Cognitive Decline in Older Adults with Epilepsy: The Cardiovascular Health Study

SUMMARY

Objective:

Cognitive decline is a major concern for older adults with epilepsy. Whether and how much faster older adults with epilepsy experience cognitive decline beyond expected age-related cognitive change remain unclear. We sought to estimate and compare rates of cognitive decline in older adults with and without epilepsy.

Methods:

The Cardiovascular Health Study is a population-based longitudinal cohort study of 5,888 US adults aged 65+. Cognitive function was assessed annually with Modified Mini-Mental State Exam (3MS) and Digit Symbol Substitution Test (DSST). We used linear mixed models to estimate average rates of decline in 3MS and DSST scores by epilepsy status (prevalent, incident, or no epilepsy), adjusted for risk factors associated with cognitive decline.

Results:

The rate of decline in 3MS was significantly faster in prevalent epilepsy (p < 0.001) and after incident epilepsy (p = 0.002) compared with no epilepsy. Prevalent epilepsy and ApoE4 had a synergistic interaction, whereby prevalent epilepsy and ApoE4 together were associated with 1.51 points faster annual decline in 3MS than would be expected if prevalent epilepsy and ApoE4 did not interact (P < 0.001). Older adults with prevalent epilepsy had significantly lower initial DSST score and faster rate of decline compared to those with no epilepsy (P <0.001).

Significance:

Faster decline in global cognitive ability seen in this study validates concerns of patients. ApoE4 allele status was an effect modifier of the relationship between cognitive decline and prevalent epilepsy. Further research is warranted to explore biological mechanisms and possible interventions to mitigate cognitive decline.

INTRODUCTION

Cognitive decline is a major concern for older adults,¹ but among patients with epilepsy, it ranks highest on a list of potential concerns.² As people age, cognitive function declines on average,³ but unknown is whether and how much faster older adults with epilepsy experience cognitive decline beyond the expected age-related cognitive change. Studies of younger patients with temporal lobe epilepsy show that there are structural and functional changes in the brain suggesting aging effects of temporal lobe epilepsy, particularly in refractory cases.^4–8 Epilepsy in adults 65 years and older may have been present since earlier in life (prevalent) or may be of new onset (incident). Studies focusing on cognitive function of older adults with epilepsy have mostly employed cross-sectional designs and have focused on patients with prevalent epilepsy.^9–12 Population-based longitudinal studies of older adults with prevalent or incident epilepsy, comparing their cognitive decline with that of older adults without epilepsy, are scarce. However, two recent longitudinal studies, one revealing a 2-fold risk of subsequent diagnosis of dementia in older veterans following the diagnosis of epilepsy,¹³ and another demonstrating a faster decline of cognition in late-onset epilepsy, with some individuals having cognitive decline prior to the first seizure,¹⁴ help support an emerging view that late-onset epilepsy and dementia may share a common underpinning such as vascular risk factors.^{15, 16} In this study, we sought to compare the rates of cognitive decline among a broader population of older adults with prevalent or incident epilepsy compared to older adults without epilepsy, using data collected in the Cardiovascular Health Study (CHS). We hypothesized that older adults with prevalent or incident epilepsy experience faster cognitive decline than older adults without epilepsy in a population-based study in the US.

METHODS

Design, setting, and participants

The Cardiovascular Health Study (CHS) is a population-based multi-center observational cohort study of coronary heart disease and stroke in 5,888 adults aged 65 years and older in whom data were collected prospectively.^{17, 18} Between 1989 and 1993, CHS enrolled participants from a random sample of men and women on the Health Care Financing Administration Medicare eligibility lists and their household members aged 65 and older residing in four U.S. communities whose populations were diverse in factors such as degree of urbanization, attained education, household income levels, and availability of healthcare: Forsyth County, North Carolina; Sacramento County, California; Washington County, Maryland; and Pittsburgh (Allegheny County), Pennsylvania. Sampling was stratified by age, sex, and race (black, white). Exclusion criteria were being institutionalized or home-bound, receiving treatment for cancer, planning to move away from the community within three years, and being unable give informed consent or respond to questions without the aid of a surrogate respondent. In-person evaluations were performed at baseline and at follow-up visits that occurred every 12 months through 1999. Baseline and follow-up data collection included questionnaires for psychosocial, medical history, physical activity, and dietary factors, medication inventories, and physical and laboratory evaluations including blood pressure, anthropometry, physical function tests, electrocardiography, echocardiography, ultrasonography, and numerous blood assays. These measures were used to identify the presence of cardiovascular disease (CVD) risk factors such as hypertension, subclinical disease such as carotid artery atherosclerosis, and clinically overt CVD. Incident CVD events including stroke occurring during follow-up were ascertained and verified. The institutional review board at each study site and at the data coordinating center approved the study, and all participants provided informed consent.

Adjudication of epilepsy

Ascertainment of epilepsy diagnosis was not part of the original CHS study. We developed an algorithm¹⁹ to identify CHS participants who had epilepsy, relying on previously validated International Classification of Diseases, Ninth Revision (ICD-9) codes for epilepsy submitted during hospitalizations that occurred during CHS follow-up, outpatient ICD-9 codes for epilepsy submitted during CHS follow-up, antiepileptic medication use, and self-report of seizures during study follow up.^20–23 Outpatient ICD-9 codes were available from the Centers for Medicare and Medicaid Services (CMS), a social insurance program that provides insurance to nearly 40 million older Americans, which recently merged CMS claims data for all CHS participants who were enrolled in CMS during follow-up. In our algorithm, epilepsy cases were defined as participants with (1) ≥2 ICD-9 codes (345.xx or 780.3x) from any setting, or (2) ≥1 ICD-9 code plus antiepileptic medication or self-report.¹⁹ We did not include participants who had a prescription of antiepileptic medication alone to exclude cases who received antiepileptic medications for non-epileptic indications such as depression, trigeminal neuralgia, or headaches. Two neurologists specializing in epilepsy independently reviewed all hospitalization and outpatient ICD-9 codes, antiepileptic medication use, and self-report data. The two neurologists independently assigned each participant’s epilepsy status, achieving kappa of 99.2%, then resolved discordant classifications by joint review and consensus.¹⁹

Exposure variable

We defined incident epilepsy cases as a subset of epilepsy cases who had no claims with epilepsy ICD-9 codes for the first 2 years after enrollment into CHS.¹⁹ The remaining epilepsy cases were regarded as prevalent epilepsy cases. The comparison group included participants with no epilepsy. We considered incident epilepsy to be a time-varying variable, updated at each outcome assessment, so cognitive function measures obtained prior to incident epilepsy were classified as no epilepsy.

