Scientific Poster on February 2020 at the International Society for CNS Clinical Trials and Methodology (ISCTM), Washington DC, USA.
In analgesia randomized clinical trials (RCTs), the magnitude and the variability of the placebo response negatively impacts the ability to demonstrate superiority of active compounds compared to placebo. The first objective of this analysis was to investigate parameters influencing the placebo response in PNP as a way to control for this major confounding factor. A second objective was to assess the similarities between these parameters and those observed in other chronic pain diseases (in particular osteoarthritic pain). This is part of a larger effort to evaluate the feasibility of building a model of the placebo response generalizable to several chronic pain diseases.
Eighty-seven PNP patients were enrolled and given placebo for 4 weeks in a blinded manner. In addition, 48 patients with osteoarthritis received blinded placebo treatment (between 1 and 3 months) in independent trials were pooled. All patients were recruited from site database and completed a psychological questionnaire at baseline assessing several components of their personality. Conventional clinical pain assessment criteria, including daily average pain score and brief pain inventory, were also collected and combined with demographics and medical history. Finally, the placebo response was estimated as the difference in pain between baseline and end of the treatment. We modelled the placebo response from patient characteristics using a support vector regression (SVR). The predictive performances were estimated in Pearson’s correlation and predictive R-squared using a repeated random sub-sampling scheme. These performance metrics were estimated while comparing the predictions of the model with the actual placebo response of the patients.
For the analyses, two populations of patients were defined:
– PNP : the PNP patients only (N = 87).
– PNP + OA : the PNP patients and OA patients (N = 135).
SVR models were built and tested on both populations. These analyses aimed at testing whether the PNP placebo response was similar to the OA placebo response.
Overall, model performance was greatest when 12 features were used. The models learned in both populations were able to predict the placebo response of the PNP patients. The Pearson’s correlations were respectively of 53.7% for the model fitted on PNP patients only and 53.9% for the model fitted on the PNP and OA population. We also tested these two models on the full set of PNP and OA patients. The Pearson’s correlations were respectively of 49.3% and 53.5%. Both models were highly statistically significant on both populations with p-values lower than 0.001. Slightly better performance was obtained with the model fitted on both PNP and OA; however, these differences were not statistically significant.
These results demonstrate the similarity in the factors predictive of the placebo response in both PNP and OA patients. In particular, a model built on PNP patients was able to predict the placebo response of both PNP and OA patients. As a covariate, the model predictions could be used to reduce the impact of the placebo response-related variance in analgesia studies. This reduction of variance could lead to increased effect size and study power.