Epidemiology of Functional Seizures Among Adults Treated at a University Hospital

Key Points Question What is the period prevalence and comorbidity of patients with functional seizures? Findings This case-control study including 3341 patients with functional seizures (period prevalence, 0.14%) identified from all patients at a university medical center found that posttraumatic stress disorder and sexual assault trauma were associated with functional seizures. Novel associations were identified, including cerebrovascular disease. Meaning The findings of this study suggest that patients who present with seizures and additional psychiatric comorbidities, sexual assault trauma history, or cerebrovascular disease should be referred for video electroencephalogram diagnostic assessment.

This supplemental material has been provided by the authors to give readers additional information about their work.eAppendix 1. Methods

Functional Seizures case/control algorithm development
Both algorithms were created by filtering criteria joined with Boolean operators.First, we required one or more convulsion or conversion disorder (ICD9 codes 300.11OR 780.39 OR ICD10 codes R56.9 OR F44.5) 1 AND the presence of regular expressions indicating functional seizures-related keywords in charts ("psychogenic nonepileptic," "pseudoseizure," "psychogenic seizure," "nonepileptic seizure", or " pnes ").Although functional seizures can present as syncopal events, we have not included this phrase in our algorithm because in patient charts such as the ones we mined, this would be a nonspecific symptom and would add noise to our algorithm.Next we required the presence of both the keyword "EEG" AND one or more EEG CPT code (Group 95812-95830: Routine Electroencephalography (EEG) Procedures and Group 95950-95967: Special EEG Testing Procedures) to ensure that patients had an EEG performed at VUMC.Eighty-two percent of potential functional seizures cases with an "EEG" keyword also had an EEG procedure CPT code present in their chart.

Chart Review
Fifty charts were randomly selected from the functional seizures algorithm identified cases and were reviewed by one rater (SBG), who was trained by a clinical neurologist (KFH).
The charts were reviewed for positive clinical diagnosis of functional seizures based on a publication by LaFrance et al describing the minimum requirements for the diagnosis of functional seizures. 2Positive predictive value (PPV) was calculated for two groups based on this paper: (a) documented, clinically established, or probable functional seizures; and (b) possible functional seizures.Based on this paper, the first criteria for a positive diagnosis of documented, clinically established, or probable functional seizures was defined as having chart history characteristics consistent with functional seizures.The second criteria was having a seizure witnessed by a clinician who either 1) reviewed the video EEG recording and found the semiology typical of functional seizures, or 2) is experienced in diagnosis of seizure disorders and found the seizure to show semiology typical of functional seizures, regardless of whether or not it was on vEEG.The third criteria was that there was no epileptiform activity in routine or sleep-deprived interictal EEG, in routine or ambulatory ictal EEG during a typical event, or immediately before, during or after ictus captured on ictal video EEG.A positive diagnosis for possible functional seizures was defined as having a history consistent with functional seizures, event matched functional seizures semiology by witness report or self-report/description, and that there was no epileptiform activity in routine or sleep-deprived interictal EEG.ƒ11 KFH was consulted as needed to adjudicate.
Prior to review of these 50 charts, SG reviewed 100 charts, and KH also reviewed 37 of these charts.The goal of this first round of chart review was to train the reviewer (SG) to identify key phrases and language used by VUMC neurologists to interpret video EEG and to confirm diagnoses.SG, KH, and LKD sat in conference for approximately 4 hours to discuss the levels of evidence observed and the certainty of diagnosis given video EEG results.
For the sexual assault trauma analysis, AL manually reviewed 52 charts to identify the presence of confirmatory statements describing a history or current experience of sexual assault trauma.
Standard error (SE) and confidence intervals (CI) for the positive predictive value of each algorithm was calculated using the following formulas, where TP = number true positive, FP = number false positive:  =  * (1 − )  +   =  ± 1.96 *

