Factors associated with placebo response rate in randomized controlled trials of antiseizure medications for focal epilepsy

Abstract

Objective

Randomized controlled trials (RCTs) are necessary to evaluate the efficacy of novel treatments for epilepsy. However, there have been concerning increases in the placebo responder rate over time. To understand these trends, we evaluated features associated with increased placebo responder rate.

Methods

Using individual‐level data from 20 focal‐onset seizure trials provided by seven pharmaceutical companies, we evaluated associations with change in seizure frequency in participants randomized to placebo. We used multivariable logistic regression to evaluate participant and study factors associated with differing rates of 50% reduction in seizure frequency during blinded placebo treatment, as compared to pre‐randomization baseline seizure frequency. In addition, we focused on the association of placebo responder rate with pre‐randomization baseline seizure frequency and country of recruitment.

Results

In the pooled analysis of 1674 participants randomized to placebo, a higher 50% responder rate (50RR) was associated with a shorter duration of epilepsy (p = .006), lower baseline seizure rate (p = .002), fewer concomitant antiseizure medications (p = .004), absence of adverse events (p < .001), more trial arms (p = .006), and geographic region (p < .001). Mixture modeling indicated a significantly higher 50RR in Bulgaria, Croatia, India, and Canada (42% in the higher group vs 22% in the lower group comprising all 40 other countries, p < 10⁻¹⁵). In addition, there was a significantly higher 50RR in participants with a baseline seizure frequency of six or fewer seizures per 28 days (29% vs 21%, p = .00018).

Significance

These results can assist future RCTs in estimating the expected placebo responder rate, which may lead to more reliable power estimates. Higher placebo responder rate was associated with markers of less‐refractory epilepsy. There were concerning significant differences in placebo responder rate by country and geographic region as well as an elevated placebo responder rate in participants with baseline seizure frequency close to the minimum eligibility criteria.

Key points.

• Clinical features associated with refractory epilepsy were associated with lower placebo responder rate.

• Placebo responder rate was higher in 4 of 44 countries.

• Placebo responder rate was significantly higher in participants with six or fewer baseline seizures per month.

• Higher placebo responder rate was associated with fewer adverse effects.

1. INTRODUCTION

Randomized placebo‐controlled trials (RCTs) are crucial to the evaluation of novel treatments for epilepsy. ¹ Despite the development and approval of numerous additional pharmacologic and non‐pharmacologic treatments for epilepsy, there remain patients with inadequate control of seizures: ~30% of patients with epilepsy continue to be medication resistant. ¹ , ² , ³ There is an enduring critical need to conduct well‐powered RCTs to evaluate the benefit of novel treatments. ⁴ , ⁵

The statistical power of these RCTs is maximized when the placebo effect is both low and consistent across all sites, so that the difference attributed to active treatment can be measured both accurately and precisely. ⁶ , ⁷ , ⁸ In a traditional RCT for medication‐resistant epilepsy, response to blinded treatment is calculated as a percent change in seizure frequency during the blinded phase (12 to 24 weeks) compared to the pre‐randomization baseline. The purpose of the baseline is twofold: (1) documentation of a trial‐qualifying retrospective seizure frequency for at least 3 months, which is then confirmed by a 4‐ to 8‐week prospective seizure diary; and (2) producing an accurate and reliable estimate of pre‐randomization seizure frequency that will be the barometer of post‐randomization seizure response. ⁹

For a novel treatment to demonstrate efficacy for seizure control, the active treatment must produce a significant reduction in seizure frequency compared to placebo. Unfortunately, the placebo 50% responder rate (50RR) has been increasing over time from 10% in 1990 to over 22% after 2020. ⁶ , ⁷ This means that more recently developed treatments have a higher bar for efficacy demonstration than previously developed treatments, since the statistical power to observe the same efficacy is reduced. To overcome these reduced effect sizes, the number of participants in trials has increased. This can contribute to increases in the cost of developing the novel treatments that are critical to improving the care of patients with epilepsy. ⁶

