Abstracts

Rationale: Current randomized clinical trial (RCT) design involves eligibility criteria in order to reduce the placebo response, yet the placebo response has been observed to be rising over the years. The ability to predict the placebo response of patients enrolled in RCTs before they are administered drug dosages would be useful to trialists. In this project, a statistical model of monthly seizure count data was used to predict the placebo response of individual patients in RCTs based on their seizure frequency (average of monthly seizure counts) and seizure standard deviation (standard deviation of monthly seizure counts).  Methods: A 2D heatmap of the expected placebo response was generated, with each seizure frequency and seizure standard deviation pair on this heatmap corresponding to one individual patient with that set of characteristics. For each seizure frequency and seizure standard deviation pair, the placebo response was calculated by averaging the placebo response of 100 trials, where all patient populations were homogenous in their seizure frequency and seizure standard deviation. Seizure counts for each patient were generated via a negative binomial (NB) distribution. No eligibility criteria and no psychological effects were applied to these simulated patients. This was done for both the 50% responder rate and the median percent change.  Results: The expected placebo response was plotted on a graph of seizure standard deviation vs. seizure frequency for both the 50% responder rate (Figure 1) and the median percent change (Figure 2). The placebo response could not be calculated for patients who have a seizure standard deviation which is lower than the square root of their seizure frequency due to mathematical restrictions from the NB distribution. Figure 1 shows that for the 50% responder rate, the placebo response is inversely proportional to the seizure frequency and is proportional to seizure standard deviation. Figure 2 shows that for the median percent change, there are large regions of the map where the median percent change does not change based on either seizure frequency nor seizure standard deviation. In one of the boundaries between these regions, the median percent changes drastically while the other boundary transition has a more gradual gradient.  Conclusions: The significance of this study is that the placebo response of an individual patient in an RCT can be estimated by calculating two simple statistical quantities derived from their seizure diary, which could provide guidance to clinicians as to whether or not a patient should continue in a trial after a screening period.  Funding: This study was funded in part by Beth Israel Deaconess Medical Center.