Outcome variables

The primary outcomes were cognitive functioning assessed by Modified Mini-Mental State Examination (3MS) and Digit Symbol Substitution Test (DSST). 3MS is a test of global cognitive performance including recall of personal information (date and place of birth), learning three words and recalling them after short and long delays, counting and spelling backwards, orientation to present time and place, naming body parts, generating a list of four-legged animals, identifying similarities, repeating a sentence verbally and in writing, reading and obeying a written command, copying two pentagons, and obeying a three-stage verbal command.²⁴ The 3MS was administered annually up to 9 times from 1990/1991 to 1998/1999. Its score ranges from zero (worst) to 100 (best), with 3MS score <78 suggested as a cutoff for dementia screening in the literature.²⁵ The 3MS is an expanded version of the Mini-Mental Status Examination, including additional items that allow for 100-point instead of 30-point scoring to achieve more fine-grained discrimination among individuals.^{24, 26} DSST, a 90-second test of information processing speed, visuomotor coordination, and attention, was administered annually up to 10 times from 1989/1990 to 1998/1999. Its score ranges from zero (worst) to 90 (best). Participants included in our analysis contributed a mean (SD) of 6.7 (2.6) 3MS scores and a mean (SD) of 7.2 (2.9) DSST scores.

Covariates

Covariates included age, sex, race (black versus white or other), education (>12^th grade or not), smoking, alcohol use, body mass index, waist circumference, systolic blood pressure, triglycerides, high density lipoprotein (HDL) cholesterol, depressive symptom score measured with Center for Epidemiologic Studies Depression Scale or CES-D (a scale ranging from 0 to 30 with higher scores indicating more depressive symptoms), self-rated health (fair or poor versus good to excellent), ApoE4 genotype (at least one e4 allele versus none), and comorbidities present at baseline. Comorbidities included hypertension, diabetes mellitus, coronary heart disease, atrial fibrillation, heart failure, stroke, and chronic kidney disease. We also considered incident stroke during follow-up in order to establish a stroke-free cohort for some analyses. Analyses that included ApoE4 genotype were limited to those with available DNA who consented to genetic studies.

Statistical analysis

We used repeated measures of cognition and linear mixed models to assess the associations of prevalent and incident epilepsy on average cognitive trajectories during aging. The linear mixed model for each cognitive outcome was as follows, with study year and incident epilepsy as time-varying variables:

$$\text{E}\left(\text{cognitive score}\right)=\text{intercept}+\text{prevalent epilepsy}+\text{incident epilepsy}+\text{study year}+\left(\text{study year }\times \text{ prevalent epilepsy}\right)+\left(\text{study year }\times \text{ incident epilepsy}\right)+\text{study year squared}+\text{adjustment covariates}+\text{random effects including intercept},\text{\hspace{0.17em}study year},\text{\hspace{0.17em}and study year squared}.$$

For each cognitive measure, Model 1 was adjusted for all demographics, health behaviors, physiologic parameters, and clinical characteristics listed above except ApoE4 genotype. Model 2 was additionally adjusted for ApoE4 genotype.

We considered effect modification of the association of epilepsy with cognitive decline by ApoE4 genotype using an extended Model 2 that also included terms for (ApoE4 × prevalent epilepsy), (ApoE4 × incident epilepsy), (study year × ApoE4), (study year × ApoE4 × prevalent epilepsy), and (study year × ApoE4 × prevalent epilepsy).

Models 1 and 2 were then reanalyzed excluding participants with history of clinical stroke at baseline and censoring cognitive measures obtained after incident stroke that occurred during CHS follow-up.

Data availability statement

De-identified data will be shared by request from any qualified investigator.

RESULTS

Among 5,888 CHS participants, 215 had prevalent epilepsy and 120 had incident epilepsy through December 31, 2010.¹⁹ For the present study, we included participants who underwent cognitive testing with 3MS (n=5,672) or DSST (n=5,790) during the first 10 years of the study, through 1999. Thus, our analysis of 3MS trajectories included 208 prevalent cases and 21 incident cases (Table 1), and our analysis of DSST trajectories included 207 prevalent and 16 incident cases, as most incident epilepsy cases occurred after the first 10 years of the study. Average age at study entry was 73 years and 58% of participants were women (Table 1). More participants in the incident epilepsy group had prevalent hypertension, atrial fibrillation, heart failure, stroke, or chronic kidney disease at study entry than those with prevalent epilepsy or without epilepsy. At baseline cognitive assessment, 8.4% of participants without epilepsy and 11.5% of participants with prevalent epilepsy scored <78 on 3MS, a suggested threshold for dementia screening. At final cognitive assessment, 18.0% of participants who remained free of epilepsy, 26.3% of participants who had developed incident epilepsy, and 28.6% of participants with prevalent epilepsy scored <78 on 3MS.

Table 1.