Extraction of sexual assault trauma from the EHR
While it is likely to be under-reported at the time of the assault, a history of sexual assault can be extracted from the medical record.Based on an initial review of 25 charts, which were randomly selected, we identified eight inclusion phrases (i.e., "his/her sexual assault", "history of sexual abuse", "history of rape", etc.) and four exclusion phrases (i.e., "denies history of sexual abuse", "no hx of sexual assault", etc.) to identify cases (Supplementary Table 5).Although ICD codes do exist for sexual assault trauma, they were infrequently used.The primary mode of identification of patients who experienced sexual assault trauma in the EHR was through the identification of regular expressions (e.g., "history of sexual abuse", "was raped") within clinical notes.
Because sex is randomly determined at conception, the assumption of random assignment to the "treatment" condition holds, thus the patient sex (male/female) was coded as the "treatment" variable.Sexual assault trauma was then coded as the mediator and functional seizures as the outcome in a model-based mediation analysis. 3We used 1,000 bootstrap comparisons to determine empirical confidence intervals and determine statistical significance of the mediation model.Additionally, we ran the mediation analysis with four different randomly generated seeds to ensure that the results and empirical p-values of the analysis remained stable across all permutations.

Development of Medical Home Population
The total study population used for algorithm development and prevalence calculations included the entire hospital population (N = 2,346,808).However, the phenome-wide association study was restricted to a subset of this total population called the "medical home" population.The medical home is a heuristic definition that restricts the total sample to a subpopulation with more complete medical record data to facilitate the comparison of comorbidity patterns between cases and controls.The definition applied required the presence of at least 5 codes over a period of at least 3 years.The reason for this restriction is to ameliorate the impact of missing data in controls which can result in a biased upward estimate in the regression coefficients in a case vs. control comparison in the PheWAS.

PheWAS of Function seizures compared to epilepsy cases
We compared algorithm identified functional seizures cases (N = 1,431) to epilepsy cases (4,715).Epilepsy cases for this analysis were defined as those that had at least three ICD codes for either generalized (GE) (FE) or focal epilepsy as previously defined. 4Finally, we repeated the analysis a final time and compared epilepsy cases (4,175)  to epilepsy controls (n = 496,890).Epilepsy controls were defined as all other patients, excluding functional seizures cases, functional seizures patients meeting exclusion criteria for functional seizures, and any patients with either one or two GE or FE ICD codes.

Functional seizures comorbidity sex-differences analysis
We performed a phenome-wide functional seizures by sex interaction analysis to determine whether comorbidities were more common among males or females with functional seizures after accounting for any baseline sexdifference in the prevalence of the comorbidity.Phenotypes that exceeded a Bonferroni corrected interaction p-value (p < 3.02E-05) demonstrated a significant diagnosis by sex interaction.

Functional seizures date of diagnosis analyses
For each patient identified as a functional seizures case, we established the date of the first clinical suspicion of functional seizures, defined as the first mention of a functional seizures keyword in clinical notes.We used the date of EEG administration (with a functional seizures keyword appearing in the patient's chart within 30 days) to define the date of the diagnosis of functional seizures.We then determined the average time from the first mention to the diagnosis for algorithm defined functional seizures cases.Furthermore, during the manual chart review we documented patient-reported onset of seizures.

Cerebrovascular disease identification
Date of diagnosis of cerebrovascular disease was identified using the date of the first ICD code of a list of ICD9 codes that were found to be significantly associated with functional seizures and epilepsy (all of which overlapped).The ICD9 code list was as follows: 430, 433, 433.8, 430.2, 433.2, 433.21, 433.3, 433.31, 430.3, 433.6, 430.1, 433.1,  433.5, 433.11, 433.12.

Calculation of BMI from the EHR
Body mass index (BMI) data was obtained from the EHR.All individuals had multiple BMI measurements from various visits to the VUMC.BMI in the EHR is prone to recording errors. 5To remove erroneous BMI records, we first z-score scaled the longitudinal BMI measurements collected within each individual, then removed any measurement with a Z-score below -3 or above 3. Next, we calculated median BMI value for each individual.
Using the cleaned median BMI values for each individual with available data, we calculated the mean and SE of BMI for functional seizures cases without epilepsy as defined by our algorithm (n = 1,605), for functional seizures cases with epilepsy as defined by our second algorithm (n = 1,058), and for controls (n = 488,398) as defined by our algorithm.We also calculated the mean and SE BMI value for epileptic patients (n = 6,186) for comparison and replication of prior studies. 6Epileptic patients were defined as anyone who had either an ICD10 code from group G40 or an ICD9 code from group 345 for either generalized or focal epilepsy, as previously reported. 1,4We used the R package ggplot2 for plotting, and a one-way ANOVA was performed in R followed by a Tukey post-hoc analysis to determine statistical significance of the difference in BMI between each group.
It is also worth noting that while all of our analyses were performed separately with and without median BMI included as a covariate.However, results for analyses excluding the median BMI covariate are not shown as they were very similar to the analyses including median BMI as a covariate.eAppendix 2. Results