The reasons for these increases in the placebo responder rate are unclear. In parallel with the rising placebo responder rate, the number of participants per site has decreased, leading RCTs to recruit from more sites and countries. ⁶ The most common reason for ineligibility for RCTs was insufficient baseline seizure frequency (BSF). ¹⁰ Simulation‐based analysis suggested that some of this elevated placebo responder rate may be attributed to regression to the mean, where a participant meets the minimum eligibility threshold when their seizure frequency is in the upper part of its range for that individual patient. ¹¹ Due to natural variability, each individual person has a range of seizure frequencies that occur month to month. ¹² , ¹³ , ¹⁴ , ¹⁵ However, after entering the blinded period, their seizure frequency declines to their lower, long‐term seizure frequency irrespective of treatment. In addition to this concept of regression to the mean, participants with fewer markers of refractory epilepsy seemed to have had a higher placebo effect in prior RCT meta‐analyses. ¹⁶ Furthermore, there have been concerning findings that the effect of active treatment has declined in some countries, without clear explanations. ¹⁷ , ¹⁸

In this study, we used data from a collection of placebo‐controlled RCTs for focal‐onset seizures to evaluate participant, study, and site characteristics that were associated with placebo responder rate. This data‐driven approach can assist trialists in identifying the key features associated with the placebo responder rate so that they can be anticipated and addressed directly. For example, the eligibility requirements and participant recruitment strategies could be modified with the goal of either reducing or more accurately predicting the placebo responder rate.

2. METHODS

This study re‐analyzed pooled results from 20 placebo‐controlled RCTs for focal‐onset epilepsy in adults and children from seven pharmaceutical companies. This included RCTs of perampanel (Eisai: National Clinical Trials Identifier [NCT] 00699972, NCT00699582, NCT00700310), ¹⁷ , ¹⁹ , ²⁰ lamotrigine (GlaxoSmithKline: NCT00113165, NCT00104416, NCT00043901), ²¹ , ²² , ²³ pregabalin (Pfizer: NCT00141258, trial numbers 1008‐011, 1008‐122, 1008‐000‐157, 1008‐034, 1008‐009), ²⁴ , ²⁵ , ²⁶ eslicarbazepine (Sunovion: NCT00957684, NCT00957047, NCT00988429), ²⁷ oxcarbazepine extended release (Supernus: NCT00772603), ¹⁸ lacosamide (UCB: NCT00136019, NCT00220415, and trial 667), ²⁸ , ²⁹ and topiramate extended release (Upsher‐Smith, NCT01142193). ³⁰ The individual‐level data from these RCTs was made available to researchers through data‐use agreements limited to participants randomized to placebo and excluding data from participants randomized to active treatment. The responder rate was quantified as the percent change in seizure frequency during the maintenance period, as compared to the pre‐randomization baseline (definition varied by trial including 4 to 12 weeks that were either a mix of retrospective or retrospective plus prospective diaries). Individual‐level participants' response to placebo was analyzed based on a 50% or greater reduction in seizure frequency during maintenance excluding titration. This 50% responder rate (or 50RR) corresponds to the primary efficacy metric recommended by the European Medicines Agency (EMA). When seizure diary data were missing due to early withdrawal, they were filled with the last observation carried forward approach.

We evaluated the association of placebo responder rate with participant‐, study‐, and site‐level factors. Participant‐level demographic factors included age, sex, race, and ethnicity. Other participant‐level factors included epilepsy duration, number of concomitant antiseizure medications (ASMs), baseline seizure frequency (or BSF), participant's time in the trial, if the participant reported an adverse event during the trial, and the number of reported adverse events. Study‐level factors included the total duration of the trial, total number of participants in the trial, number of treatment arms in the trial, and year of the start of the study. Site‐level factors included number of participants per site, number of study sites, number of geographic regions, specific geographic region, number of countries, and specific countries. To promote privacy and reduce identifiability, the identity of each specific site within each country was not shared, and it could not be re‐identified based on the individual patient‐level data.

We evaluated the association of responder rate with these factors using either mixed‐effects logistic regression (yes/no factors) or rank analysis of covariance (ANCOVA; continuous factors), stratified by both country and trial. We evaluated these associations for each individual factor as well as with a generalized estimating equation to account for covariation across factors. In this first analysis of country‐associated differences, we compared non‐U.S. sites to the reference of the United States because the United States was the single country that recruited the most participants and had a middle‐range 50RR.

