Epilepsy Research (2014) 108, 792—801
journal homepage: www.elsevier.com/locate/epilepsyres
Prevalence of epilepsy in rural Kansas Elizabeth Ablah a, Dale C. Hesdorffer b, Yi Liu c, Angelia M. Paschal d, Suzanne Hawley e, David Thurman f, W. Allen Hauser b,g,∗ , The Prevalence of Epilepsy in Rural Kansas Study Group1 a
Department of Preventive Medicine and Public Health, University of Kansas School of Medicine — Wichita, United States b Department of Epidemiology and GH Sergievsky Center, Columbia University, United States c Department of Biostatistics and GH Sergievsky Center, Columbia University, United States d Department of Health Science, University of Alabama, United States e Department of Public Health Sciences, Wichita State University, United States f Department of Neurology, Emory University, United States g Department of Neurology and GH Sergievsky Center, Columbia University, United States Received 18 January 2013; received in revised form 9 December 2013; accepted 16 January 2014 Available online 13 February 2014
KEYWORDS Epilepsy; Prevalence; Rural; Capture—recapture
Summary Purpose: To determine the prevalence of active epilepsy in two southeastern rural Kansas counties. Methods: Medical records were abstracted from the emergency rooms, out- and inpatient services and clinics of 9 hospitals, from 10 doctors’ offices, and 1 nursing home in and surrounding the two counties. Letters were mailed from hospitals and doctors’ offices to invite their potentially eligible patients to participate in an interview. Medical record information and the interview, when available, were used for the final determination of active epilepsy, seizure type, etiology, syndrome, age, and gender in consensus conferences. Prevalence of epilepsy was calculated, and capture—recapture methodology, which estimates prevalence based on what is known about the population, was employed to assess active epilepsy in the two counties. Results: This study identified 404 individuals with active prevalent epilepsy who visited at least one of the 20 facilities during the observation period. The overall prevalence of active epilepsy was 7.2 per 1000. The seizure type for 71.3% of prevalent cases was unknown; among the 76 cases with known and classifiable seizure type, 55.3% had focal with secondary generalized seizures. Among the 222 cases with classifiable etiology, 53.1% were idiopathic/cryptogenic. About 75% (n = 301) were captured at only one center, 72% (n = 75) of the remaining 103 patients were captured at two centers, and 28 patients were identified at three or more centers. The
∗ Corresponding author at: Department of Epidemiology and GH Sergievsky Center, Columbia University, 680 West 168 Street, New York, NY 10032, United States. Tel.: +1 914 760 3144; fax: +1 212 305 2426. E-mail address: [email protected] (W.A. Hauser). 1 See Appendix A for ‘The Prevalence of Epilepsy in Rural Kansas Study Group’ members.
0920-1211/$ — see front matter © 2014 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.eplepsyres.2014.01.001
Epilepsy prevalence
793 capture—recapture assessment yielded an estimation of 982 prevalent patients. The overall estimated prevalence of epilepsy in the two Kansas counties using capture—recapture was 17 per 1000. Conclusions: The crude prevalence of epilepsy, using medical record survey methods, was similar to, but on the high end, of other total population prevalence studies in the United States. The capture—recapture assessment suggested that epilepsy prevalence might be considerably higher than the crude prevalence. © 2014 Elsevier B.V. All rights reserved.
