This 6-month prospective RCT assessed the effects of SMART vs. WL on self-reported NHEs defined as a seizure, ED visit, hospitalization or a self-harm attempt. Adults with epilepsy were enrolled at two participating sites, the University of Iowa and the Case Western Reserve University. Building on the success of a completed efficacy trial, the current RCT was an effectiveness trial which attempted to reach a diverse group of PWE, and which was delivered entirely remotely via internet or phone. The primary study outcome was change from baseline to 6-month follow-up in NHE counts. Secondary outcomes included factors relevant to PWE such as depression, epilepsy-related QOL, functional status, general health, stigma, self-efficacy, epilepsy self-management competency and social support. Assessments were conducted at screening, baseline, 13 weeks follow-up, and 6-month follow-up time-points. A detailed description of the RCT methods has been described elsewhere [7]. The study was registered at ClinicalTrials.gov (NCT04705441).

Study inclusion criteria included a self-reported diagnosis of epilepsy and age ≥ 18 years with at least 1 NHE within the previous 6 months. Given the effectiveness focus, there were only a limited number of study exclusion criteria, which included being at immediate risk of self-harm, dementia, pregnancy, or those unable to read/understand English. The study team prioritized engagement and enrollment of PWE with complex illness due to their living in rural settings and/or having poorly controlled seizures. Participants were recruited using a variety of outreach efforts including reaching out to clinicians at epilepsy centers and soliciting referrals of their patients who might benefit from SMART, solicitation of referrals from local/regional epilepsy social service agencies, outreach via free or paid advertising such as Research Match [6], querying the electronic health records (EHRs) of two medical health systems, use of IRB-approved advertising and in-person outreach. We attempted to oversample PWE who resided in rural settings by filtering EHRs for ZIP codes that aligned with more rural locations on the Rural Urban Continuum Code (RUCC), developed by the U.S. Department of Agriculture classification scheme that distinguishes metropolitan counties by the population size of their metro area, and nonmetropolitan counties by degree of urbanization and adjacency to a metro area [8]. Each county in the U.S. is assigned one of the nine codes with 1 being most urban and 9 being most rural. Clinicians from epilepsy centers tended to refer people who had more frequent seizures and might potentially benefit from SMART. The study was approved by the local institutional review board (IRB) and all individuals provided written informed consent.

Participants were randomly assigned, on a 1:1 basis, to receive either SMART (N = 80) or WL (N = 80) using a computer-generated random number generator. The randomization scheme was unavailable to study staff interacting with participants. While participants and SMART interventionists were not blinded (due to the nature of the intervention), data analysts had no interaction with participants or interventionists. Participants were enrolled in 10 sequential “cohorts” such that once 16 individuals were enrolled (about half randomized to SMART vs. WL) the individuals allocated to SMART in that cohort began their SMART group sessions.

The experimental arm of this RCT delivered the SMART curriculum, divided into 2 phases. Phase 1 comprised 8 group-format 45-60-minute sessions with 5-10 PWE in each group and occurred over a period of 8-12 weeks. Phase 2 comprised a short series of follow-up phone calls. The SMART intervention is an adjunct to standard epilepsy management, including use of antiseizure medications as prescribed by the participant’s regular (non-research) epilepsy care clinician.

Each of the Phase 1 SMART group sessions was jointly delivered by a Nurse Educator and a Peer Educator. Nurse Educators (N=6, 5 women, 1 man) were individuals who were already participating in clinical research at the participating sites, but did not need to have any previous experience in epilepsy care. The Peer Educators (N=6, 2 women, 4 men) were referred by site clinicians or by epilepsy-service organizations. Some PWE were self-referred based on previous experience serving as a SMART Peer Educator. Nurse and Peer Educators participated in an initial 2-day virtual training supplemented with online and written materials.

Phase 1 of SMART was delivered remotely using secure videoconferencing. Telephone call-in was available for those without internet access. SMART group participants received e-mail and phone/text reminders of upcoming sessions, and make-up sessions were available. SMART Phase 2 phone follow-up consisted of 3 brief (no more than 15 minutes) monthly telephone sessions conducted by the Nurse Educator to address ongoing issues of ESM. The nurse also served as a facilitator of linkage between the PWE’s epilepsy care clinicians by providing SMART program status updates to providers.