After this first simplified analysis by country, we performed additional analyses based on each country of recruitment. We evaluated if individual countries or regions had a significantly elevated placebo responder rate compared to others (one‐vs‐all, p < .05). To address concerns with multiple comparisons due to evaluating each country separately, we used mixture modeling to use a data‐driven approach to test the hypothesis that there were groups of countries with higher or lower placebo responder rates. ³¹ To determine the number of groups, we used a forward selection approach, where a group was added only if a log‐likelihood test showed a significant improvement of residual deviance, when adjusting for the number of groups (chi‐square, p < .05). Countries that recruited fewer than 10 participants were included in the reference (not high) placebo responder rate group based on a paucity of evidence for a higher or lower responder rate than other countries.

We performed an additional analysis of responder rate based on BSF. In that analysis, BSF was grouped into integer ranges (e.g., 4–4.99 seizures/28 days) and the responder rate for ranges of BSF compared to others was evaluated with Fisher binomial exact tests. Due to the propensity for multiple testing because of testing multiple thresholds for BSF, we also used mixture modeling to confirm the appropriate choice of threshold. ³¹

3. RESULTS

Data from 1674 participants with focal‐onset epilepsy randomized to placebo were available from seven pharmaceutical companies with an earliest start date of June 1998 and latest end date of December 2012 (Sunovion 424, Eisai 399, UCB 364, Pfizer 199, Supernus 121, Upsher‐Smith 125, and Glaxo‐Smith‐Kline [GSK] 118) of which 395 (23.6%) were 50% responders.

3.1. Factors associated with placebo responder rate

First, we evaluated univariate associations of participant‐, study‐, and site‐level factors with 50RR (Table 1; see Table S1 for factors without a significant association with 50RR). Higher placebo responder rate was associated with shorter epilepsy duration, lower baseline seizure frequency, fewer concomitant ASMs, absence of adverse events, fewer adverse events, smaller number of trial arms, and fewer participating countries. There were complex associations of placebo responder rate with geographic region.

TABLE 1.

Demographics and characteristics associated with 50% responder rate (p < .05).

	Non‐responder	50% responder	p
Participants, n (%)	1279 (76.4%)	395 (23.6%)
Years of age (SD)	37.5 (11.9)	38.6 (12.2)	.11
Female sex	652 (51%)	200 (51%)	.95
Epilepsy duration, years [IQR]	21.5 [12.1–31.0]	19.5 [9.6–28.9]	.006
Baseline seizure freq/28 days [IQR]	9.5 [5.7–21.7]	8.0 [5.0–16.1]	.002
Concomitant ASMs, n [IQR]	2.0 [2.0–3.0]	2.0 [1.0–2.2]	.004
Adverse events, n (%)	892 (70%)	236 (60%)	<.001
Adverse events, number [IQR]	2.0 [.0–4.0]	1.0 [.0–4.0]	.005
Number of trial arms			.006
2	201 (16%)	89 (23%)
3	718 (56%)	211 (53%)
4+	360 (28%)	95 (24%)
Countries per trial, number [IQR]	12.0 [8.0–16.0]	11.0 [8.0–16.0]	.023
Geographic region			<.0001
Africa	12 (0.9%)	5 (1.3%)
North America	328 (26%)	99 (25%)
South America	129 (10%)	38 (10%)
Asia	153 (12%)	63 (16%)
Eastern Europe	331 (26%)	132 (33%)
Western Europe	261 (20%)	46 (12%)
Oceania	53 (4%)	6 (1.5%)

Note: The value listed for continuous values is the median. The denominator for each characteristic if the number of either non‐responders or 50% responders (e.g., 0.9% of non‐responders were from Africa). Significance was based on Fisher exact tests, Wilcoxon signed‐rank tests, and chi‐square tests of independence. For characteristics without significant associations and multivariable associations, see Tables S1 and S2.

Abbreviations: freq, frequency; IQR, interquartile range; n, number of participants; SD, standard deviation.

