 per 100,000 person years, which is higher than in most developed countries (Hauser et al., 1993; Olafsson et al., 1996, 2005; Banerjee & Hauser, 1997; Annegers et al., 1999; Christensen et al., 2007) and also higher than in Asian countries (Mac et al., 2007; Yemadje et al., 2011). The prevalence was 5.85/1,000 and in close agreement with that of previous studies (Hauser et al., 1991; Banerjee & Hauser, 1997; ILAE, 1997; Forsgren et al., 2005) in Western countries, but slightly higher than that in Singapore, India, and China (Mac et al., 2007; Yemadje et al., 2011) and lower than Latin America and sub-Saharan Africa (Yemadje et al., 2011). However, it should be noted that the methodology in our national survey was different from the door-to-door surveys adopted in these previous studies, and external data validation suggests that we have slightly overestimated the prevalence. Age-specific incidence patterns of epilepsy point to risk factors clustering in certain age groups. A U-shaped incidence distribution is usually reported in economically developed countries (Hauser et al., 1993; Olafsson et al., 1996, 2005; Banerjee & Hauser, 1997; Christensen et al., 2007). In contrast, in relatively poorly resourced countries, the peak incidence rates are often found in the young and middle-aged (Lavados et al., 1992; Rwiza et al., 1992; Tekle-Haimanot et al., 1997). In our study, it is interesting to note that the pattern of age-specific incidence in East Taiwan was similar to that reported in relatively poorly resourced countries, whereas the other three regions had patterns more akin to those observed in developed countries. This raises the question as to whether SES is an important determinant of the prevalence of epilepsy among the young and middle-aged. Many diseases increase the risk of symptomatic epilepsy in these age groups. It is, therefore, reasonable to suggest that better resourcing and policy making would reduce the level of these diseases, thereby leading to a reduction in the incidence of epilepsy. Longitudinal studies of epilepsy, with reference to the improvement of public resources, may help to answer this question.
In Taiwan, we previously studied the epidemiology of epilepsy in Keelung city using results from a community-based multiple screening program (Chen et al., 2006). The prevalence of active epilepsy among those >30 years old was 2.77/1,000, which is lower than that found in the present study. Because public attitudes toward patients with epilepsy in Asian countries may be more negative than those in Western countries (Caveness & Gallup, 1980; Lai et al., 1990; Jensen & Dam, 1992; Chung et al., 1995), it is possible that stigma leads people with epilepsy to underreport the condition to investigators during screening. Therefore, the current study may be a more accurate depiction of the number receiving medical treatment for epilepsy in Taiwan. Of course, with use of the NHI database we were not able to identify the new (asymptomatic) cases that would have been detected by screening (Chen et al., 2006) or the door-to-door surveys.
Because the data set used in this study is derived from NHI claims data, it could be argued that the ICD coding would be less accurate than that of a conventional neurologic survey. There are five justifications for using the NHI data set to estimate the prevalence and incidence of epilepsy and elucidate geographic variations. First, because neurologists are scattered in clinics across Taiwan as well as being based in medical centers or hospitals and our health insurance system encourages individuals with chronic epilepsy to have regular clinical visits to their neurologist, the disease is either managed directly by a neurologist or indirectly following referral by other physicians. Second, because there are up to three ICD codes independently recorded for each individual claim (including comorbidity), the accuracy of our method for identifying cases is enhanced. For example, for a patient with chronic epilepsy with DM and hypertension, the three ICD codes would be recorded as 345, 401, and 250 when the consultation was related to epilepsy and as 250, 401, and 345 when it was related to diabetes. To be identified as a patient with epilepsy, an individual had to make at least two claims where one of the three ICD codes was 345.xx. Third, the accuracy of our procedure for identifying cases was further enhanced by further requiring that the patient was being prescribed an AED. This may lead to the underestimation of the prevalence or incidence of epilepsy, but it means that false positives in this analysis are unlikely. Fourth, the accuracy of claim codes is regularly checked by audit committees formed of experts and located in each of six regions of Taiwan. Typically, each committee selects a random sample of medical records to review to verify the accuracy of claim data. Finally, to estimate the degree of measurement errors, a formal external validation was made by cross-checking primary data from a previous community-based survey with our NHI data set. After adjusting for sensitivity and specificity, the prevalence over the period was overestimated by ∼13%. However, because the validated sample focused only on one city and adults aged 30 years or older, further validation may be required.
This study was financially supported by the Department of Health, Executive Yuan (DOH95-TD-M-113-002).
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.