EEG keyword presence in charts and CPT code presence are strongly correlated
Across the entire VUMC-EHR, 66,936 patients had an EEG CPT code, 109,523 patients had "EEG" written in their charts, and 62,249 patients had both an EEG CPT code and keyword (93% of patients with an EEG CPT code and 57% of patients with an EEG keyword).

Mental, neurological, and cerebrovascular disorders are associated with epilepsy
We also performed a PheWAS analysis between epilepsy cases and epilepsy controls, which showed 47 out of 71 mental disorder codes were significantly associated with epilepsy, as well as 60 out of 82 neurological Phecodes (Supplementary Figure 3).We also saw that 35 associations with circulatory system codes were significant, several of which were among the highest associations with epilepsy.
Additionally, PheWAS comparing comorbidity patterns between functional seizures cases and epilepsy cases showed that 72% of the psychiatric, 42% of the neurological and 20% of the circulatory system phenotypes that were associated with functional seizures, remained associated with functional seizures even when the comparison group was limited to epilepsy cases (Figure 1B).

Comorbidity patterns for functional seizures do not differ by sex
There were no significant differences between males and females in the prevalence of each functional seizures comorbidity.

BMI is significantly higher in functional seizures patients than controls
One-way ANOVA and a post-hoc Tukey test showed that functional seizures cases without concurrent epilepsy had significantly higher median BMI measurement than algorithm-defined controls (include result here) or epilepsy patients (include result here) (Supplementary Figure 6).Additionally, a greater proportion of algorithm-defined functional seizures cases without concurrent epilepsy were considered obese (BMI>30) (39%) compared to algorithm-defined controls (33%), epileptic patients (27%), and functional seizures patients with concurrent epilepsy (37%)(Supplementary Patients were exclusively identified using natural language processing of patients' clinical notes using various inclusion phrases and exclusion phrases.The inclusion phrases and exclusion phrases used are both reported here.eTable 6. Functional seizures case status and sex are significantly associated with sexual assault in the VUMC-EHR Age at diagnosis (years) eFigure 6. Functional seizures patients have greater BMIs than the control group and epilepsy patients Average and SE of median BMI are shown for controls as identified by our algorithm, epilepsy patients as identified by ICD codes, functional seizures cases as identified by our algorithm, and functional seizures cases with epilepsy as identified by our second algorithm.All participants analyzed had a median age of records over 18.The sample size and percentage of patients from each group that were considered obese (BMI over 30) is also shown.

eFigure 2 . 3 .
Flow chart of algorithm used to identify functional seizures patients irrespective of their epilepsy case/control status eFigure PheWAS of VUMC EHR ICD-code defined epilepsy cases versus controls.Results are plotted by category of phenotypes, with each category shown in a different color.Only the top 20 associations are labeled to increase visibility of the graph.Additionally, upturned triangles represent positive associations with functional seizures case status, while downturned triangles represent negative associations with functional seizures case status.A horizontal blue line indicates the nominal p-value level 0.05, while a horizontal red line indicates the Bonferroni corrected p-value.eFigure 4. Histogram of the number of years from date of functional seizures diagnosis to cerebrovascular diagnosis in 92 patients who had both CVD and functional seizures Negative values indicate patients who were diagnosed with CVD before functional seizures, while positive values indicate patients who were diagnosed with functional seizures before CVD.# days from PNES dx to 1st cerebro vascular ICD years from functional seizures diagnosis to CVD diagnosis eFigure 5. Temporal analysis of the development of functional seizures and cerebrovascular disease in 92 patients A. B. C. Patients are binned into three groups: A. Patients who developed functional seizures before CVD (n = 27); B. Patients who developed functional seizures and CVD within 3 months of each other (n = 21); and C. Patients