Attendance of SMART sessions was recorded, and acceptability was assessed at the end of each group series by a brief self-rated survey. Following Fraser [9], and to help ensure fidelity, non-interventionist research staff monitored and randomly assessed 25% of group sessions, with each fidelity dimension rated on a 1-10 scale.

Individuals randomized to WL received standard epilepsy management, including use of antiseizure medications as prescribed by the participant’s regular (non-research) epilepsy care clinician. After they completed their 13-week and 6-month assessments, which were identical to the SMART group assessments, individuals randomized to WL then received the SMART intervention (data not presented here).

Research assessments were conducted remotely using REDCap (Research Electronic Data Capture), a secure web application for building and managing online surveys and databases [10]. We collected sociodemographic information, select clinical variables (epilepsy age of onset/duration, type of epilepsy, somatic comorbidity) and standardized measures of epilepsy outcomes. Health literacy was evaluated using the Rapid Estimate of Adult Literacy in Medicine-Revised (REALM-R) [11,12]. Individuals unable or unwilling to complete REDCap surveys could complete assessments by phone. Individuals were assessed at study screening, baseline, 13-week and 6-month timepoints.

The primary outcome was differences in proportions of PWE with reductions in NHEs from baseline to 6 months follow up (SMART vs. WL), defined as any improvement, ≥10%, ≥20%, ≥30%, ≥40%, and ≥50% improvement. NHEs were defined as seizures, ED visits, hospitalizations, and self-harm attempts identified via self-report. In addition, we assessed counts of NHEs in each category, including seizure frequency, and length of stay in days for hospitalization NHEs. As an a priori validation of self-reported NHEs, we counted ED visit and hospitalization encounters documented in the EHRs of individuals in the primary enrolling site in Iowa (32 out of 108 = 33% total enrolled at this site).

Additional key assessments included evaluation of depression, functioning, epilepsy-related QOL, ESM competence, epilepsy stigma, social support and epilepsy self-efficacy. Depressive symptom severity was assessed using the 9-item Patient Health Questionnaire (PHQ-9), a widely used and validated self-rated depression scale [13]. On the PHQ-9, scores range from 0-27, with higher scores indicating worse depression severity. Functional health status was assessed using the 36-Item Short Form Survey (SF-36), a multi-purpose, short-form health survey that yields two psychometrically-based components: a physical component summary and mental component summary [14]. On each component of the SF-36, scores range from 0 (worst functioning) to 100 (best functioning). Epilepsy-related QOL was assessed with the Quality of Life in Epilepsy (QOLIE-31), a self-rated instrument which contains 7 sub-components including seizure worry, overall QOL, emotional well-being, energy-fatigue, cognitive functioning, medication effect, and social function, each with a range of 0 (lower QOL) to 100 (higher QOL) [15,16]. QOLIE-31 total score can be converted to a T-score with a range from 11 (lower QOL) to 73 (higher QOL).

ESM competency was measured with the Epilepsy Self-Management Scale (ESMS); scores range from 1-190 with higher scores indicating better self-management of epilepsy [13]. Stigma in PWE was measured using the Epilepsy Stigma Scale (ESS) [17,18]. In the ESS, scores range from 7-70; each item is rated on a 7-point scale from strongly disagree to strongly agree, with higher numbers indicating greater perceived stigma. Social support was measured with the 12-item Multidimensional Scale of Perceived Social Support (MSPSS), that measures perception of social support provided by family and friends, as well as satisfaction with that support [19]. The MSPSS score ranges from 1-84 with higher scores indicating better social support. Self-efficacy was measured using the 33-item Epilepsy Self-Efficacy Scale (ESES) with scores ranging from 0-330 and higher scores indicating better self-efficacy [17-21].

To help inform future efforts in scale-up and implementation [22], the study team conducted process evaluations in two main areas: 1) PWE’s perception of the SMART program, and 2) “Reach” among referral partners. PWE perceptions were assessed via a set of open-ended questions regarding their satisfaction with the program, comprising Likert-style questions assessing positive/negative/neutral components of the program and whether they would recommend participation in SMART to a family member or friend with epilepsy. To assess reach, referral sources were tracked for all study screens as well as successful enrollments.

Statistical analyses were performed using STATA v17.0 (StataCorp, TX). Descriptive analyses compared individuals assigned to SMART or WL using Fisher’s exact test and non-parametric Wilcoxon-Mann-Whitney tests when appropriate. Analyses were performed for the intention-to-treat population (ITT, all participants enrolled to receive the SMART trial; N=160) and in a per-protocol (PP) sample.