Background Epilepsy is among the most common of all neurologic conditions (U.S. Department of Health and Human Services, 2008). There are an estimated 50 million people with epilepsy in the world (WHO, 2012), with an estimated prevalence of active epilepsy of 4—7 per 1000 persons in developed countries (Sander, 2003) and 5—74 per 1000 persons in developing countries (Preux and Druet-Cabanc, 2005; Ngugi et al., 2010). In the United States, an estimated 2.2 million individuals have epilepsy (Institute of Medicine, 2012). As a spectrum disease with symptoms that may mimic other conditions, epilepsy can be especially difficult to diagnose, making an accurate census of the disease even more of a challenge. Beyond difficulties inherent in diagnosing the condition itself, the logistic realities of determining the prevalence of epilepsy in a given region serve as impediments to determining widespread and reliable prevalence estimates for epilepsy. Because of the time and expense of such efforts, rural and other at-risk populations are less likely to be canvassed in a manner to reliably determine prevalence (Asawavichienjinda et al., 2002). Despite the inherent difficulties in determining prevalence, there exists a large body of literature centered on epilepsy prevalence across the world. In the large number of studies (more than 5000 since 1965) (Banerjee et al., 2009) conducted on the epidemiology of epilepsy and the wide variety of methodologies used and areas studied, studies of urban communities have predominated (Benn et al., 2009; Kelvin et al., 2007). Rural communities in the United States remain under-studied for epilepsy prevalence. Although most studies focus on urban populations in the United States, epilepsy is estimated to be more prevalent in rural communities than in urban communities (Haerer et al., 1986; Hollingsworth, 1978; Baumann et al., 1977). Studies of rural populations worldwide have suggested epilepsy prevalence between 2.7 per 1000 in Italy (Reggio et al., 1996) and 17.6 per 1000 in Chile (Lavados et al., 1992). Prevalence varies dramatically by region and often corresponds to the countries’ degree of development or industrialization (Strzelczyk et al., 2008). Within countries, epilepsy can be as much as twice as common in rural areas than urban areas (Gourie-Devi et al., 2004; Aziz et al., 1997). Few rural, population-based prevalence studies have been conducted in the United States. Moreover, studies that have included racially and ethnically diverse populations suggest a higher prevalence in minorities, yet these studies have generally been of both minority and of low socio-economic status populations, and there have been few attempts to separate these factors. The current study sought to determine the prevalence of active epilepsy in two rural,
predominantly Caucasian southeastern Kansas counties for the year 2008.
Methods Labette and Montgomery counties were selected for this study because of their rural, low socio-economic status, and predominately Caucasian population.
Labette County According to the 2000 census (United States Census Bureau, 2001a), there were 22,835 people (51.1% female) in Labette County. The median income for a household in the county was $30,875, the per capita income was $15,525, and 12.7% of the population was below the poverty line. The racial makeup of the county was 89.3% White and 4.7% Black or African American, and 3.1% of the population was Hispanic or Latino.
Montgomery County There were 36,252 people (51.8% female) in Montgomery County (United States Census Bureau, 2001b). The median income for a household in the county was $30,997, the per capita income was $16,421, and 12.60% of the population was below the poverty line. The racial makeup of the county was 85.8% White and 6.1% Black or African American, and 3.1% of the population was Hispanic or Latino. For 2006—2010, the percent of persons 25 years and older with a bachelor’s degree or higher was 17.8% in Labette County and 18.6% in Montgomery County; these are considerably lower than the United States’ average (29.3%) (United States Census Bureau, 2012a,b,c).
Definitions Epilepsy was defined as a condition manifest by recurrent unprovoked epileptic seizures. Active epilepsy was defined as having a history of recurrent, unprovoked seizures and having at least one unprovoked seizure or having taken anticonvulsant medication within five years of the prevalence year of 2008 (Hauser et al., 1991).
Procedures Those with active epilepsy in 2008 were identified at one or more of 20 hospitals, clinics, doctors’ offices, or a nursing home either in the two counties or at presumed epilepsy specialist referral centers ‘‘near’’ (within 300 miles) these
794
E. Ablah et al.
Figure 1
Map of participating referral neurologists near Labette and Montgomery counties.
