Prediction of placebo response in Osteoarthritis improves estimation of the treatment effect: Impact on drug development

Placebell©™, developed by Tools4Patient, is a proven method to predict the placebo response in chronic pain diseases. The technology is built on a machine learning-based model and reduces the impact of the placebo response in clinical trials regardless of therapeutic area. This approach considers multiple factors on an individual patient basis, including demographics, medical history, baseline disease intensity and certain previously identified important psychological traits as measured by our proprietary questionnaire, the Multi-dimentional Psychological Questionnaire (MPsQ). These data are integrated using machine learning to calculate a baseline covariate for each patient that is highly predictive of the placebo response. It can be used in statistical analyses to adjust for the spectrum of placebo responsiveness in a clinical trial population. This adjustment can used to improve the discrimination between drug treatment and placebo treatment in randomized controlled trial to increase drug development success rates.  Here, we review recent results obtained using the chronic pain Placebell©™ model in a randomized controlled trial (RCT) in osteoarthritis (OA). 

OA is common musculoskeletal disease associated with aging. It is mainly developed in knee, hip and hand joints as well as facets joints of the spine. As a degenerative disease, OA can evolve asymptomatically over a prolonged period (up to 20 years) before patients complain of pain and stiffness in the affected joint(s). Where these symptoms can be worsened with activity. Development of pharmacological treatments, following long-term administration in randomized clinical trials (RCTs), showed a large amount of uncertainty around estimation of efficacy (pain improvement) compared with placebo (1).  

Assessment of treatment efficacy (both historical and experimental treatments) are known to be blurred by the placebo effect. Indeed, the amount of reduction in pain, and improvement in function and joint stiffness, associated with the placebo response could reach respectively 75, 71 and 83% of the treatment efficacy (2). In addition, the range of pain reduction can vary from 47% after intra-articular corticosteroid injection to 91% after joint lavage (2), although, the size of these effects varies widely between clinical trials or published meta-analyses. The spectrum of placebo responsiveness in a clinical trial population creates additional complexity as it further increases variability in the data. Based on this, understanding the drivers of variability in the placebo response between patients can lead to development of predictive models of placebo response.  With this information, variability can be reduced, which would improve interpretation of RCTs and accelerate the development of new treatments.  

Placebell©™ can be used in OA and similar diseases in which efficacy is characterized using patient-reported outcomes to reduce the interference of the placebo effect and improve assessment of the drug effect. The performance and applicability of PlacebellI©™ has recently been demonstrated in a Phase 2 RCT conducted by a biotech sponsor in subjects with moderate to severe painful knee OA. The trial aimed to assess the efficacy of single-dose intra-articular (IA) injection of an experimental drug compared to placebo. Efficacy was evaluated using the change from baseline to 3 months in WOMAC-Pain (primary), WOMAC physical function, WOMAC-stiffness and average pain score. The machine learning-based model was built following a gaussian process with a linear kernel using data from placebo-treated patients from several chronic pain studies conducted by Tools4Patient and was pre-specified in the statistical analysis. The predictive performance of the model was tested both in the placebo-treated group and in the entire population by comparing the predicted response to the actual observed response using Pearson’s correlation. This correlation was shown to be statistically significant at p<0.001 for all primary and secondary endpoints when considering both placebo-treated patients and all patients. The Placebell model explained 35 % of the variance related to the placebo response for WOMAC-Pain in placebo-treated patients and ranged between 21% and 33% for the other endpoints evaluated.  Placebell©™ further demonstrated an improvement in the precision of the estimation of treatment effect by almost 40%. 

In conclusion, Placebell©™, when applied to RCT in OA chronic pain, significantly predicts placebo response on primary and secondary endpoint (p<0.001) with a decrease of the variance associated to the placebo response by 35% (WOMAC pain; the primary endpoint). The integration of the baseline covariate in statistical analysis increased the precision of the estimation of the treatment effect by 40%. Based on this, Placebell©™ has been proven to be a valuable method to improve efficacy evaluation in randomized controlled trials in both the placebo and drug-treated patients. Placebell©™ is being developed and applied in diseases and indications ranging from pain to neuropsychiatric disease to ophthalmology.

Register to Tools4Patient upcoming webinar


1.  Dario Gregori et al. Association of Pharmacological Treatments With Long-term Pain Control in Patients With Knee Osteoarthritis A Systematic Review and Meta-analysis. JAMA. 2018;320(24):2564-2579. doi:10.1001/jama.2018.19319  

2.  W. Zhang The powerful placebo effect in osteoarthritis. Clin Exp Rheumatol 2019; 37 (Suppl. 120): S118-S123.