Digital biomarkers: a new way to look at diseases?

We usually use biomarkers from body tissue or blood to diagnose diseases and monitor their progression. This requires taking and analyzing samples from patients at regular points in time. Two new studies shed light on an easier and less expensive method: using wearable sensors to collect movement data and AI to analyze them.

In this interview with MEDICA-tradefair.com, Prof. Aldo Faisal talks about the use of movement data as digital biomarkers for Friedreich’s Ataxia and Duchenne Muscular Dystrophy and the application AI to predict the progression of these diseases. In the future, this could even predict the onset of new ones.

Prof. Faisal, you have researched digital biomarkers. What exactly are they?

Prof. Aldo Faisal: Biomarkers, in a conventional sense, are a way to measure the state of health. This is usually done by taking a tissue or blood sample. Digital biomarkers do not require physical sampling. Instead, data is used, the output of digital devices or sensors like motion sensors. However, currently, digital biomarkers are often not better biomarkers, they are just reproductions of conventional measures with digital means. We believe Artificial Intelligence can revolutionize that.

Why should we be interested in digital markers?

Faisal: They can become important for anyone who develops new treatments, be it pharmaceutical or technology-based. You need to prove to the regulators that a treatment works in a clinical trial to bring it to the market. If the state of the patients in a trial changes, this is shown by biomarkers. This is the reason why studies are very expensive: You need to collect a lot of samples from many patients for a long time and analyze them. Existing biomarkers are imprecise, it takes up to two years to detect if a patient improved with treatments.

There are about 9,000 diseases, so-called rare diseases, for which currently we do not have therapies or the therapies we have are not very good and their development is expensive. There are no good biomarkers that show the progression of these diseases well. In our studies, we used AI and wearable motion capture technology to discover digital biomarkers for two movement-related diseases that could be used instead of conventional biomarkers. Our markers are based on the study of natural behavior. The Greek word is ethology, and so our approach is called ethomic biomarkers – in analogy to genomics which is focused on genetics.

Which diseases are those?

Faisal: One of them is Friedreich's Ataxia. It affects energy production in cells, so people are less and less able to move. It can impact the brain and heart as well. It is typically adults between 20 to 40 who get this disease. The other is Duchenne Muscular Dystrophy. It gradually destroys all muscles. Patients often die by age 25.

Our AI algorithm has looked to understand movement behavior in healthy people to then discover what we called ‘behavior fingerprints’ of these diseases. We are now able to effectively detect disease progression in half the time compared to the gold standard biomarker used in the pharmaceutical industry. We can also do this with a much smaller number of patients. This means that ethomic biomarkers can accelerate clinicals trials, make them smaller, cheaper, and thus less risky. This will help a lot of parties that are interested in developing new treatments, even for rare diseases, at much lower cost and risk.

What kind of technology did you use to measure movement?

Faisal: We used wearable sensors, because currently, this is the best technology available to collect movement data. But for our ethomic biomarker it is not important what sensor technology we use, as long as we get movement data. The measurement could as well be based on camera, radar, or any other form of motion capture technology.

How were your studies conducted?

Faisal: We are a team of AI researchers, biomedical engineers, and clinicians. We collaborated closely with clinical researchers. One is Prof. Richard Festenstein from Imperial College for Friedreich’s Ataxia. The others are Prof. Thomas Voit and Dr. Valeria Ricotti from Great Ormond Street Hospital for Duchenne.

At the beginning of the studies, we had a standard clinical assessment with each participant. As part of this assessment, we used wearable sensors, but the patients did not only wear them during the hospital assessment, but also during daily life. We collected their movement data at three points in time: at the beginning of the studies, after nine months and 18 months. During this time, we observed how their conventional biomarkers developed.

Afterwards, we also used our digital biomarkers to predict disease progression based on the digital behavioral data. For Duchenne, we could predict the exact time course of disease progression for every single patient individually from the data collected on day one, that is because daily life behavior is a very rich data source.

For Friedreich’s, we were able to do the same and even more: because it is a genetic disease that affects the amount of the protein called Frataxin that is made in cells. This amount changes with disease. We could predict from the movement data what those patients’ protein profile on a given day looked like, so whether they had more or less of the protein than another day. This means, for the first time, we can replace analyzing blood samples by analyzing behavioral data.

The process of making proteins out of DNA is called transcriptomics. We were able to do what we called Behavioral Transcriptomics: reading out behavior and working out what a gene activity profile looks like.

What value do you estimate to come from this for screening or monitoring disorders?

Faisal: The biggest immediate impact is going to be for anyone who develops therapies, pharmaceutical or other. We can cut the cost, time, and risk associated with development. Most drugs and therapies do not fail at the discovery stage, but during the development process, when they are taken into trials and their effect is not as big as expected in terms of conventional markers. My colleague Richard Festenstein who is developing such a treatment told me, "the patients can turn around in bed now", but the conventional markers cannot capture this improvement – that is why we need ethomic biomarkers and we are making these available as a service.

We also think that our technology can be useful for diagnostic purposes, as it captures movement data all the time and can detect patterns that humans would not notice. In fact, the next work that we are going to publish is about diagnostic purposes.

Ultimately, the ethomic biomarker technology could end up monitoring all of us for the onset of disease. We are currently moving beyond movement-related diseases and conducting trials regarding mental health. With the appropriate methods, this technology could even be applied to cardiovascular diseases or symptomless brain cancers.