To evaluate the combined and conditionally independent effects of these multiple associations, the multivariable model also demonstrated significant associations of placebo responder rate with shorter epilepsy duration, absence of adverse events, smaller number of trial arms, non‐U.S. site, and longer time in the blinded phase of trial (Figure 1 and Table S2). In this multivariable model, there was an additional association of older age with higher placebo responder rate (odds ratio [OR] 1.2 per decade of age, 95% confidence interval [CI] 1.07–1.36, p = .002), but no significant association with BSF (OR 0.98 per 10 seizures/28 days, 95% CI 0.92–1.04, p = .51).

FIGURE 1.

Multivariable logistic regression demonstrating characteristics associated with placebo 50% responder rate (50RR, *p < .05). For all characteristics in the regression, refer to Figure S1.

3.2. Placebo responder rate varied by country

To evaluate the complex relationship between geographic region and placebo responder rate (Table 1, Tables S1 and S2), we further evaluated the variation in 50% placebo responder rate by country (Figure 2, Table S3). We addressed the challenge of multiple testing using mixture modeling, which suggested that there were two groups of 50RR: high 50RR (42%, 95% CI 33%–50%) and not high 50RR (22%, 95% CI 19%–24%; see Table S3 for detailed results). Bulgaria, Croatia, India, and Canada were significantly more likely to be in the high 50RR group.

FIGURE 2.

Differences in placebo 50% responder rate (50RR) based on country of recruitment. Countries with average (reference) 50RR are in gray. Countries with higher 50RR are shades of red, whereas countries with lower 50RR are in shades of blue. Countries other than those highlighted (Bulgaria, Croatia, Canada, and India) with high 50RR recruited insufficient participants to detect a significant difference from the average country (see Table S3). Countries with no participants were not visualized (e.g., most African nations).

When these countries' individual responder rate was compared to all other countries (one‐vs‐all), there was also evidence for an elevated placebo responder rate in Bulgaria (16/28, 57%, 95% confidence interval 39%–75% uncorrected p = .00011, false discovery rate [FDR] corrected p = .0048), Croatia (14/22, 42%, 95% CI 27%–61%, uncorrected p = .012, FDR corrected p > .50), India (19/52, 37%, 95% CI 23%–50%, uncorrected p = .028), and Canada (13/23, 36%, 95% CI 22%–53%, uncorrected p = .078). Eleven countries had fewer than 10 recruited participants and therefore had insufficient data to conclude they were not in the reference group with a not high 50RR (Table S3).

In addition to those four countries with high placebo responder rate, Australia had a significantly lower 50% placebo responder rate compared to all other countries (5/58, 9%, 95% CI 1%–18%, uncorrected one‐vs‐all p = .0099, FDR corrected p > .40). The mixture model that considered three groups identified a high group, reference (not high) group, and a very low responder rate group (the latter of which included Australia only), but there was insufficient evidence that the three‐group model was superior to the two‐group model (p = .29, Table S3).

3.3. Association of placebo responder rate with baseline seizure frequency

To evaluate if there was empiric evidence for increase placebo responder rate from regression to the mean, we evaluated the association of 50RR with baseline 28‐day seizure frequency (Figure 3). Participants with six or fewer baseline seizures per 28 days had a significantly higher placebo 50RR compared to other BSFs (29% vs 21%, p = .00018; see Table S4 for detailed results). Overall, 27% of participants randomized to placebo had six or fewer baseline seizures per 28 days.

FIGURE 3.

Placebo 50% responder rate (50RR) associated with baseline seizure frequency. Dashed vertical bar indicates that 50RR was higher in participants with six or fewer baseline seizures per 28 days compared to more than six (p = .00018). Shading reflects binomial exact standard error equivalents with Gaussian smoothing of log baseline seizure frequency with standard deviation of 1.3 seizures per 28 days.

4. DISCUSSION

This study of the clinical characteristics associated with placebo responder rate across 20 placebo‐controlled randomized trials can inform future trials to better predict the expected placebo responder rate, which is key to estimating statistical power and setting recruitment goals. The unexpected differences in placebo response in particular countries and in participants with the lowest baseline seizure frequency for eligibility were concerning.