Baseline characteristics of participants included in 3MS analysis

Characteristic*	All participants (n = 5,672)	Participants without epilepsy at baseline or during follow-up (n =5,443)	Participants with prevalent epilepsy at baseline (n = 208)	Participants with incident epilepsy during follow-up (n = 21)
Demographics
Age, y, mean (SD)	73.2 (5.5)	73.2 (5.5)	73.0 (5.0)	73.5 (4.6)
Female, %	57.8	57.9	55.8	57.1
Black race, %	16.0	15.8	20.2	19.1
Any education beyond 12th grade, %	43.4	43.5	44.7	9.5
Among those without any education beyond 12th grade: years of education through 12th grade, mean (SD)	10.1 (2.4)	10.1 (2.4)	10.1 (2.4)	9.6 (2.5)
Health behaviors
Former smoking, %	41.8	41.7	43.5	42.9
Current smoking, %	11.7	11.6	15.0	28.6
Any current alcohol use, %	50.3	50.3	51.2	47.6
Among drinkers: drinks/wk, mean (SD)	4.9 (8.3)	4.9 (8.2)	7.2 (10.4)	2.5 (2.1)
Physiological and clinical characteristics
Body mass index, kg/m2, mean (SD)	26.7 (4.7)	26.7 (4.7)	26.4 (4.3)	26.2 (4.5)
Waist circumference, cm, mean (SD)	94.5 (13.2)	94.6 (13.2)	93.6 (12.9)	92.5 (14.5)
Hypertension, %	58.5	58.5	57.7	71.4
Systolic blood pressure, mm Hg, mean (SD)	136.4 (21.7)	136.4 (21.7)	137.2 (22.5)	133.5 (29.5)
Triglycerides, mg/dL, mean (SD)	139.5 (77.1)	139.7 (75.9)	136.4 (105.8)	114.2 (53.5)
HDL cholesterol, mg/dL, mean (SD)	54.3 (15.8)	54.2 (15.7)	56.2 (16.8)	60.6 (18.6)
Diabetes, %	15.6	15.5	18.9	10.0
Coronary heart disease, %	19.0	18.9	22.6	19.1
Atrial fibrillation, %	2.7	2.7	2.9	9.5
Heart failure, %	4.2	4.2	2.4	9.5
Stroke, %	4.0	4.0	4.8	19.1
Chronic kidney disease (eGFR < 60), %	39.1	39.0	41.2	55.0
Depressive symptom score, mean (SD)	4.7 (4.6)	4.6 (4.6)	5.2 (4.7)	5.6 (6.0)
Fair/poor self-rated health, %	21.5	21.4	26.9	15.0
Genetics
≥1 ApoE4 allele, %	25.4	25.3	27.3	22.2

^*Numbers of participants with missing values (number with prevalent and incident epilepsy): 16 for education, 6 for smoking status, 25 for alcohol use, 19 for body mass index, 34 for waist circumference, 7 for hypertension, 9 for systolic blood pressure, 50 for triglycerides, 58 for HDL cholesterol, 93 for diabetes, 74 for chronic kidney disease, 10 for depressive symptom score, 130 for self-rated health, and 626 for ApoE4 genotype. Means and percentages were calculated using all available data for each characteristic.

3MS

Main analysis.

We estimated longitudinal 3MS trajectories over a mean of 6.7 annual 3MS administrations. The rate of cognitive decline accelerated over time as participants aged (Figure 1, Panel A). The rate of decline was significantly faster in prevalent epilepsy (p < 0.001) and after incident epilepsy (p = 0.002) compared with no epilepsy, adjusted for demographics, health behaviors, and physiological and clinical characteristics (Table 2). Mean 3MS score in those without epilepsy declined on average 0.4 points per year from baseline through year 4 of follow-up (from mean initial score of 89.8 to mean of 88.3 at year 4), then declined more quickly at a rate of 1.2 points per year on average from year 4 to year 8 of follow-up (from mean of 88.3 to 83.3), as seen in Table 2 under Model 1. Although initial mean 3MS score was similar in prevalent epilepsy and no epilepsy, mean 3MS score declined faster in those with prevalent epilepsy, with a decline of 1.1 point per year on average from baseline to year 4 (from mean initial score of 89.1 to mean of 84.5 at year 4), and then 2.0 points per year on average from year 4 to year 8 (from mean score of 84.5 to 76.5). In participants who developed incident epilepsy during follow-up, mean decline in 3MS score after epilepsy onset was 2.1 points per year through year 4 (from mean initial score of 92.1 to mean of 88.0 at year 4), and then 2.7 points per year on average from year 4 to year 8 (from mean of 88.0 to 77.2). Confidence intervals suggest that in this population, older adults with prevalent epilepsy may decline 0.44 to 1.11 points faster in mean 3MS score per year relative to their peers without epilepsy, while older adults with incident epilepsy may subsequently decline 0.55 to 2.39 points faster in mean 3MS score per year compared with people who do not have epilepsy (Table 2).

Figure 1.

Estimated trajectories of mean 3MS and DSST scores by epilepsy status

Trajectories shown in the graphs are derived from linear mixed models: E(Cognitive score) = β₀ (intercept) + β₁ × prevalent epilepsy + β₂ × incident epilepsy + β₃ × year + β₄ × year × prevalent epilepsy + β₅ × year × incident epilepsy + β₆ × year squared + β_i × adjustment variables + random effects. Adjustment variables were baseline values of age, sex, race, education, smoking, alcohol use, body mass index, waist circumference, hypertension, systolic blood pressure, triglycerides, HDL cholesterol, diabetes, coronary heart disease, atrial fibrillation, heart failure, stroke, chronic kidney disease, depressive symptom score, and self-rated health (Model 1 in Table 2). Trajectories in incident epilepsy do not begin at analysis year 0 because cognitive test scores were first obtained after incident epilepsy onset in analysis year 2 for 3MS analysis, and in analysis year 3 for DSST analysis.

Table 2.