counties. Though there was an initial order in which the sites were searched, in the end, many sites were searched simultaneously over four years, starting in 2006. There were nine hospitals and ten clinics and doctors’ offices and one nursing home searched overall. In Labette County, both of the two hospitals and two of the eight doctor’s offices agreed to participate. Parsons State Hospital, a statewide referral center, resides in Labette County. Cases reviewed from Parsons State Hospital were limited to residents of the two counties prior to the time of admission. Among the people admitted to this facility from the two counties, there were no active prevalent epilepsy cases among those residents. In Montgomery County, both of the two hospitals, two of the seven doctor’s offices, and one nursing home agreed to participate. Four hospitals and six doctor’s offices outside of the two counties who treated individuals from the two counties also participated. Eight neurologists participated: the two neurologists in the two counties, three neurologists in Wichita (one at a tertiary epilepsy center), and two neurologists at the University of Kansas Medical Center in Kansas City and the director of the epilepsy program at the University of Oklahoma (Fig. 1). Other neurologists in the region (Overland Park, KS, Tulsa, OK, Joplin, MO, Springfield, MO) were contacted and declined participation. All facilities providing electroencephalography (EEG) services in the two counties participated in the study, and their referrals to these laboratories were reviewed. Nurses from the local Area Health Education Center (AHEC) conducted the initial case finding and abstracted
medical records with ICD 9 entries of 345.xx or 780.39 from the medical facilities. Medical records were abstracted for zip code of residence, seizure information, date of birth, gender, and EEG or imaging findings (which were seldom available). It was not possible to obtain race and ethnicity from most medical records. Upon obtaining relevant information from medical records, consensus conferences were convened to determine which of the potential cases might be active prevalent epilepsy. For those considered to have prevalent epilepsy as well as those with insufficient information to make a definitive diagnosis of epilepsy, AHEC staff prepared letters to be mailed from hospitals and doctors’ offices to invite their potentially eligible patients to participate in an interview. Consenting cases were then interviewed by the AHEC nurses using a structured interview (Ottman et al., 2010). The interviews included details of the seizure phenomenology, quality of life, employment, and sources of medical care. Approximately 10% of identified cases were interviewed. Medical record information and the interview, when available, were used for the final determination of active epilepsy, seizure type, etiology, syndrome, age, and gender. The final epilepsy classification was done in a consensus conference led by one epileptologist team member. The study protocol was approved by the Human Subjects Committee at the University of Kansas School of Medicine-Wichita, the Institutional Review Board at Columbia University, and institutional review boards at all participating institutions.
Epilepsy prevalence
795
Figure 2
Flow chart of cases meeting inclusion criteria to classification.
Factors examined Prevalence was calculated according to age, etiology, seizure type, and syndrome. Age was categorized as 80 years for the calculations of prevalence. For the capture recapture analysis, broader age categories were used to make more stable estimates. Etiology was classified according to etiologic categories (e.g., idiopathic/cryptogenic, remote symptomatic, progressive symptomatic, and other/unknown (cases lacked sufficient information to classify etiology)) (United States Census Bureau, 2012b). Seizure type was classified as primary generalized, focal with or without secondary generalization, both focal and generalized, spasms, unclassified (when seizure type could not be determined despite adequate descriptions), and unknown (when there was no information on seizure type available). Syndrome was categorized into broad categories including primary generalized, other generalized, focal with or without secondary generalization (localization related), both focal and generalized, no definitive features of focal or generalized epilepsy, and unknown (when there was no information upon which to classify syndrome) (United States Census Bureau, 2012c).