eAppendix 1 .
Methods eAppendix 2. Results eTable 1. Criteria for Algorithm to Identify Functional Seizures Patients, Excluding Those Who Have Concurrent Epilepsy eTable 2. Criteria for Algorithm to Identify Functional Seizures Patients Regardless of Epilepsy Status eTable 3. Demographics for 92 Patients Who Were Diagnosed With Both Cerebrovascular Disease and Functional Seizures eTable 4. Demographics for 735 Patients Who Were Diagnosed With Both Cerebrovascular Disease and Epilepsy eTable 5. Criteria to Identify Sexual Assault Trauma (SAT) Patients eTable 6. Functional Seizures Case Status and Sex Are Significantly Associated With Sexual Assault in the VUMC-EHR eTable 7. Body Mass Index information for Functional Seizures and Epilepsy Cases as Compared to Controls eFigure 1. Functional Seizures Patients Have Longer, More Dense Medical Records as Compared to Controls eFigure 2. Flow Chart of Algorithm Used to Identify Functional Seizures Patients Irrespective of Their Epilepsy Case/Control Status eFigure 3. PheWAS of VUMC EHR ICD-Code Defined Epilepsy Cases Versus Controls eFigure 4. Histogram of the Number of Years from Date of Functional Seizures Diagnosis to Cerebrovascular Diagnosis in 92 Patients Who Had Both CVD and Functional Seizures eFigure 5. Temporal Analysis of the Development of Functional Seizures and Cerebrovascular Disease in 92 Patients eFigure 6. Functional Seizures Patients Have Greater Bmis Than the Control Group and Epilepsy Patients

Table 7 )
. eTable 1. Criteria for algorithm to identify functional seizures patients, excluding those who have concurrent epilepsy Demographics are also shown for three subgroups of these patients: those who developed Epilepsy before CVD (n = 208); those who developed Epilepsy and CVD within three months of each other (n = 153); and those who developed CVD before Epilepsy (n = 374).The proportion of males and females, race, ethnicity, density of records, median ages across the medical record, age at first CVD code, age at first generalized or focal epilepsy code, and percentages of CVD subcategories are shown for each group.ICD codes for cerebral hemorrhage included 430, 430.2, 430.3, and 430.1.ICD codes for cerebral artery occlusion included 433.2, 433.21, 433.1, and 433.11.ICD codes for cerebral ischemia included 433.3 and 433.31.ICD codes for cerebrovascular diseases included 433, 433.8, and 433.6.The ICD code for cerebral aneurysm was 433.5, and the ICD code for cerebral atherosclerosis was 433.12.eTable 5. Criteria to identify sexual assault trauma patients regression functional seizures ~ SAT + sex + SAT*sex was performed to determine whether functional seizures case/control status, sex, or the interaction of functional seizures case status were correlated with sexual assault case status.P-values, odds ratio (OR) and the 95% confidence interval (CI) are presented for each variable.The multivariable logistic regression used included these variables and additional covariates for median age of record, median BMI, race, and the density of records.eTable 7. Body Mass Index information for functional seizures and epilepsy cases as compared to controls Average and SE of median BMI are shown for controls as identified by our algorithm, epilepsy patients as identified by ICD codes, functional seizures cases as identified by our algorithm, and functional seizures cases with epilepsy as identified by our second algorithm.All participants analyzed had a median age of records over 18.The sample size and percentage of patients from each group that were considered obese (BMI over 30) is also shown.eFigure 1. Functional seizures patients have longer, more dense medical records as compared to controls A.B.
A. Density histogram of length of record is plotted in years for functional seizures cases (blue, purple) versus controls (red, green) in both medical home (red, blue) and outside of it (green, purple).Additionally, the median value each group is plotted as a vertical dashed line.B. Density histogram of records per year in functional seizures cases (blue) and controls (red) within the medical home.Median values are plotted as vertical dashed line for both groups.