Abstracts

Rationale: Clinical trials for new epilepsy therapies rely heavily on self-reported seizure counts, although some studies suggest they underestimate seizures. Most trials find evidence of improved seizure frequency in both placebo and therapy arms. One meta-analysis noted more marked placebo improvement in pediatric patients than adults. Apparent improvement in placebo arms has been considered a combination of regression to the mean, classical conditioning, and other psychological factors including true placebo effect. We investigated patterns consistent with placebo improvement, but devoid of any true placebo. Methods: Data from seizure diaries entered into seizuretracker.com from its inception in November 2007 until April 2014 were studied, including 684,825 seizures from 8,569 patients. Patients were "enrolled" in a simulated clinical trial involving patients randomly placed into "placebo" and "therapy" arms. Time 0 was identified as the first seizure documented per patient. Seizure counts over a 3-month baseline period and a 3-month test period were measured, with a 1 month "ramping up" period between the two. Patients needed at least two and at most 900 seizures (10 seizures/day) during the baseline to be eligible. The simulation was repeated 673 times, each representing a trial performed after waiting a specific number of days after the first recorded seizure (X axis in the figure), covering nearly a 2 year time span. Results: A "placebo improvement" was seen throughout the 2 year period, though most prominently in the first 3 months--seen in both the 50% responder rate and the seizure freedom rates. For trials starting at time 0, the 50% responder and seizure free rates were 45% and 20%, respectively. The effect was more pronounced in patients under age 18 than adults, regardless of simulation starting time. The seizure freedom rate accounts for a portion of the responder rate variability, though a relatively constant 20% of patients improve in their seizure frequency without achieving seizure freedom throughout all simulated trials. Conclusions: Strong placebo improvement was demonstrated in simulated placebo controlled clinical trials. Although regression to the mean may play a role for the first 3 months, additional factors improve seizures even two years after time 0. No psychological factors typically associated with a clinical trial could have been at play, including a patient-provider interaction and classical conditioning. Furthermore, patients that stop using seizuretracker.com cannot account for ongoing improvement noted across time, because the difference between responder rate and seizure free rate remains high throughout. That difference appears to be a relatively constant fifth of the study population, regardless of when the simulation is run. Additionally, the difference between pediatric and adult patients seen in other clinical trials appeared in these simulated trials as well, suggesting a common mechanism. Seizure recording itself may have therapeutic benefit accounting for a significant proportion of placebo improvement.