Regression coefficients and estimated trajectories of mean 3MS and DSST scores by epilepsy status

	Models estimating trajectory of mean 3MS score‡						Models estimating trajectories of mean DSST score‡
	Model 1 (n = 5,341)			Model 2 (n = 4,794)			Model 1 (n = 5,327)			Model 2 (n = 4,780)
	Est	95% CI	P	Est	95% CI	P	Est	95% CI	P	Est	95% CI	P
Regression coefficients *
β0: Intercept (no epilepsy, year 0)	89.8	(89.6, 90.0)	<0.001	89.8	(89.6, 90.1)	<0.001	37.1	(36.8, 37.4)	<0.001	37.2	(36.9, 37.5)	<0.001
β1: Prevalent epilepsy	−0.64	(−1.69, 0.41)	0.233	−0.31	(−1.41, 0.80)	0.584	−1.88	(−3.33, −0.42)	0.011	−1.64	(−3.15, −0.13)	0.033
β2: Incident epilepsy	5.64	(−0.19, 11.5)	0.058	4.36	(−1.33, 10.1)	0.133	−5.53	(−12.5, 1.48)	0.122	−5.54	(−12.6, 1.48)	0.122
β3: Year	0.06	(−0.03, 0.14)	0.206	0.05	(−0.04, 0.14)	0.274	0.47	(0.39, 0.55)	<0.001	0.47	(0.39, 0.56)	<0.001
β4: Year × prevalent epilepsy	−0.77	(−1.11, −0.44)	<0.001	−0.78	(−1.12, −0.44)	<0.001	−0.32	(−0.52, −0.13)	0.001	−0.25	(−0.45, −0.05)	0.013
β5: Year × incident epilepsy	−1.47	(−2.39, −0.55)	0.002	−0.82	(−1.74, 0.09)	0.078	0.56	(−0.42, 1.54)	0.261	0.55	(−0.43, 1.53)	0.268
β6: Year squared	−0.11	(−0.12, −0.10)	<0.001	−0.10	(−0.11, −0.09)	<0.001	−0.13	(−0.14, −0.12)	<0.001	−0.13	(−0.14, −0.12)	<0.001
Estimated trajectories †
Initial cognitive score
No epilepsy	89.8	(89.6, 90.0)	ref	89.8	(89.6, 90.1)	ref	37.1	(36.8, 37.4)	ref	37.2	(36.9, 37.5)	ref
Prevalent epilepsy	89.1	(88.1, 90.2)	0.233	89.5	(88.5, 90.6)	0.584	35.2	(33.8, 36.6)	0.011	35.6	(34.1, 37.1)	0.033
Incident epilepsy	92.1	(87.8, 96.4)	0.278	92.3	(88.1, 96.5)	0.256	33.5	(29.0, 38.0)	0.117	33.6	(29.1, 38.1)	0.117
Average annual decline from initial score through year 4
No epilepsy	0.4	(0.3, 0.4)	ref	0.3	(0.3, 0.4)	ref	0.0	(−0.0, 0.1)	ref	0.0	(−0.0, 0.1)	ref
Prevalent epilepsy	1.1	(0.8, 1.5)	<0.001	1.1	(0.8, 1.5)	<0.001	0.4	(0.2, 0.6)	0.001	0.3	(0.1, 0.5)	0.013
Incident epilepsy	2.1	(1.1, 3.0)	0.002	1.4	(0.4, 2.3)	0.078	−0.1	(−1.1, 0.8)	0.261	−0.1	(−1.1, 0.8)	0.268
Average annual decline from year 4 through end of follow-up
No epilepsy	1.2	(1.1, 1.3)	ref	1.1	(1.0, 1.2)	ref	1.2	(1.1, 1.2)	ref	1.2	(1.1, 1.2)	ref
Prevalent epilepsy	2.0	(1.7, 2.4)	<0.001	1.9	(1.6, 2.2)	<0.001	1.5	(1.3, 1.7)	0.001	1.4	(1.2, 1.6)	0.013
Incident epilepsy	2.7	(1.8, 3.6)	0.002	1.9	(1.0, 2.9)	0.078	0.6	(−0.4, 1.6)	0.261	0.6	(−0.4, 1.6)	0.268

^*Regression coefficients: Linear mixed models are E(Cognitive score) = β₀ (intercept) + β₁ × prevalent epilepsy + β₂ × incident epilepsy + β₃ × year + β₄ × year × prevalent epilepsy + β₅ × year × incident epilepsy + β₆ × year squared + β_i × adjustment variables + random effects. Each P value is from testing the null hypothesis that the coefficient = 0 in the population.

^†Estimated trajectories: For no epilepsy, initial cognitive score is estimated as the intercept (β₀). Average annual decline in cognitive score is estimated using coefficients for year and year squared (β₃ and β₆). For prevalent epilepsy, initial cognitive score is estimated using the intercept and coefficient for prevalent epilepsy (β₀ and β₁). Average annual decline in cognitive score is estimated using coefficients for year, year × prevalent epilepsy, and year squared (β₃, β₄, and β₆). Each P value is from testing the null hypotheses that initial score or average annual decline for prevalent epilepsy equals that for no epilepsy in the population (reference). For incident epilepsy, cognitive test scores were first obtained after incident epilepsy onset in analysis year 2 for 3MS analysis or in analysis year 3 for DSST. Therefore, initial cognitive score is estimated at analysis year 2 (for 3MS analysis) or at analysis year 3 (for DSST analysis) using the intercept and coefficients for incident epilepsy, year, year × incident epilepsy, and year squared (β₀, β₂, β₃, β₅, and β₆). Average annual decline in cognitive score is estimated using coefficients for year, year × incident epilepsy, and year squared (β₃, β₅, and β₆). Each P value is from testing the null hypotheses that initial score or average annual decline for incident epilepsy equals that for no epilepsy in the population (reference). End of follow-up was analysis year 8 for 3MS analysis and analysis year 9 for DSST analysis.

^‡Model 1: Adjusted for baseline values of age, sex, race, education, smoking, alcohol use, body mass index, waist circumference, hypertension, systolic blood pressure, triglycerides, HDL cholesterol, diabetes, coronary heart disease, atrial fibrillation, heart failure, stroke, chronic kidney disease, depressive symptom score, and self-rated health. For 3MS analysis, 331 of 5,672 eligible participants were excluded due to missing values of adjustment variables, leaving n = 5,341 for analysis. For DSST analysis, 463 of eligible 5,790 participants were excluded due to missing values of adjustment variables, leaving n = 5,327 for analysis. Model 2: Adjusted for Model 1 variables plus ApoE4 genotype. For 3MS analysis, an additional 547 participants were excluded due to missing ApoE4 genotype, leaving 4,794 for analysis. For DSST analysis, an additional 547 participants were excluded due to missing ApoE4 genotype, leaving 4,780 for analysis.