The participant‐related features associated with a lower placebo responder rate were also markers of long‐term medication‐refractory epilepsy. ² , ¹⁶ These included longer duration of epilepsy, higher BSF, and more concomitant ASMs. The reduced placebo responder rate in these participants was intuitive because these participants' seizure frequency may have stabilized over the many years that they had epilepsy. Their medication resistance, defined by a lack of variation in seizure frequency in response to ASMs, may also potentially indicate less variation in seizure frequency over time. However, this explanation is speculative because previous analyses of the natural variability of seizure frequency have not evaluated if reduced variability was associated with these factors. The association of these markers of medication resistance with reduced placebo responder rate in participants with these markers of refractory epilepsy also has been observed in other trial re‐analyses. ¹⁶

The additional association of lower placebo responder rate with a higher number and more frequent reporting of adverse events was the opposite of what might be expected. We expected that negative experiences on treatment (e.g., an adverse event) would lead participants to guess that they were randomized to active treatment, and thereby should benefit from treatment. ³² Instead, the negative experience of an adverse effect was potentially associated with a negative view of treatment, and thereby a lower placebo response. This co‐occurrence of poor seizure response and adverse events has been described as a nocebo effect in other trials for epilepsy, which defined the nocebo effect as any seizure or other adverse experiences on placebo. ⁸ , ³³ Alternatively, this association could be explained by participants who had a very positive anticipation of treatment and, in turn, had both a higher 50RR and fewer adverse effects.

Although the above associations had some biologically plausible explanations, the variability by country and BSF did not have clear pathophysiologically‐based explanations and, therefore, is concerning.

The increased cost and challenge of recruiting participants into trials has prompted sponsors to expand to include countries and centers that did not contribute previously. ³⁴ It was reassuring that placebo 50RR was similar in 40 of 44 countries (91%), despite their vast differences in health service models, funding structures, clinical practice, availability of ASMs, history of conducting clinical trials, and other differences not considered in this analysis. However, there were four countries (9%) with significantly higher placebo responder rates, which represented three continents that recruited 7% of placebo participants. Each of these countries recruited between 22 and 57 participants, which provided sufficient data to conclude that the placebo responder rate was elevated, even after considering this moderate sample size and multiple testing corrections. In addition, many other countries with smaller numbers of participants may have had elevated placebo responder rate, but there was insufficient evidence to demonstrate that statistically (e.g., New Zealand with one 50% responder of one participant recruited). To maintain participant and site privacy, we only evaluated placebo responder rate based on country. There often were multiple sites within each country that may have had elevated, lower, or average placebo responder rate, so we emphasize that our results may not pertain to each site within each country.

Expanding recruitment to countries and sites with elevated placebo responder rates may have created a vicious cycle, where challenges in recruiting led to expanding to recruit from new centers, some of which had high placebo responder rates which, in turn, reduced statistical power, and thereby required more participants to be recruited. ⁶ , ¹⁷ , ¹⁸ , ²⁰ , ³⁵ , ³⁶ Although improved diversity, equity, and inclusion in trials is important to results being generalizable to the global population who will use the novel active treatment, ³⁷ , ³⁸ , ³⁹ , ⁴⁰ the elevated placebo responder rate in these countries hinders the ability of trials to reliably estimate the effect of active treatment, as compared to placebo. For example, the highest country‐specific 50RR was 57% (16/28). If all participants were recruited from that country in a 180‐participant per arm trial, the active treatment 50RR would have to be more than 70% to demonstrate a significant benefit (p < .05). However, the typical average treatment responder rate was 40% ⁷ , ¹² , ⁴¹ ; therefore, these elevated placebo responder rates may lead to inaccurately concluding that a novel treatment is ineffective. False‐negative trials can have catastrophic results on future drug development.

The potential reason for elevated placebo responder rates in particular countries is unclear, but may be related to recruitment by centers with less subspecialty expertise in the details of performing clinical trials of treatments for seizures. ⁴² , ⁴³ , ⁴⁴ To bridge this difference in expertise, trials can enlist independent experts with subspeciality expertise in seizures and clinical trials (e.g., the European Consortium for Epilepsy Trials and Epilepsy Study Consortium, Inc. [ESCI]. Regarding conflicts of interest: Dr. French is president of the ESCI and Dr. Kerr is an adjudicator). ³⁴ Adjudication can include correcting misclassifications of seizure type (e.g., description including lapse in awareness, but the site investigator indicated focal aware seizures) as well as raising concern that the seizure description suggested that the episodes were not seizures (e.g., recurrent syncope). ³⁴ By clarifying the data from these patients, adjudication may improve the reliability of the results of trials, both in the countries where we saw statistically significant elevations of placebo responder rate rates, as well as other countries with elevated rates but insufficient data to reach statistical conclusions.