Statistical analysis Prevalence of active epilepsy in 2008 was calculated per 1000, using the extrapolated 2008 U.S. Census Bureau estimates for the two counties (United States Census Bureau, 2008). An active prevalence case was a person with a diagnosis of epilepsy, alive, living in the study counties in 2008, and known to have taken antiepileptic medications or had a seizure in the preceding five years. Prevalence was age adjusted to the US 2000 population (Klein and Schoenborn, 2001). Capture—recapture method People with prevalent epilepsy did not necessarily seek care with just one hospital or physician, and visits to more
than one facility were possible. It was also possible that some people with prevalent epilepsy never visited a hospital or physician’s office during the study period. As such, capture—recapture methodology, designed to avoid case ascertainment bias, was employed to estimate the number of people with active epilepsy in the two counties, based upon what was known about the study population. Therefore, independent lists of patients from each participating hospital, clinic, and doctor’s office where cases were identified were developed. Since most capture—recapture analyses assume equal likelihood of identification at each center, a separate analysis was performed excluding lists identified at epilepsy referral centers, since such cases may not have represented independent ascertainment. Log-linear models are the foundation of the capture—recapture assessments used to estimate the prevalence of disease in a given population (Baillargeon and Rivest, 2007). The Rcapture Package in R2 was used to conduct the capture—recapture analysis (R Development Core Team, 2011). This package was used to produce an estimate of the total number of prevalent people in the two-county region during 2008, given information about the observed capture history for patients in the sample. Since the observation period was relatively short, the assumption was that the population was closed (Baillargeon and Rivest, 2007; Chao, 2001). In addition, the assumption was that the probability of being captured t times could vary across individuals. There was no assumption of temporal variation, nor was there an assumption that the probability of being captured would differ before and after the first capture (Baillargeon and Rivest, 2007). The Rcapture package contains several different formulae for capture—recapture calculations, but the results reported are based upon Chao’s methodology (Chao, 2001). The Chao methodology is an accurate estimator in situations where there is heterogeneity in capture probability, and if many subjects have very small capture probabilities so that they are caught only once or twice. In most cases, the Chao estimate provided the most conservative estimate of numbers of people with epilepsy. It was also the method which for the most part yielded the lowest Akaike information criterion (AIC) and optimal distribution of
796
E. Ablah et al.
Prevlaence per 1,000
12 10 8 6
Males
4
Females
2
All Cases
0
Age in Years
Figure 3
Prevalence by age and gender.
Pearson residuals. The deviance was evaluated to verify that the model was a good fit. The capture—recapture analysis was conducted for the total sample and after stratification by gender and age groups (
journal homepage: www.elsevier.com/locate/epilepsyres
Prevalence of epilepsy in rural Kansas Elizabeth Ablah a, Dale C. Hesdorffer b, Yi Liu c, Angelia M. Paschal d, Suzanne Hawley e, David Thurman f, W. Allen Hauser b,g,∗ , The Prevalence of Epilepsy in Rural Kansas Study Group1 a
Department of Preventive Medicine and Public Health, University of Kansas School of Medicine — Wichita, United States b Department of Epidemiology and GH Sergievsky Center, Columbia University, United States c Department of Biostatistics and GH Sergievsky Center, Columbia University, United States d Department of Health Science, University of Alabama, United States e Department of Public Health Sciences, Wichita State University, United States f Department of Neurology, Emory University, United States g Department of Neurology and GH Sergievsky Center, Columbia University, United States Received 18 January 2013; received in revised form 9 December 2013; accepted 16 January 2014 Available online 13 February 2014
KEYWORDS Epilepsy; Prevalence; Rural; Capture—recapture
Summary Purpose: To determine the prevalence of active epilepsy in two southeastern rural Kansas counties. Methods: Medical records were abstracted from the emergency rooms, out- and inpatient services and clinics of 9 hospitals, from 10 doctors’ offices, and 1 nursing home in and surrounding the two counties. Letters were mailed from hospitals and doctors’ offices to invite their potentially eligible patients to participate in an interview. Medical record information and the interview, when available, were used for the final determination of active epilepsy, seizure type, etiology, syndrome, age, and gender in consensus conferences. Prevalence of epilepsy was calculated, and capture—recapture methodology, which estimates prevalence based on what is known about the population, was employed to assess active epilepsy in the two counties. Results: This study identified 404 individuals with active prevalent epilepsy who visited at least one of the 20 facilities during the observation period. The overall prevalence of active epilepsy was 7.2 per 1000. The seizure type for 71.3% of prevalent cases was unknown; among the 76 cases with known and classifiable seizure type, 55.3% had focal with secondary generalized seizures. Among the 222 cases with classifiable etiology, 53.1% were idiopathic/cryptogenic. About 75% (n = 301) were captured at only one center, 72% (n = 75) of the remaining 103 patients were captured at two centers, and 28 patients were identified at three or more centers. The
∗ Corresponding author at: Department of Epidemiology and GH Sergievsky Center, Columbia University, 680 West 168 Street, New York, NY 10032, United States. Tel.: +1 914 760 3144; fax: +1 212 305 2426. E-mail address: [email protected] (W.A. Hauser). 1 See Appendix A for ‘The Prevalence of Epilepsy in Rural Kansas Study Group’ members.