Role of ApoE4 allele.

When we further adjusted for ApoE4 allele status (Table 2, Model 2), the significant relationship of prevalent epilepsy with cognitive decline found in Model 1 remained unchanged, suggesting that ApoE4 allele status did not confound the prevalent-epilepsy-cognition relationship. However, for incident epilepsy, the average rate of decline per year in Model 2 was smaller than that found in Model 1, indicating that ApoE4 status confounded the incident-epilepsy-cognition relationship in our cohort (Table 2).

When we investigated statistical interaction of ApoE4 with epilepsy, we observed a significant interaction with prevalent epilepsy (p < 0.001) but not with incident epilepsy (p = 0.613). As shown in Figure 2 and Table 3, the association of prevalent epilepsy with faster decline in 3MS was stronger in the presence of ApoE4 than in the absence of ApoE4. From baseline through year 4, in the absence of ApoE4, average annual decline in 3MS was 0.6 points in prevalent epilepsy compared with 0.3 points in no epilepsy, a difference of 0.3 points per year. However, in the presence of ApoE4, average annual decline in 3MS was 2.5 points in prevalent epilepsy compared with 0.6 points in no epilepsy, a difference of 1.9 points per year. This pattern of interaction also applied to the second half of follow-up with faster rates of decline. The regression coefficient for the year × prevalent epilepsy × ApoE4 interaction term was −1.51 (95% CI −2.26, −0.75; P < 0.001), meaning that the joint presence of prevalent epilepsy and ApoE4 was synergistic, leading to 1.51 points faster annual decline in 3MS score (95% CI: 0.75 to 2.26 points faster) than would be expected if prevalent epilepsy and ApoE4 did not interact.

Figure 2.

Estimated trajectories of mean 3MS score by epilepsy status and ApoE4 genotype

Trajectories shown in the graph are derived from the linear mixed model E(3MS score) = β₀ (intercept) + β₁ × prevalent epilepsy + β₂ × ApoE4 genotype + β₃ × year + β₄ × year × prevalent epilepsy + β₅ × year × ApoE4 genotype + β₆ × year × prevalent epilepsy × ApoE4 genotype + β₇ × year squared + β_i × adjustment variables + random effects. Adjustment variables were baseline values of age, sex, race, education, smoking, alcohol use, body mass index, waist circumference, hypertension, systolic blood pressure, triglycerides, HDL cholesterol, diabetes, coronary heart disease, atrial fibrillation, heart failure, stroke, chronic kidney disease, depressive symptom score, and self-rated health. The regression coefficient β₆ for the 3-way interaction term year × prevalent epilepsy × ApoE4 genotype was −1.51 (95% CI: −2.26, −0.75; P < 0.001). This indicates that the joint presence of prevalent epilepsy and ApoE4 was associated with 1.51 points faster annual decline in 3MS score than would be expected if there were no interaction.

Table 3.

Estimated trajectories of mean 3MS score by epilepsy status and ApoE4 genotype status

	ApoE4 absent		ApoE4 present
	Est	95% CI	Est	95% CI
Estimated trajectories *
Average annual decline from initial score through year 4
No epilepsy	0.3	(0.2, 0.3)	0.6	(0.5, 0.7)
Prevalent epilepsy	0.6	(0.2, 1.0)	2.5	(1.8, 3.1)
Average annual decline from year 4 through end of follow-up
No epilepsy	1.0	(0.9, 1.1)	1.4	(1.2, 1.5)
Prevalent epilepsy	1.4	(1.0, 1.8)	3.2	(2.6, 3.9)

^*Linear mixed model is E(3MS score) = β₀ (intercept) + β₁ × prevalent epilepsy + β₂ × ApoE4 genotype + β₃ × year + β₄ × year × prevalent epilepsy + β₅ × year × ApoE4 genotype + β₆ × year × prevalent epilepsy × ApoE4 genotype + β₇ × year squared + β_i × adjustment variables + random effects. Adjustment variables include baseline values of age, sex, race, education, smoking, alcohol use, body mass index, waist circumference, hypertension, systolic blood pressure, triglycerides, HDL cholesterol, diabetes, coronary heart disease, atrial fibrillation, heart failure, stroke, chronic kidney disease, depressive symptom score, and self-rated health. To interpret the interaction of ApoE4 genotype with prevalent epilepsy, consider the following regression coefficients: First, the regression coefficient β₄ for the 2-way interaction term year × prevalent epilepsy was −0.34 (95% CI: −0.74, 0.06; P = 0.100), indicating that in the absence of ApoE4, annual decline in 3MS score was 0.34 points faster in prevalent epilepsy versus no epilepsy. Second, the regression coefficient β₅ for the 2-way interaction term year × ApoE4 genotype was −0.36 (95% CI: −0.51, −0.21; P < 0.001), indicating that in the absence of prevalent epilepsy, annual decline in 3MS score was 0.36 points faster with ApoE4 present versus absent. Finally, the regression coefficient β₆ for the 3-way interaction term year × prevalent epilepsy × ApoE4 genotype was −1.51 (95% CI: −2.26, −0.75; P < 0.001). This indicates that the joint presence of prevalent epilepsy and ApoE4 was associated with 1.51 points faster annual decline in 3MS score than would be expected from simply summing the association of prevalent epilepsy with annual decline in 3MS score (β₄) with the association of ApoE4 with annual decline in 3MS score (β₅). Because β₆ < 0, we conclude that the association of prevalent epilepsy with annual decline in 3MS score was significantly stronger in the presence of ApoE4. In other words, prevalent epilepsy and ApoE4 combined synergistically to predict a significantly faster rate of decline in 3MS score than would be expected if prevalent epilepsy and ApoE4 had independent relationships with the rate of decline in 3MS score.

