Predicting the placebo response

The placebo effect is a fascinating but poorly understood phenomenon that often occurs in clinical development. The majority of randomised controlled clinical trials will have a placebo arm where patients will be given a placebo, often a sugar pill, to see if the drug is having an effect.

Patients in the placebo arm will not know they are being given a fake drug but sometimes, just thinking that they are getting the real treatment will have an impact on their disease. In other cases, patients receiving the real drug can also be affected by the placebo effect.

While a patient becoming better (or perceiving they are) by simply believing they had the drug is great for the individual patient, on the whole, the placebo effect is a nightmare for clinical research.

Belgian clinical trial tech firm Tools4Patient has developed Placebell, a tool to help predict which patients are likely to be placebo-responsive by assessing the personality and psychological traits of an individual patient along with some other factors associated with placebo response.

Placebell combines machine learning-based algorithms with individual patient personality data to predict the individual placebo response in a patient before the first dose of a drug is administered to estimate the amount of benefit that patients in a clinical trial are likely to get that is purely attributable to the placebo effect. The predictive tool aims to improve success rates with no statistical or operational risk to a study.

The firm recently presented data at the MDS Congress 2021 that predicted the placebo response in Parkinson’s disease (PD) in a multi-centre, multi-national clinical study.

The trial saw 94 people with Parkinson’s given an oral treatment for three months, which they were unaware was a placebo.

The researchers then constructed mathematical models of the data, using factors like disease severity at the start of the study and patients’ psychological traits. By applying these models, they then calculated how much variability might be removed from patient responses. For example, in part three of the Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS), the part of the test that evaluates motor symptoms, the models were estimated to remove 33.2% of the variability.

We spoke with Tools4Patient CEO Dominique Demolle to discuss the placebo response phenomenon, the need to take into account individual’s personalities in clinical trials and how Placebell is helping avoid trial failures due to the placebo response.

The placebo response is a very complex phenomenon that encompasses different components. For example, the site investigator may have an influence on the placebo response related to empathy, where they take care of the patient and really raise their expectations. Then there is the design of the trial. If you have a trial where there is a placebo arm or a number of treatment arms, again you may raise the expectation of the patient that they have a higher chance to have the drug.

Apart from that, you also have the specific nature of the patient. It is well described in the literature that depending on your personality traits, you can be more likely to be a placebo responder.

For decades there has been investigation into the relationship between psychology and the placebo response – we call that the placebo effect. There is a connection between the psychology and what is happening biologically in your body. In pain for example, If you are a placebo responder, your brain will induce its own production of endorphins – so there is really something psychobiological happening.

Diseases like Parkinson’s and psychiatric disorders like depression and schizophrenia are also affected by this phenomenon. If this effect is predominant, then you are at risk of trial failure because you will have difficulties making the distinction between this placebo response and effect compared to the effect of the drug, and it may happen at a very late stage, even in Phase III.

The concept we had when developing Placebell was based on the fact that a key component, the psychology of the patient, so far has not been taken into account to evaluate this placebo response in clinical trials.

A lot has been done by academic groups, but we observed nothing that was implementable in clinical studies. In a trial, there are a lot of variabilities, and we measure a lot of objective things about the patient like biomarkers, imaging and blood pressure. Everything except what we see as a key element – the personality.

We realised if we could really have access to that information to reduce variability in the clinical trial, it would be a tremendous plus for the analysis of the data. With Placebell we collect the psychology of the patient with a simple questionnaire and then we used machine learning to see if we could really link that to the placebo response.

Through an algorithm, we are able to associate each patient participating in a trial with a placebo score. Just like you would have the age, weight, gender of the patient – a unique value associated to a patient which you may take into account in your statistical analysis. It’s like using any covariate in statistical analysis.

Of course, we will not disclose the IP but there are clearly traits that have been associated to the placebo response in the literature, like optimism. But, in reality, it is a combination of things – it’s the personality, the level of expectation, the social support.

With Placebell we collect different types of data that combined together are going to predict which patient is going to be a placebo responder or not. This is where machine learning is important because it’s a lot of information that you have to process and at the end of the day you need a single score.

When we developed Placebell we conducted our own trial. This is how we developed the algorithm. Then we conducted validation by applying this algorithm in independent studies, to make sure that the models we had were really predictive of the placebo response and know that those models are defined to be applied in the next coming trial. So, we are not developing the model as we are conducting the trials.

We have predefined models that we apply in a coming study and this is very important because the fact that it’s predefined means you don’t have bias.

Parkinson’s disease is one of the diseases, particularly now with the generation of disease-modifying drugs, where you have a placebo response, especially in the early stages of the disease. This is a handicap for new treatments in development. When we started the development of Placebell in PD, we connected with different investigators and key opinion leaders in that area to embark on the project with us.

Based on the work we had done in pain, they were very enthusiastic. Together we completed a multi-site trial in the US and Europe with a protocol that was developed in partnership with those academic centres. Thanks to that we have been able to mimic what we had done in pain, which was to identify those key parameters that were related to placebo response prediction in PD, and these results we recently released at the MDS Congress.

It was extremely encouraging for that community to see that progress could be made in placebo response management in PD, in addition to work that is already ongoing to try to tackle this big issue, especially for a disease like PD which is devastating.

When this placebo response happens in Phase II, for example, you may disregard the compound. If you have tactics that help to reduce that risk, then it’s a fundamental advantage. Firstly, to not repeat a non-conclusive trial.

Secondly, to not kill a good compound, but certainly then to equip you with the information to move to Phase III. I would say that in a disease like PD, the placebo response starts to be really damaging in Phase II. Then, of course, you have still to manage the placebo response because it may still, unfortunately, happen in Phase III.

When you start your clinical study, you define the statistical power you would like to achieve and from there you calculate your sample size. What Placebell does is reduce the variance.

Let’s assume you define the trial at 80% power, and to achieve that without Placebell you need 100 patients. If you apply Placebell, because you are going to reduce your noise you can improve your statistical power and you will need fewer patients to achieve that power. This is really the power of the technology.

If you wanted an absolutely no-risk approach, you could keep your 100 patients and because of the additional information that you have brought thanks to Placebell and the variance reduction, the power would be more like 92%.

It is great for any indication where pain is one of the efficacy endpoints like rheumatoid arthritis or irritable bowel disease. But now we are being contacted by companies running programmes in areas like immunology and psychiatry, [the latter of which] which is clearly an area where placebo response has been a big challenge.

The technology is so simple, it’s a questionnaire. You administer the questionnaire to the patient once in the trial – that’s it. In my opinion, patients should be given the questionnaire for all trials.

We are so influenced by our behaviour, our sense of self. It’s very surprising that in clinical research, the only component that we don’t take into account so far is the psychology of the patient.

We can really influence our own health, or the progression of the disease and there have been tremendous studies so it’s just unbelievable it’s not taken into account, especially now that we are moving to decentralised clinical studies.

We are human and who we are will influence the way we are going to behave in a study. When you have trials where some part of it is decentralised and some part is traditional, inevitably there will be some variability, so you need to get a sense of what is there that may induce that and really characterise that better.

When you have a baseline covariate and take that into account it will reduce the variability and noise coming from the placebo effect and will improve study power, decreasing your risk of failure. You’re in a better position to see a difference, if there is a difference, between the placebo group and the treated group. It is very simple to implement.