0920-1211/$ — see front matter © 2014 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.eplepsyres.2014.01.001
Epilepsy prevalence
793 capture—recapture assessment yielded an estimation of 982 prevalent patients. The overall estimated prevalence of epilepsy in the two Kansas counties using capture—recapture was 17 per 1000. Conclusions: The crude prevalence of epilepsy, using medical record survey methods, was similar to, but on the high end, of other total population prevalence studies in the United States. The capture—recapture assessment suggested that epilepsy prevalence might be considerably higher than the crude prevalence. © 2014 Elsevier B.V. All rights reserved.
Background Epilepsy is among the most common of all neurologic conditions (U.S. Department of Health and Human Services, 2008). There are an estimated 50 million people with epilepsy in the world (WHO, 2012), with an estimated prevalence of active epilepsy of 4—7 per 1000 persons in developed countries (Sander, 2003) and 5—74 per 1000 persons in developing countries (Preux and Druet-Cabanc, 2005; Ngugi et al., 2010). In the United States, an estimated 2.2 million individuals have epilepsy (Institute of Medicine, 2012). As a spectrum disease with symptoms that may mimic other conditions, epilepsy can be especially difficult to diagnose, making an accurate census of the disease even more of a challenge. Beyond difficulties inherent in diagnosing the condition itself, the logistic realities of determining the prevalence of epilepsy in a given region serve as impediments to determining widespread and reliable prevalence estimates for epilepsy. Because of the time and expense of such efforts, rural and other at-risk populations are less likely to be canvassed in a manner to reliably determine prevalence (Asawavichienjinda et al., 2002). Despite the inherent difficulties in determining prevalence, there exists a large body of literature centered on epilepsy prevalence across the world. In the large number of studies (more than 5000 since 1965) (Banerjee et al., 2009) conducted on the epidemiology of epilepsy and the wide variety of methodologies used and areas studied, studies of urban communities have predominated (Benn et al., 2009; Kelvin et al., 2007). Rural communities in the United States remain under-studied for epilepsy prevalence. Although most studies focus on urban populations in the United States, epilepsy is estimated to be more prevalent in rural communities than in urban communities (Haerer et al., 1986; Hollingsworth, 1978; Baumann et al., 1977). Studies of rural populations worldwide have suggested epilepsy prevalence between 2.7 per 1000 in Italy (Reggio et al., 1996) and 17.6 per 1000 in Chile (Lavados et al., 1992). Prevalence varies dramatically by region and often corresponds to the countries’ degree of development or industrialization (Strzelczyk et al., 2008). Within countries, epilepsy can be as much as twice as common in rural areas than urban areas (Gourie-Devi et al., 2004; Aziz et al., 1997). Few rural, population-based prevalence studies have been conducted in the United States. Moreover, studies that have included racially and ethnically diverse populations suggest a higher prevalence in minorities, yet these studies have generally been of both minority and of low socio-economic status populations, and there have been few attempts to separate these factors. The current study sought to determine the prevalence of active epilepsy in two rural,
predominantly Caucasian southeastern Kansas counties for the year 2008.
Methods Labette and Montgomery counties were selected for this study because of their rural, low socio-economic status, and predominately Caucasian population.