Role of clinical stroke.

When we excluded participants with prior history of clinical stroke at baseline and censored 3MS scores obtained after incident stroke during follow-up, the associations of prevalent and incident epilepsy with rates of cognitive decline were similar to the main analysis but attenuated (Table 4). These results suggest that clinical stroke partially, but not fully, accounts for the faster cognitive decline we observed in epilepsy.

Table 4.

Regression coefficients and estimated trajectories of mean 3MS and DSST scores by epilepsy status in absence of clinical stroke

	Models estimating trajectory of mean 3MS score‡						Models estimating trajectories of mean DSST score‡
	Model 1 (n = 5,092)			Model 2 (n = 4,572)			Model 1 (n = 5,122)			Model 2 (n = 4,596)
	Est	95% CI	P	Est	95% CI	P	Est	95% CI	P	Est	95% CI	P
Regression coefficients *
β0: Intercept (no epilepsy, year 0)	89.9	(89.7, 90.1)	<0.001	90.0	(89.8, 90.2)	<0.001	37.3	(37.0, 37.6)	<0.001	37.4	(37.1, 37.7)	<0.001
β1: Prevalent epilepsy	−0.92	(−1.94, 0.09)	0.075	−0.62	(−1.68, 0.45)	0.256	−2.34	(−3.82, −0.87)	0.002	−2.04	(−3.58, −0.51)	0.009
β2: Incident epilepsy	4.75	(−1.07, 10.6)	0.110	3.90	(−1.79, 9.59)	0.179	−4.39	(−11.9, 3.09)	0.250	−4.34	(−11.8, 3.15)	0.256
β3: Year	0.13	(0.04, 0.21)	0.003	0.11	(0.02, 0.20)	0.013	0.49	(0.41, 0.57)	<0.001	0.50	(0.41, 0.58)	<0.001
β4: Year × prevalent epilepsy	−0.63	(−0.94, −0.32)	<0.001	−0.60	(−0.91, −0.28)	<0.001	−0.19	(−0.39, 0.01)	0.060	−0.13	(−0.34, 0.07)	0.195
β5: Year × incident epilepsy	−0.92	(−1.94, 0.11)	0.079	−0.47	(−1.50, 0.56)	0.368	0.51	(−0.58, 1.60)	0.361	0.49	(−0.60, 1.58)	0.381
β6: Year squared	−0.10	(−0.11, −0.09)	<0.001	−0.09	(−0.10, −0.07)	<0.001	−0.12	(−0.13, −0.11)	<0.001	−0.12	(−0.13, −0.11)	<0.001
Estimated trajectories †
Initial cognitive score
No epilepsy	89.9	(89.7, 90.1)	ref	90.0	(89.8, 90.2)	ref	37.3	(37.0, 37.6)	ref	37.4	(37.1, 37.7)	ref
Prevalent epilepsy	89.0	(88.0, 90.0)	0.075	89.4	(88.3, 90.4)	0.256	34.9	(33.5, 36.4)	0.002	35.4	(33.8, 36.9)	0.009
Incident epilepsy	92.7	(88.4, 96.9)	0.198	92.8	(88.7, 96.9)	0.180	34.8	(30.0, 39.5)	0.302	34.9	(30.2, 39.7)	0.304
Average annual decline from initial score through year 4
No epilepsy	0.3	(0.2, 0.3)	ref	0.2	(0.2, 0.3)	ref	0.0	(−0.0, 0.1)	ref	−0.0	(−0.1, 0.0)	ref
Prevalent epilepsy	0.9	(0.6, 1.2)	<0.001	0.8	(0.5, 1.1)	<0.001	0.2	(−0.0, 0.4)	0.060	0.1	(−0.1, 0.3)	0.195
Incident epilepsy	1.4	(0.4, 2.4)	0.079	0.9	(−0.2, 1.9)	0.368	−0.1	(−1.2, 1.0)	0.361	−0.1	(−1.2, 1.0)	0.381
Average annual decline from year 4 through end of follow-up
No epilepsy	1.0	(0.9, 1.1)	ref	0.9	(0.8, 1.0)	ref	1.1	(1.1, 1.2)	ref	1.1	(1.0, 1.1)	ref
Prevalent epilepsy	1.7	(1.4, 2.0)	<0.001	1.5	(1.2, 1.8)	<0.001	1.3	(1.1, 1.5)	0.060	1.2	(1.0, 1.4)	0.195
Incident epilepsy	2.0	(0.9, 3.0)	0.079	1.4	(0.4, 2.4)	0.368	0.6	(−0.5, 1.7)	0.361	0.6	(−0.5, 1.7)	0.381

^*Regression coefficients: Linear mixed models are E(Cognitive score) = β₀ (intercept) + β₁ × prevalent epilepsy + β₂ × incident epilepsy + β₃ × year + β₄ × year × prevalent epilepsy + β₅ × year × incident epilepsy + β₆ × year squared + β_i × adjustment variables + random effects. Each P value is from testing the null hypothesis that the coefficient = 0 in the population.

^†Estimated trajectories: For no epilepsy, initial cognitive score is estimated as the intercept (β₀). Average annual decline in cognitive score is estimated using coefficients for year and year squared (β₃ and β₆). For prevalent epilepsy, initial cognitive score is estimated using the intercept and coefficient for prevalent epilepsy (β₀ and β₁). Average annual decline in cognitive score is estimated using coefficients for year, year × prevalent epilepsy, and year squared (β₃, β₄, and β₆). Each P value is from testing the null hypotheses that initial score or average annual decline for prevalent epilepsy equals that for no epilepsy in the population (reference). For incident epilepsy, cognitive test scores were first obtained after incident epilepsy onset in analysis year 2 for 3MS analysis or in analysis year 3 for DSST. Therefore, initial cognitive score is estimated at analysis year 2 (for 3MS analysis) or at analysis year 3 (for DSST analysis) using the intercept and coefficients for incident epilepsy, year, year × incident epilepsy, and year squared (β₀, β₂, β₃, β₅, and β₆). Average annual decline in cognitive score is estimated using coefficients for year, year × incident epilepsy, and year squared (β₃, β₅, and β₆). Each P value is from testing the null hypotheses that initial score or average annual decline for incident epilepsy equals that for no epilepsy in the population (reference). End of follow-up was analysis year 8 for 3MS analysis and analysis year 9 for DSST analysis.

