The Blind Leading the Uncontrolled: How Does Seizure Frequency Improve Over Time in Drug-Resistant Epilepsy?

Commentary

Numerous randomized-controlled trials attest to the efficacy of antiseizure medications (ASMs), epilepsy surgery, and neuromodulation for people with drug-resistant epilepsy. Blinded, randomized trials are one of the most powerful tools we employ to seek truth in medicine. These trials tell us how helpful these therapies, like a new medication or novel stimulator, are when compared to placebo, sham, or alternative treatments. By virtue of the rigor inherent in their design and conduct, we can confidently counsel people with drug-resistant epilepsy regarding the improvements these therapies realistically offer over the short term.

However, the controlled and blinded portion of these trials is typically brief. Beyond this window, we lack high-quality (randomized, blinded) evidence of what these therapies achieve long-term compared with sham or standard of care. Our best evidence is instead derived from long-term, open-label extension studies. These studies span years but, for ethical and logistical reasons, lack blinding and comparator groups. It is ethically problematic to maintain placebo or sham arms for years when effective alternative therapies exist, and financing extended multiyear trials is logistically prohibitive. The long-term, open-label studies we have are nevertheless unequivocally useful. They help define persistent safety, show consistent seizure control, and attest to maintained quality-of-life improvements.

Intriguingly, these long-term extension trials, regardless of the medication studied or neuromodulatory therapy examined, tend to show two consistent features. First, most show improved seizure control that is greater than what was found in the blinded periods. Second, most show a consistent improvement over time, which some have argued shows a disease-modifying effect. For example, at 4 years, cenobamate showed a 76.1% reduction in seizure frequency, ¹ lacosamide had a 62.5% reduction, ² responsive neurostimulation (RNS) had 63.3%, ³ and deep brain stimulation (DBS) had 66%. ⁴ These proportions are strikingly similar. Yet, because we are not comparing this gradual improvement to sham or placebo, the source of this continued benefit remains opaque.

This uncertainty has not stifled speculation. Optimistically, some believe these improvements reflect a disease-modifying effect of the therapy itself. This potential plasticity may result from direct influence on neural networks (as with RNS or DBS) or via intrinsic homeostatic mechanisms enabled by a reduced seizure burden. A less hopeful interpretation is that these improvements are artifacts of study design rather than consequences of the intervention. Observed improvements could stem from the dropout of non-responders, reporting fatigue (patients becoming less rigorous with diaries over years), or regression to the mean—patients often enroll in trials when their condition is at its nadir, ensuring that natural fluctuation looks like improvement. ⁵

A middle ground is that there may be factors that improve seizure control beyond the precise drug or device used, that are common to actively managed patients. For example, enrolling in a trial may lead to overall improved contact with the healthcare system and management of comorbidities—a Hawthorne effect for the participating patients. ⁶ People with epilepsy may also improve their own management over time, for instance by achieving a better understanding of triggers, like sleep and medication adherence.

In a timely and well-designed effort to help differentiate among these hypotheses, the Human Epilepsy Project 2 (HEP2), a consortium spread across 10 hospital systems in the United States, studied 146 people with drug-resistant focal epilepsy for 3 years. ⁷ Crucially, this was an observational study; physicians managed patients at their discretion without protocol-mandated interventions. This allowed for an analysis of the “natural history” of patients with stable regimens versus those receiving new interventions.

As might have been predicted by prior open-label studies, just over two-thirds of patients (68.3%) experienced seizure frequency reduction when comparing the second half to the first half of their study participation. The overall cohort showed a mean modeled seizure frequency reduction of approximately 69%. About half of participants (55%) had an ASM added during the study, and of those, two-thirds experienced seizure reduction—though only 6 achieved three or more months of seizure freedom directly following the addition.

Thirty-five patients had implanted devices for neuromodulation (VNS, RNS, or DBS), with 14 devices implanted during the study. While a Friedman test indicated that devices influenced prospective seizure trajectory post-implantation, seizure trajectories in participants with devices did not significantly differ from those without devices—a notable finding. The single participant who underwent resective epilepsy surgery during the study achieved seizure freedom, attesting to the continued power of surgical interventions that remove epileptic tissue.

Where does this leave us? A nihilistic interpretation is that drug-resistant epilepsy tends to improve regardless of the specific management choice. A more hopeful interpretation—equally consistent with these data—is that “active management” itself drives improvement. Supporting this, the HEP2 authors note that among the minority of patients achieving 3 months of seizure freedom, 82% had a medication change near that time. This suggests that the improvement is not purely passive “natural history,” but the result of expert clinicians relentlessly optimizing therapy.

We still lack definitive evidence to identify which hypothesis explains these long-term improvements, as we lack randomized-controlled data at these time horizons. Since multiyear sham-controlled trials are fraught, future research should interrogate the alternative hypotheses. If “diary fatigue” is suspected, objective measures (eg, RNS detections or wearables) could be used alongside diaries. If regression to the mean is the culprit, delayed-start designs could be employed, similar to the pivotal trial for temporal lobectomy where patients were randomized to surgery or a 1-year waiting list. ⁸

Ultimately, the HEP2 trial is both humbling and reassuring. It is humbling because it highlights how much remains unknown about treatment effects beyond the brief window of controlled trials. Yet it is reassuring because these patients, managed by experts, experienced meaningfully improved seizure control—even if the precise mechanism remains distinct from the specific drug or device employed. Importantly, this new study does not call into question the efficacy of treatments validated by randomized trials—they work. Rather, it raises the compelling question of why this additional improvement occurs and why it appears so uniform across modalities. Answering this matters deeply for people with epilepsy and represents a rich direction for future research.