May 6, 2021

Predicting the placebo response in OA to improve the precision of the treatment effect estimation

The current abstract is the result of a collaboration between Unity biotechnology and Tools4 patient As you know, the placebo response is one of the major sources of variability in randomized clinical trials As a result, many trials fail due to a high placebo response. Unfortunately, this is also true in OA studies. However, within Tools4Patient, we have developed a Machine-Learning model named Placebell©™ that can overcome this placebo challenge. This machine -learning model is predictive of the placebo response of subjects participating in RCTs. In other words, for each subject, the model associates a placebo covariate that represents the ability to be a placebo responder. The objective of this abstract is to present the results obtained in a phase two study where we used the technology to improve the estimation of the treatment effect. The UNITY study was conducted to investigate the safety and efficacy of a single dose intra articular administration of a new compound. One hundred and eighty patients with moderate to severe painful knee OA were enrolled in this study. Subjects were randomized in 4 different groups. The efficacy measurements were the change from baseline of the WOMAC pain score as well as WOMAC functioning sub-score, the average daily pain intensity and patient global assessment. 

The predictive models of endpoints were built with historical tools4patient data from chronic pain patient.  

Obviously, the models were fixed before any study analysis. In this study, using predictive model and baseline data of subjects , including psychological profile of patients, the model was able to explain almost 30% of the variance of the WOMAC pain score. The prediction of the model and the observed response were correlated at 53 %. The correlation was highly significant . In addition, the same type of correlation was observed for all secondary endpoints. Moreover, prediction of placebo response could be used to characterize subject population. As shown in this graphic, you have the evolution of the change from baseline of the WOMAC pain score in patients predicted to have high placebo response compared to patients predicted to have low placebo response using the Placebell model. The graph represents the change from baseline of the WOMAC-pain for both groups all along the study.  As you can see, there is a huge difference between the evolution of the average score between the two groups. The high responders in blue have a much larger response than low responders in grey. Finally, the predictions of the predictions of the Placebell model can also be used as a covariate to improve the estimation of the treatment effect. In this study, this adjustment increased by 40% the precision of the estimated treatment effect. 

To conclude: 

Tools4Patient has developed a machine-learning model predictive of the placebo response in chronic pain RCTs. This model has been applied successfully in a study conducted by Unity biotechnology. Highly significant predictions were obtained for all primary and secondary endpoints. In particular, the model explained 30 % of the WOMAC-Pain variance. This can obviously improve importantly the study power. 

Related Posts

Multimedia

Video Interview at Partnerships in Clinical Trials, Europe 2017

Dominique Demolle, CEO speaks with Andy Burrows of PCT TV about the early vision and development of Placebell©™...

Read More
Multimedia

Interview on The Bio Report Podcast

Erica Smith, VP of Business Development, speaks with Daniel Levin of The Bio Report Podcast about the Placebell©™...

Read More
Multimedia

Video Interview at Biotech Showcase 2019

Erica Smith, VP of Business Development, discusses the benefits of the Placebell©™ approach in video interview at Biotech...

Read More