^‡Model 1: Adjusted for baseline values of age, sex, race, education, smoking, alcohol use, body mass index, waist circumference, hypertension, systolic blood pressure, triglycerides, HDL cholesterol, diabetes, coronary heart disease, atrial fibrillation, heart failure, chronic kidney disease, depressive symptom score, and self-rated health. For 3MS analysis, 331 of 5,672 eligible participants were excluded due to missing values of adjustment variables, and an additional 249 were excluded due to prevalent stroke or incident stroke prior to contributing a 3MS score, leaving n = 5,092 for analysis. For DSST analysis, 463 of eligible 5,790 participants were excluded due to missing values of adjustment variables, and an additional 205 were excluded due to prevalent stroke or incident stroke prior to contributing a DSST score, leaving n = 5,122 for analysis. Model 2: Adjusted for Model 1 variables plus ApoE4 genotype. For 3MS analysis, an additional 520 participants were excluded due to missing ApoE4 genotype, leaving 4,572 for analysis. For DSST analysis, an additional 526 participants were excluded due to missing ApoE4 genotype, leaving 4,596 for analysis.

DSST

Main analysis.

We estimated longitudinal DSST trajectories over a mean of 7.2 annual DSST administrations. The rate of decline in information processing accelerated over time for all three groups (Figure 1, Panel B). The rate of decline was significantly faster in prevalent epilepsy compared with no epilepsy (p = 0.001). However, the rate of decline after incident epilepsy was not significantly different than the rate of decline without epilepsy (p = 0.261; Table 2). Mean DSST score in those without epilepsy changed minimally through year 4 (from mean initial score of 37.1 to mean of 36.9 at year 4), then declined 1.2 points per year on average from year 4 to year 9 of follow-up (from mean of 36.9 to 31.0), as seen in Table 2, Model 1. In prevalent epilepsy, initial DSST score was significantly lower than initial score without epilepsy, and the rate of decline was marginally steeper, with mean DSST score declining 0.4 points per year on average through year 4 (from mean initial score of 35.2 to mean of 33.8 at year 4), and then declining 1.5 points per year on average from year 4 to year 9 (from mean score of 33.8 to 26.2). After incident epilepsy, mean DSST score remained stable through year 4, and then declined 0.6 points per year on average from year 4 to year 9 (from mean of 33.6 to 30.5). Confidence intervals suggest that in the population, older adults with prevalent epilepsy may decline 0.13 to 0.52 points faster in mean DSST score per year than older adults without epilepsy (Table 2).

Role of ApoE4 allele.

The significant association of prevalent epilepsy with faster decline in DSST found in Model 1 remained significant and similar in magnitude in Model 2, suggesting that ApoE4 status did not confound the relationship (Table 2). We did not observe a significant interaction of ApoE4 with prevalent epilepsy (p = 0.342) or with incident epilepsy (p = 0.907).

Role of clinical stroke.

When we modeled cognitive decline in the absence of clinical stroke, initial mean DSST score was still significantly lower in prevalent epilepsy than without epilepsy, however the rate of decline in DSST was no longer significantly different in prevalent epilepsy compared with no epilepsy (Table 3).

DISCUSSION

In this longitudinal study of older adults, global cognitive function declined more rapidly in individuals with prevalent epilepsy (onset prior to age 65) and incident epilepsy (onset at age 65 or older) compared to those without epilepsy, even after adjusting for risk factors associated with cognitive decline. When we adjusted for ApoE4 status, associations of prevalent and incident epilepsy with decline in 3MS were attenuated. We found that ApoE4 status modified the relationship of prevalent epilepsy with 3MS such that epilepsy patients with an ApoE4 allele had more rapid cognitive decline than would be expected if epilepsy and ApoE4 acted independently. In secondary models excluding participants with prior clinical stroke and censoring data after new clinical stroke, decline in 3MS remained significantly faster in participants with prevalent epilepsy than in those without epilepsy. For incident epilepsy, however, decline in 3MS was no longer significantly different than in those without epilepsy.

Information processing speed declined significantly faster in prevalent epilepsy compared with no epilepsy. However, no significant difference in decline of DSST score was found between incident epilepsy and no epilepsy. In the absence of clinical stroke, decline in information processing speed was not significantly different across the three groups.

Cognitive difficulty ranks highest on a list of potential concerns among epilepsy patients.² However, few studies have examined cognitive outcomes in older adults with epilepsy. Five cross-sectional studies and a longitudinal study tested mostly small samples of epilepsy patients with earlier onset,^{9–12, 27, 28} with only one cross-sectional study examining incident epilepsy in older adults.²⁸ In these six studies, older adults with epilepsy were found to have greater deficits compared with those without epilepsy across multiple cognitive domains, including short- and long-term visual and verbal memory, executive functions, attention, and processing speed. The cross-sectional study of incident epilepsy found that 43% of patients had markedly adverse executive function before epilepsy treatment was even begun.²⁸

Underlying mechanisms that hasten cognitive decline in older adults with prevalent epilepsy may differ from those with incident epilepsy first occurring in late life. In our study, data on duration of epilepsy was not available in individuals with prevalent epilepsy. Accelerated cognitive decline among people with long-standing epilepsy may be due to cumulative effects of years of seizures²⁹ or interictal epileptiform potentials.³⁰ In addition, poor cognitive performances in older adults with epilepsy is correlated with polytherapy of antiepileptic medications.¹¹ As highlighted by Sen et al.,¹⁵ there are few studies in epilepsy that have addressed the relationship of comorbidities such as vascular risk factors to cognitive course. Identification of modifiable vascular risk factors associated with cognitive decline in epilepsy would not only inform mechanisms of cognitive decline but may provide therapeutic opportunities to slow cognitive decline.¹⁵