Labette County According to the 2000 census (United States Census Bureau, 2001a), there were 22,835 people (51.1% female) in Labette County. The median income for a household in the county was $30,875, the per capita income was $15,525, and 12.7% of the population was below the poverty line. The racial makeup of the county was 89.3% White and 4.7% Black or African American, and 3.1% of the population was Hispanic or Latino.
Montgomery County There were 36,252 people (51.8% female) in Montgomery County (United States Census Bureau, 2001b). The median income for a household in the county was $30,997, the per capita income was $16,421, and 12.60% of the population was below the poverty line. The racial makeup of the county was 85.8% White and 6.1% Black or African American, and 3.1% of the population was Hispanic or Latino. For 2006—2010, the percent of persons 25 years and older with a bachelor’s degree or higher was 17.8% in Labette County and 18.6% in Montgomery County; these are considerably lower than the United States’ average (29.3%) (United States Census Bureau, 2012a,b,c).
Definitions Epilepsy was defined as a condition manifest by recurrent unprovoked epileptic seizures. Active epilepsy was defined as having a history of recurrent, unprovoked seizures and having at least one unprovoked seizure or having taken anticonvulsant medication within five years of the prevalence year of 2008 (Hauser et al., 1991).
Procedures Those with active epilepsy in 2008 were identified at one or more of 20 hospitals, clinics, doctors’ offices, or a nursing home either in the two counties or at presumed epilepsy specialist referral centers ‘‘near’’ (within 300 miles) these
794
E. Ablah et al.
Figure 1
Map of participating referral neurologists near Labette and Montgomery counties.
counties. Though there was an initial order in which the sites were searched, in the end, many sites were searched simultaneously over four years, starting in 2006. There were nine hospitals and ten clinics and doctors’ offices and one nursing home searched overall. In Labette County, both of the two hospitals and two of the eight doctor’s offices agreed to participate. Parsons State Hospital, a statewide referral center, resides in Labette County. Cases reviewed from Parsons State Hospital were limited to residents of the two counties prior to the time of admission. Among the people admitted to this facility from the two counties, there were no active prevalent epilepsy cases among those residents. In Montgomery County, both of the two hospitals, two of the seven doctor’s offices, and one nursing home agreed to participate. Four hospitals and six doctor’s offices outside of the two counties who treated individuals from the two counties also participated. Eight neurologists participated: the two neurologists in the two counties, three neurologists in Wichita (one at a tertiary epilepsy center), and two neurologists at the University of Kansas Medical Center in Kansas City and the director of the epilepsy program at the University of Oklahoma (Fig. 1). Other neurologists in the region (Overland Park, KS, Tulsa, OK, Joplin, MO, Springfield, MO) were contacted and declined participation. All facilities providing electroencephalography (EEG) services in the two counties participated in the study, and their referrals to these laboratories were reviewed. Nurses from the local Area Health Education Center (AHEC) conducted the initial case finding and abstracted
medical records with ICD 9 entries of 345.xx or 780.39 from the medical facilities. Medical records were abstracted for zip code of residence, seizure information, date of birth, gender, and EEG or imaging findings (which were seldom available). It was not possible to obtain race and ethnicity from most medical records. Upon obtaining relevant information from medical records, consensus conferences were convened to determine which of the potential cases might be active prevalent epilepsy. For those considered to have prevalent epilepsy as well as those with insufficient information to make a definitive diagnosis of epilepsy, AHEC staff prepared letters to be mailed from hospitals and doctors’ offices to invite their potentially eligible patients to participate in an interview. Consenting cases were then interviewed by the AHEC nurses using a structured interview (Ottman et al., 2010). The interviews included details of the seizure phenomenology, quality of life, employment, and sources of medical care. Approximately 10% of identified cases were interviewed. Medical record information and the interview, when available, were used for the final determination of active epilepsy, seizure type, etiology, syndrome, age, and gender. The final epilepsy classification was done in a consensus conference led by one epileptologist team member. The study protocol was approved by the Human Subjects Committee at the University of Kansas School of Medicine-Wichita, the Institutional Review Board at Columbia University, and institutional review boards at all participating institutions.