Because of study disruptions and delays due to the COVID-19 pandemic, which began shortly after this study began, there was substantial variability in the time lag between study baseline and when each cohort of individuals (group of 16 PWE with half randomized to SMART and half randomized to WL) began their SMART group sessions. The pandemic also impacted scheduling of follow-up visits. To account for this variable observation period at follow-up, in the ITT sample, NHEs and seizure counts were pro-rated by calculating NHE and seizure counts over each participant’s observation period and re-scaled to derive NHE and seizures counts over the prior 6 months. The primary outcome to assess SMART vs. WL differences in proportions of PWE with NHE reductions from baseline to 6 months follow-up was defined as: 1) any improvement, 2) improvement that reached each the clinically relevant thresholds of ≥10%, ≥20%, ≥30%, ≥40%, and ≥50%, and 3) change in NHEs and seizures over 6 months. Logistic regression models with the binary dependent variable of any improvement or improvement at specific thresholds at 6 months and the intervention as the independent variable, used a clustered sandwich estimator for the variance-covariance matrix allowing for intragroup correlations by RUCC category, and were restricted to those reporting <500 NHEs or seizures at baseline and at 6-month follow up. Change in NHEs and seizures over 6 months were inverse rank normalized, and the effect of SMART was evaluated using linear regression models allowing for intragroup correlations by RUCC category and restricted to those with <180 NHEs or seizures at baseline and at 6-month follow up (those averaging at most 1 event per day).

All comparisons were repeated restricting to a PP sample which was defined as those who received the intervention within 2 months of baseline visit (+1 week for anticipated normal scheduling variability) and subsequently SMART and WL participants completing the 6-month follow-up visit within 180 days (+2 weeks for anticipated normal scheduling variability). This sample included 58 participants (20 SMART, 38 WL).

For secondary outcomes, longitudinal linear mixed models with a subject-level random effect, and an intervention-by-time (days from baseline) interaction spanning baseline, week 13, and 6-month observations for PHQ-9, QOLIE-31 T-score, 7 QOLIE-31 sub-scores, SF-36 physical and mental component summaries, ESMS, ESS, ESES, and MSPSS, similarly inverse rank normalizing each outcome and adjusting the variance-covariance matrix to allow for intragroup correlations for RUCC levels.

Amongst those receiving SMART, we also assessed if number of sessions attended were associated with primary and secondary outcomes using similar models described above. A two-sided, Type I error rate of 5% was used for all tests.

Figure 1 illustrates overall study flow. There were 176 PWE screened and 160 randomized. For the 6-month follow-up outcome there were a total of 65 PWE in the analyzable SMART sample (81.0% of original randomized allocation) and 63 in the analyzable WL sample (79.0%). Six-month study attrition was 20.0% overall (N=32 lost to follow-up or withdrew consent). Individuals lost to follow up were unable to be contacted by study staff and thus the reason for drop out is not known. Reasons for study withdrawal when this was available (N= 6, 3.75 % of the overall sample) included three individuals reporting no longer having time to participate and three individuals reporting they were no longer interested in participating. Baseline characteristics including age, gender, race, education, employment level, health literacy, epilepsy age of onset, epilepsy duration, past 6- month NHE counts and depression severity between those retained in the analysis and those who dropped out showed no significant differences, suggesting a low likelihood of substantial selection bias.

Demographic and clinical characteristics, shown in Table 1, had a mean age of 39.4 (SD 12.2) years, 66.9% (N=107) women and 82.5% (N=132) white. A slight majority (N=83, 51.9%) had publicly funded health insurance (Medicaid, Medicare, VA). Just over 31% (N=50) were from more rural regions (RUCC of 4-9), 56.9% unemployed, and 40.3% with yearly incomes of less than $25,000 US dollars. Mean depression scores were in mild/moderate severity range with a mean PHQ-9 of 10.5 (SD 7). Participants had epilepsy for a mean duration of 15.5 (SD 13.8) years. The most common seizure type was bilateral tonic-clonic, reported by 49.8% of individuals.