However, this difference in subspecialty expertise likely does not address all factors contributing to elevated placebo effect because, for example, most Canadian centers had a long history of subspecialty expertise in epilepsy and clinical trials. Further studies where data across multiple trials with identified site or country information are needed to identify and address all factors contributing to placebo response.

The most common challenge in recruiting participants for trials was meeting the minimum seizure frequency eligibility criterion. ¹⁰ Investigators, sponsors, and participants all hope that participants will benefit from the study drug; therefore, everyone involved is incentivized to recruit a sufficient number and quality of participants to demonstrate that benefit. Especially if it is known that a certain study is struggling to recruit, there can be pressure to assist in recruitment to meet the mutual goal of successful study completion. This hope to benefit patients and other incentives can lead to enrolling a participant after a 3‐month period with an eligible seizure frequency (e.g., four seizures/month), even if their usual seizure frequency is below the minimum (e.g., two seizures/month). ¹¹ If seizure frequency remains elevated during the 4‐ to 8‐week prospective baseline period, then seizure frequency during the blinded treatment period may undergo regression to that long‐term usual rate, which could qualify as a placebo response. ¹¹

Unfortunately, our finding of an elevated placebo responder rate for participants with fewer than six baseline seizures per month provides empiric data that matches the predictions of simulation‐based studies of regression to the mean. ¹¹ The solution to this problem is not as simple as raising the minimum seizure frequency to six per month. Raising the minimum eligibility required would exacerbate the recruitment problem that very few patients had seizure frequencies high enough to be eligible for trials. In these trials, 27% of participants had six or fewer baseline seizures per month; therefore, at least 27% more participants would have been needed to be screened and randomized to conduct these trials. Even if the minimum was raised, elevation of placebo responder rate from regression to the mean could shift to affect participants with six to eight baseline seizures per month, which would not fix the problem.

Conversely, lowering the minimum eligibility requirement may reduce the influence of regression to the mean for participants with six or fewer baseline seizures per month, but it may shift the problem to lower seizure frequencies. In addition, inclusion of lower seizure frequencies may require longer blinded treatment periods or more participants to achieve the same statistical power. ⁴⁵ Longer trials are both more expensive and more challenging to recruit for due to requiring more effort from participants and sites, as well as incurring more risk of adverse events, including sudden unexpected death in epilepsy (SUDEP), when exposed to placebo. ⁴⁶

Therefore, the solution to this problem of regression to the mean is unclear. Ideally, regression to the mean may apply equally to placebo and active treatment and thereby maintain the magnitude of difference, but prior analyses showed that when placebo response increased, there was a concomitant decrease in active treatment response. ¹⁷ , ¹⁸ Alternatively, simulation‐based work proposed that the effect of regression to the mean reduced when these minimum criteria were only applied to the retrospective seizure diary reported during participant screening, and those minimum criteria were relaxed when applied to the prospective baseline seizure diary before randomization. ¹¹ However, that approach may incentivize overestimation of retrospective seizure frequency to enter the trial, which would harm the statistical power of the trial due to participants with lower‐than‐expected seizure frequencies. ⁴⁷ Further studies are needed to identify the benefits and limitations of seizure frequency–based eligibility criteria.