As stroke is the most common cause of incident epilepsy in older adults³¹ and poststroke patients have faster decline in cognitive function over long-term,³² cognitive decline would likely accelerate in those who had a stroke prior to epilepsy. We found that the faster rate of decline in 3MS in the incident epilepsy group was no longer significant when participants with stroke were removed. However, given the small size of the incident epilepsy group, we had low statistical power to demonstrate a difference in cognitive decline. Mechanisms for accelerated cognitive decline in poststroke patients are thought to be (1) unmasking or exacerbation of neurodegenerative disease,³³ and (2) immune response to stroke causing oxidative stress, hippocampal dysfunction, and ongoing cerebrovascular injury.³⁴ In prevalent epilepsy, however, more rapid decline in 3MS remained significant even when participants with stroke were removed, suggesting a possibility for a greater degree of negative cognitive impact from epilepsy burden in this group.

Dementia is the second most common risk factor for epilepsy in older adults. Among a cohort of Alzheimer’s disease (AD) patients in the UK, risk of incident seizure was five times higher among AD patients compared with a non-AD group.³⁵ more people have dementia. Faster decline in global cognition in the incident epilepsy group that we observed, when adjusted for demographics and vascular risk factors, raises the question of whether more patients in our incident epilepsy group already had dementia. When we further adjusted for ApoE4 genotype, a strong marker for dementia, we no longer observed a significant relationship between incident epilepsy and rate of cognitive decline. We did not censor data after dementia, as we did for stroke, because adjudicated dementia diagnosis was performed only in a subgroup of CHS participants. Possibly some with incident epilepsy could have developed epilepsy during the prodromal stage of dementia, as reported previously.³⁶ A combined analysis of eight case-control studies showed that epilepsy was more common among cases of AD in the years preceding the onset of AD.³⁷ Another possible explanation for the accelerated decline of global cognition in the incident epilepsy group could be that some participants had vascular dementia, because the association was no longer significant when individuals with prior stroke were excluded.

One of the most salient findings in our study was the role of ApoE4 allele status as a significant effect modifier in the relationship between cognitive decline and prevalent epilepsy. Older adults with prevalent epilepsy who had ApoE4 had much more rapid decline than older adults with prevalent epilepsy without ApoE4 or older adults without epilepsy who had ApoE4 allele. Significant interaction between ApoE4 status and epilepsy duration has been shown in studies of patients with temporal lobe epilepsy, such that E4 carriers with a long duration of epilepsy had the poorest memory performance.³⁸ For example, in one study, patients with temporal lobe epilepsy with E4 allele and epilepsy duration ≥25.5 years had the highest risk (OR, 32.29; 95% CI, 5.23–195.72) of developing verbal memory impairment in comparison with those without E4 allele and shorter duration of epilepsy (<4 years).³⁹ Additionally, adults with childhood-onset epilepsy who were ApoE4 carriers seem to have a higher burden of brain amyloid as measured with position emission tomography at late middle age, when compared to matched controls, suggesting a relationship between ApoE4 genotype, amyloid pathology, and epilepsy.⁴⁰ These findings raise additional questions such as usefulness of screening epilepsy patients for ApoE4 status and whether intense seizure control may guard against accelerated cognitive decline.

Our study had several strengths, notably inclusion of longitudinal cognitive data assessed prospectively at a population-wide level. CHS also employed rigorous methods of adjudicating stroke cases. Cognitive assessments were repeated over 10 years. We were also able to adjust for rigorously measured risk factors of abnormal cognitive aging such as vascular, lifestyle factors, and ApoE4.

Our study also had important limitations. First, CHS was designed to examine risks of cardiovascular disease in older adults. Therefore, we did not have data on seizures types, duration of epilepsy in participants with prevalent epilepsy, or seizure control, which potentially affect cognitive trajectory. Additionally, data used in our analyses were collected from 1989 to 1999, when clinical examinations and cognitive evaluations were repeated at annual CHS clinic visits, and thus may not be representative of modern populations, potentially limiting generalizability. Second, despite our large sample, the number of incident epilepsy cases who had cognitive measures assessed during the first ten years of the study was small, likely limiting power. Third, the instruments used in this study to assess cognitive changes are screening instruments. 3MS has discriminant ability for dementia²⁶ and even mild cognitive impairment from healthy aging.⁴¹ However, comprehensive neuropsychological evaluation is the gold standard for identifying cognitive decline. Fourth, we did not assess separate effects of epilepsy and AED on cognitive outcome. Since cognitive assessments were made between 1989 and 1999, older medications such as phenobarbital, phenytoin, and carbamazepine were likely to have been used. While older antiepileptic medications adversely affect cognition relative to nondrug conditions,^{42, 43} it is unclear whether chronic use of them would lead to more rapid decline of cognition over time with steeper slope than age-related decline. The association we observed could be due to the combined effects of epilepsy and AED. Finally, epilepsy case ascertainment relied on claims data and consequently some cases may have been misclassified.

In conclusion, prevalent epilepsy was associated with faster decline in global cognitive ability, especially in the presence of ApoE4, and with lower initial score and faster decline in processing speed. Incident epilepsy was associated with faster subsequent decline in global cognitive ability, although the decline became attenuated when adjusting for ApoE4. Association between global cognition and incident epilepsy was no longer significant after excluding participants with prior clinical stroke and censoring data after new clinical stroke during follow-up, possibly due to the limited sample size of incident epilepsy. Biological mechanisms to explain these association, and clinical implications of screening for ApoE4 status in individuals with epilepsy of early onset should be explored in future studies.

Key Point Box.

• Global cognitive function declined more rapidly in prevalent epilepsy and incident epilepsy compared with no epilepsy.

• Information processing speed declined more rapidly in prevalent epilepsy compared with no epilepsy.

• epilepsy patients with an ApoE4 allele had more rapid cognitive decline than would be expected if epilepsy and ApoE4 acted independently.