The overwhelming proportion of NHEs were seizures. Given that recall of exact seizure counts among those with very seizure recurrence might not be reliable, those with seizure counts > 500 in the prior 6 months were not included in seizure or NHE count ITT analyses. Of 152 PWE with analyzable seizure data (not missing or not above 500), the mean seizure count in the 6 months prior to baseline was 20.6 (SD 41.0). It is notable that the range of past seizure counts was particularly broad in the WL arm (0-375) although this was not statistically significant compared to SMART (0-180). While 41.3% (N=66) of the sample had at least one ED visit and 23.1% (N=37) had at least 1 hospitalization in the 6 months prior to study enrollment and the mean number of hospitalization NHEs was 0.34 (SD 0.96); overall, non-seizure NHEs were relatively rare events.

For individuals in SMART, the average number of group sessions attended (out of a possible maximum of 8) was 5.9 (SD 2.8). Of the 80 individuals randomized to SMART, 9 (11.3%) attended no sessions. There were no adverse events related to study participation.

In unadjusted comparisons of SMART vs. WL, there was no significant difference between rates of any improvement in NHE and non-significant trends for SMART participants to have selected thresholds for improvement in NHEs compared to WL (Table 2). From the logistic regression models that allowed for intraclass correlation by RUCC category of residence (Table 2), SMART was significantly associated with 2-fold increased odds of having at least 10-30% improvement in the number of NHEs compared to WL (p<0.05). An examination of PWE who averaged having NHEs or seizures of not more than about once/day (mean of 180 NHEs or 180 seizures at baseline and at 6 months follow-up) found that individuals in SMART had significantly less seizures (p<0.05) and a trend for fewer NHEs (p=0.054) compared to WL. Figure 2 notes NHE frequency distributions for SMART at baseline and at 6 months (A) and for WL at baseline and at 6 months (B).

There was no significant difference between SMART and WL in non-seizure NHEs including hospital length of stay in days (data not shown). In the modestly sized PP analyses (20 SMART, 38 WL), there was no evidence for associations with any of these outcomes (data not shown). Number of SMART sessions attended was also not associated with these outcomes. With respect to accuracy (percent agreement) of self-report vs. health record hospital-based NHE documentation, baseline ED visit accuracy was 78% and baseline hospitalization accuracy was 91% while 6-month ED visit accuracy was 82% and 6-month hospitalization accuracy was 97%.

When restricting to those who attended at 8 session and comparing to WL, similar significant results were obtained for 10% and 20% improvement (p<0.05) and trending association for decreases in NHEs and seizures (p<0.1) (Supplementary Table 1). When comparing those who attended 5 or fewer sessions to WL, there were significant 10-40% improvements (p<0.05) with non-significant trends for decreases in NHEs and seizure counts (p<0.15) (Supplementary Table 2).

Table 3 shows changes from baseline to 13 weeks and 6 months in depressive symptoms severity (PHQ-9), epilepsy-related QOL (QOLIE-31 T-score (composite measure/T-score and 7 sub-scales)), functional health status (SF-36, physical and mental health components), epilepsy self-management competency (ESMS), epilepsy stigma (ESS), social support (MSPSS) and epilepsy self-efficacy (ESES). In the ITT sample, SMART was associated with significantly improved PHQ-9 (p=0.014), QOLIE-31 seizure worry (p=<0.001) and social functioning (p=0.021), ESES (p=0.046) and ESMS (p<0.001). In the PP sample SMART was associated with significantly improved QOLIE-31 T-score (p=0.011) and social functioning (p=0.007), ESS (p=0.04), ESES (p=0.01), ESMS (p=0.006) and SF-36 mental health component (p=0.014). There were no significant differences between SMART and WL on other variables. Change in any of the secondary variables was not significantly associated with SMART session attendance.

As noted in Table 4, the majority of PWE enrolled (approximately 59% for both screen and randomized participants) were referred from clinicians. Epilepsy-specific social support entities such as the Epilepsy Foundation referred approximately 30%. In the group that was randomized to SMART, 31 participants responded to the acceptability survey. Of those that completed the acceptability survey, the majority, 23 (74.2%) strongly agreed or agreed that SMART covered most of the important issues and 21 (68.0%) strongly agreed or agreed that SMART addressed issues important to their particular situation. A majority, 19 (61.3%) strongly agreed or agreed that the benefit of SMART exceeded the burden or hassle of attending. Additionally, 22 (71.0%) felt that the number of sessions was about right, while 25 (81.0%) felt that the length of each SMART session was about right.