Clinical and demographic information for our sample population is shown in Table 1. Among the 53 patients included in this study, the median age at the time of the study was 28 years, with ages ranging from 8 to 91 years. The median age at VNS implantation was 18 years, ranging from 2 to 83 years. Forty patients (75.5%) had childhood-onset epilepsy. The average duration of VNS therapy was 11.2 years, ranging from 0.25 to 24 years. There were 22 patients who underwent VNS therapy for over a decade. Overall, a total of 19 patients (35.8%) were responders, and 17 of them (89.5%) had childhood-onset epilepsy. Twenty-eight patients (52.8%) were diagnosed with developmental epileptic encephalopathy, including 14 patients (26.4%) with Lennox-Gastaut Syndrome (LGS). Ten patients (18.9%) had a neuronal migration disorder. Over the 24 years, twelve patients discontinued VNS therapy, including four deaths, three from lack of efficacy, four due to complications from the VNS implantation procedure, and one due to exercise intolerance. Twenty-two individuals (41.5%) experienced less than a 50% reduction in seizure frequency and were classified as non-responders. Nonetheless, they opted to continue with VNS therapy, as it offered modest benefits and caused minimal side effects. The retention rate was 77.4%. Thirty-three patients received VNS models with automatic stimulation (AutoStim). Seventeen of them remained to be responders, while 16 remained to be non-responders. Overall, the new models showed no significant additional benefit in seizure control.

We examined the continuous variables and their associations with VNS outcomes, including age at VNS implantation, epilepsy duration, output current, and duty cycle. None were statistically significant by the Wilcoxon rank-sum test (Table 2). The median epilepsy duration before VNS placement in years was lower among VNS responders (5 years) vs. non-responders (11.5 years); however, the overlap in the distribution between these groups was substantial (1–34 years for responders, 0.25-30 years for non-responders), and the difference was not statistically significant (p = 0.13) (Table 2). We then plotted receiver operating characteristics (ROC) curves for each variable to calculate optimum threshold values for each variable and the sensitivity and specificity they provide (Fig. 1). We found that epilepsy duration < 5 years identified responders in our population with 85% sensitivity and 47% specificity (Table 3). 100% of responders had output current > 1 mA; however, this cutoff was only 15% specific. Conversely, a duty cycle > 25% was 91% specific for VNS response in our population but only 16% sensitive. The ROC curves also suggested that age at VNS implantation < 18 years might affect the response. We found that the area under the ROC curve (AUC) for epilepsy duration and age at implantation were higher than those of the VNS parameters, although we noted that the 95% confidence interval (CI) for every AUC contained 0.5, the value associated with random chance.

The receiver operating characteristics (ROC) of continuous variables. TPF = True Positive Fraction, equivalent to sensitivity. FPF = False Positive Fraction, equivalent to 1 – Specificity. The solid line indicates the ROC curve; the dashed line indicates the diagonal (signifies random chance); the point indicates the optimum threshold by the Youden Index (see Table 3)

We then did a binary univariate analysis of categorical variables, including epilepsy types and surgery before VNS, and continuous variables based on threshold values from the ROC curves to try to identify potential predictors of a good VNS response (Table 4). 64% of patients who had an epilepsy duration of < 5 years were responders compared to 26% with longer epilepsy duration, a difference that was statistically significant (p = 0.021). There were 5 patients (9%) who had epilepsy surgery prior to VNS placement. It is interesting that none of the patients with prior surgical resections were classified as responders to VNS (and all had epilepsy duration for more than 5 years at VNS placement) (Table 4).

Even though VNS was only approved to treat focal epilepsy in the US, it has been used off-label for patients with generalized epilepsy. We included two patients with refractory Juvenile Myoclonic Epilepsy (JME) and one patient with refractory Juvenile Absence Epilepsy (JAE), two with epilepsy with myoclonic atonic seizures (EMAtS) and two patients with epilepsy with generalized tonic-clonic seizures alone (EGTCS). Their EEGs showed generalized spike and wave complexes. Magnetic resonance imaging (MRI) of the brain demonstrated no focal lesions. All of them except one were responders. The non-responder reported magnets helped abort myoclonus clusters. VNS helped decrease all seizure types in these patients, including GTCs, absence seizures, and atonic seizures. The responder rate of patients with generalized epilepsy (86%) was significantly higher than that of patients with focal epilepsy (29%) or combined epilepsy (28%) (Table 4). The p-value was 0.016, indicating generalized epilepsy is a predictor for good response.

VNS parameters, including a duty cycle >25% and an output current>1 mA, did not prove to be significant predictors for VNS response (Table 4). While some evidence in the literature suggests that increased duty cycles could significantly increase efficacy [24], we did not find this to be the case in our population (p = 0.65). Rapid cycling, with off-time ≤1.1 min, was employed in six patients who did not have a significant seizure reduction after reaching the standard target settings. Three of them with duty cycles of 41%, 35%, and 29% were responders, while another three patients with duty cycles of 41%, 49%, and 35% were non-responders. There was no significant difference in output current between responders and non-responders (p=0.15).

In a multivariate logistic regression analysis (Table 5), we found that epilepsy duration of < 5 years was a strong independent predictor for successful VNS response (OR = 7.23, p = 0.01) after adjusting for epilepsy type, epilepsy surgery prior to VNS, age at VNS implantation, duty cycle, and output current. Similarly, generalized epilepsy was also a predictor for a good response compared to focal epilepsy, combined epilepsy, and other covariates (OR = 8.62, p = 0.027). The effects of epilepsy duration of < 5 years and generalized epilepsy were clearly independent of one another, as only 1 of 7 (14%) generalized epilepsy patients had epilepsy duration of < 5 years at VNS implantation, compared to 6 of 22 (27%) focal epilepsy patients and 7 of 11 (64%) combined epilepsy patients.