Epilepsy prevalence
795
Figure 2
Flow chart of cases meeting inclusion criteria to classification.
Factors examined Prevalence was calculated according to age, etiology, seizure type, and syndrome. Age was categorized as 80 years for the calculations of prevalence. For the capture recapture analysis, broader age categories were used to make more stable estimates. Etiology was classified according to etiologic categories (e.g., idiopathic/cryptogenic, remote symptomatic, progressive symptomatic, and other/unknown (cases lacked sufficient information to classify etiology)) (United States Census Bureau, 2012b). Seizure type was classified as primary generalized, focal with or without secondary generalization, both focal and generalized, spasms, unclassified (when seizure type could not be determined despite adequate descriptions), and unknown (when there was no information on seizure type available). Syndrome was categorized into broad categories including primary generalized, other generalized, focal with or without secondary generalization (localization related), both focal and generalized, no definitive features of focal or generalized epilepsy, and unknown (when there was no information upon which to classify syndrome) (United States Census Bureau, 2012c).
Statistical analysis Prevalence of active epilepsy in 2008 was calculated per 1000, using the extrapolated 2008 U.S. Census Bureau estimates for the two counties (United States Census Bureau, 2008). An active prevalence case was a person with a diagnosis of epilepsy, alive, living in the study counties in 2008, and known to have taken antiepileptic medications or had a seizure in the preceding five years. Prevalence was age adjusted to the US 2000 population (Klein and Schoenborn, 2001). Capture—recapture method People with prevalent epilepsy did not necessarily seek care with just one hospital or physician, and visits to more
than one facility were possible. It was also possible that some people with prevalent epilepsy never visited a hospital or physician’s office during the study period. As such, capture—recapture methodology, designed to avoid case ascertainment bias, was employed to estimate the number of people with active epilepsy in the two counties, based upon what was known about the study population. Therefore, independent lists of patients from each participating hospital, clinic, and doctor’s office where cases were identified were developed. Since most capture—recapture analyses assume equal likelihood of identification at each center, a separate analysis was performed excluding lists identified at epilepsy referral centers, since such cases may not have represented independent ascertainment. Log-linear models are the foundation of the capture—recapture assessments used to estimate the prevalence of disease in a given population (Baillargeon and Rivest, 2007). The Rcapture Package in R2 was used to conduct the capture—recapture analysis (R Development Core Team, 2011). This package was used to produce an estimate of the total number of prevalent people in the two-county region during 2008, given information about the observed capture history for patients in the sample. Since the observation period was relatively short, the assumption was that the population was closed (Baillargeon and Rivest, 2007; Chao, 2001). In addition, the assumption was that the probability of being captured t times could vary across individuals. There was no assumption of temporal variation, nor was there an assumption that the probability of being captured would differ before and after the first capture (Baillargeon and Rivest, 2007). The Rcapture package contains several different formulae for capture—recapture calculations, but the results reported are based upon Chao’s methodology (Chao, 2001). The Chao methodology is an accurate estimator in situations where there is heterogeneity in capture probability, and if many subjects have very small capture probabilities so that they are caught only once or twice. In most cases, the Chao estimate provided the most conservative estimate of numbers of people with epilepsy. It was also the method which for the most part yielded the lowest Akaike information criterion (AIC) and optimal distribution of
796
E. Ablah et al.
Prevlaence per 1,000
12 10 8 6
Males
4
Females
2
All Cases
0
Age in Years
Figure 3
Prevalence by age and gender.
Pearson residuals. The deviance was evaluated to verify that the model was a good fit. The capture—recapture analysis was conducted for the total sample and after stratification by gender and age groups (