Our approach of focusing on placebo responder rate has important limitations. Although we focus on the impact of placebo responder rate on statistical power, we did not calculate an effect size by comparing placebo responder rate to the effect of active treatment. This was because, due to the diversity of active treatments, the variation in effect size could be attributed either to changes in placebo responder rate or to efficacy of active treatment. In additionally, although the inclusion and exclusion criteria of all trials were very similar, they were not identical, which may contribute to some of our observed associations. In particular, trials with a later start date had a higher placebo responder rate and included a broader range of countries. Our regression found conditionally independent associations with both country and start date; however, this cross‐correlation and selection bias could contribute to elevated placebo responder rates in some countries. The last end date of trials in this analysis was December 2012, and placebo responder rate has continued to increase after that date. ⁶ , ⁷ However, more recent trials were not available for this analysis (e.g., cenobamate) and older trials of commonly used ASMs were no longer available for re‐analysis (e.g., levetiracetam). In addition, to promote participant and site privacy and identifiability, we were provided data only at the resolution of country and could not evaluate if elevated placebo responder rate could be attributed to inclusion of specific centers, even within countries that otherwise had low placebo responder rate. Based on these results, future analyses could evaluate site‐level associations with placebo responder rate, including years of experience with epilepsy clinical trials, subspecialty board certification or other characteristics of site investigators and staff, number of screened participants later deemed to be ineligible by independent adjudicators, proportion of missing data, and other quality metrics. For the same reasons of privacy and identifiability, we also did not analyze the U.S. Food and Drug Administration's (FDA) primary efficacy outcome of median percent reduction in seizure frequency, which has higher statistical power than 50RR. ⁴¹

5. CONCLUSION

These associations of participant‐, study‐, and site‐level factors with placebo responder rate provide important information to guide the development of eligibility criteria for future RCTs as well as inform power analyses. The association of higher placebo responder rate with lower duration of epilepsy, lower concomitant ASMs, and lower baseline seizure frequency may be attributed to fewer markers of long‐term, presumably static, refractory epilepsy. However, the specific association of elevated placebo responder rate for participants with a baseline of six or fewer baseline seizures raised concern for regression to the mean. Furthermore, the finding that certain countries had a higher placebo responder rate was concerning for a potential vicious cycle of challenges in recruiting participants, leading to recruiting from new centers, but these new centers limited statistical power by having elevated placebo responder rate.

AUTHOR CONTRIBUTIONS

Dr. Kerr performed the final analyses of the data and drafted the manuscript. Dr. Suprun performed the initial data analysis, supervised by Dr. Bagiella. Mr. Kok, Mr. Reddy, and Ms. McFarlane assisted in statistical comparisons. Dr. French supervised and provided input at all stages of the project. All authors reviewed manuscript drafts and approved the final version of the work.

CONFLICT OF INTEREST STATEMENT

No pharmaceutical company contributed to this manuscript in any stage of development. Dr. Kerr received personal compensation as Associate Editor of Epilepsia; writes review articles for Medlink Neurology; is a paid consultant for SK Life Sciences, Biohaven Pharmaceuticals, UCB Pharmaceuticals, Jazz Pharmaceuticals, Cerebral Therapeutics, Ventus, Epygenix, Harmony, and Epitel; and has collaborative or data use agreements with Eisai, Janssen, Radius Health, and GSK. Dr. Suprun is an employee at Janssen Pharmaceuticals. Mr. Reddy, Mr. Kok, and Ms. McFarlane have no relevant disclosures. Dr. French receives salary support from the Epilepsy Foundation and for consulting work and/or attending scientific advisory boards on behalf of the Epilepsy Study Consortium for Aeonian/Aeovian, Alterity Therapeutics Limited, Anavex, Arkin Holdings, Angelini Pharma S.p.A, Arvelle Therapeutics, Inc., Athenen Therapeutics/Carnot Pharma, Autifony Therapeutics Limited, Baergic Bio, Biogen, Biohaven Pharmaceuticals, BioMarin Pharmaceutical Inc., BioXcel Therapeutics, Bloom Science Inc., BridgeBio Pharma Inc., Camp4 Therapeutics Corporation, Cerebral Therapeutics, Cerevel, Clinical Education Alliance, Coda Biotherapeutics, Corlieve Therapeutics, Eisai, Eliem Therapeutics, Encoded Therapeutics, Encoded Therapeutics, Engage Therapeutics, Engrail, Epalex, Epihunter, Epiminder, Epitel Inc., Equilibre BioPharmaceuticals, Greenwich Biosciences, Grin Therapeutics, GW Pharma, Janssen Phamaceutica, Jazz Pharmaceuticals, Knopp Biosciences, Lipocine, LivaNova, Longboard Pharmaceuticals, Lundbeck, Marinus, Mend Neuroscience, Marck, NeuCyte Inc., Neumirna Therapeutics, Neurocrine, Neuroelectives USA Corporation, Neuronetics Inc., Neuropace, NxGen Medicine Inc., Ono Pharmaceutical Co., Otsuka Pharmaceutical Development, Ovid Therapeutics Inc., Paladin Labs, Passage Bio, Pfizer, Praxis, Pure Tech LTY Inc., Rafa Laboratories Ltd., SK Life Sciences, Sofinnova, Stoke, Supernus, Synergia Medical, Takeda, UCB Inc., Ventus Therapeutics, Xenon, Xeris, Zogenix, and Zynerba. Dr. French also has received research support from the Epilepsy Study Consortium (Funded by Andrews Foundation, Eisai, Engage, Lundbeck, Pfizer, SK Life Science, Sunovion, UCB, and Vogelstein Foundation), the Epilepsy Study Consortium/Epilepsy Foundation (Funded by UCB), GW/FACES, and NINDS. She is on the editorial board of Lancet Neurology and Neurology Today. She is Chief Medical/Innovation Officer of the Epilepsy Foundation. She has received travel reimbursement related to research, advisory meetings, or presentation of results at scientific meetings from the Epilepsy Study Consortium, the Epilepsy Foundation, Angelini Pharma S.p.A., Clinical Education Alliance, NeuCyte, Inc., Neurocrine, Praxis, and Xenon. Dr. Kwan was supported by the NHMRC Investigator Grant (GNT2025849). His institution has received research grants from Eisai, Jazz Pharmaceuticals, LivaNova, and UCB Pharmaceuticals unrelated to this work; he/his institution has received consultancy fees from Angelini, Eisai, LivaNova, SK Life Science, and UCB Pharmaceuticals unrelated to this work.

ETHICS STATEMENT

We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

Supporting information

Figure S1. Multivariable logistic regression demonstrating all associations with placebo 50% responder rate (50RR, *p < .05). For a selection of the significant associations, refer to Figure 1.

Table S1. Associations between participant‐, site‐, and study‐level characteristics associated with 50% responder rate. For continuous variables, the median is listed when an IQR is provided and the mean when an SD is provided. The median is listed for continuous variables. The denominator for each characteristic was the number of either non‐responders or 50% responders (e.g., 0.9% of non‐responders were from Africa). Significance was based on Fisher exact tests, Wilcoxon signed‐rank tests, and chi‐square tests of independence. For characteristics without significant associations and multivariable associations, see Table 1 and Table S2. Abbreviations: number of participants (n), standard deviation (SD), interquartile range (IQR), frequency (Freq).

Table S2. Exact values of the multivariable associations between characteristics and placebo 50RR. ORs >1 indicate increased placebo 50RRs. Abbreviations: 50% responder rate, 50RR; days d; OR, odds ratio.

Table S3. Placebo 50RR for each country, and the one‐vs‐all logistic regression p‐value for difference from reference. Bold underline indicates countries grouped by mixture modeling into a higher placebo response group. In mixture modeling, the log likelihood of two2 groups was 99.54 (3 degrees of freedom) and the log likelihood of 3 groups was 97.67 (5 degrees of freedom). Therefore, there was insufficient evidence to suggest that there were 3 groups (deviance difference, p = .29). Abbreviations: 50% responder rate, 50RR; Fisher exact 95% confidence interval by country (95% CI).

Table S4. Association of placebo 50RR with BSF per 28 days, grouped within integer bins, and the significance of the difference between 50RR based on thresholds above (≥) or below (<) the chosen baseline seizure frequencies. Mixture modeling indicated two groups of 50RR (indicated in bold), with the 50RR of participants with six or fewer PSC being 29% compared to 21% in participants with seven or more PSC. Abbreviations: 50% responder rate, 50RR; BSF, baseline seizure frequency; CI, confidence interval; OR, odds ratio; PSC, pre‐randomization average monthly seizure count.

Kerr WT, Suprun M, Kok N, Reddy AS, McFarlane KN, Kwan P, et al. Factors associated with placebo response rate in randomized controlled trials of antiseizure medications for focal epilepsy. Epilepsia. 2025;66:407–416. 10.1111/epi.18197

DATA AVAILABILITY STATEMENT

The individual‐level data for this analysis were provided by each pharmaceutical company according to their individual policies. Requests for data access should be directed to these companies.