Knowledge discovery through Neuroscience Data Integration

min. read
July 7, 2019

Knowledge discovery through Neuroscience Data Integration

Our CEO Stephen Larson hosted a webinar with Dan Karlin, a psychiatrist, former drug developer for Pfizer, currently General Manager of HealthMode. 

Knowledge discovery through Neuroscience Data Integration

In the past few years, wearable devices that can collect medically relevant data has formed a global market which is projected to reach $12.1 billion by 2021, with the US being the largest market. Such an exponential rise in data-generating devices naturally goes together with the astronomical growth of healthcare data.

In this new era it is consumers who directly collect data about health or disease management through digital health technologies. By sifting through these data, data scientists are in hot pursuit of so-called "digital biomarkers", or signatures of health-related conditions, which ultimately are hoped to be able to explain, influence or predict health-related outcomes.

Of particular interest are those digital biomarkers that can help in the area of CNS-related drug discovery, as diseases like Alzheimer's continue to be difficult to develop effective drugs for.

The topic is highly relevant to MetaCell, and recently our CEO Stephen Larson hosted a webinar with Dan Karlin, a psychiatrist, former drug developer for Pfizer, currently General Manager of HealthMode. 

The following Q&A is a short summary of that live talk - which you can watch in full here.

SL: Hi Dan, thank you for your time today. I’d like to start with a challenging question... What do you think the impact of digital biomarkers and wearables is going to be in the healthcare and pharma industries?

DK: Digital biomarkers are expected to significantly help in bringing healthcare from a reactive towards a more preventive approach, as researchers will not only be able to explain diseases better, but more and more data will be available to analyze what healthy, normal states signify and to predict future health outcomes.

In Neuroscience in particular, we’ve come quite a long way over the past 5-6 years in improving the intersection of technology, drug development, measurement, and illnesses. But a lot more can and should be done, especially on how data is used to drive forward the science, drug discovery, and clinical care.

SL: Love how you set the scene… tell us more!

DK: Take measurements for drug development - they haven’t progressed as much over the past few decades, whilsts costs keep increasing. In neuroscience especially, we have run out of low hanging fruits, or new minerals that save lives, however there’s an evolving requirement for drugs to be safer and safer, and more effective, so development costs and are going up and timelines are getting longer.

SL: What can be done to improve measurements?

DK: The two points above means nothing on their own. They’re data, but only de-contextualized points. If we know they’re weights, that gives us a bit of context so we start to be able to turn data into information.

But for example, if we look at scales - especially modern ones that store data online - we get pretty accurate timelines, and suddenly the same points with context look a lot different!

However you could pick two different days, a few weeks apart, and draw very different conclusions about change over time. See below.

If our goal is to demonstrate the effect of a drug on weight, at this point we’re in trouble. The reality of the world differs from what we measure, therefore we must change the way we observe data in clinical trials. Understanding the wider context is key.

With all of today’s connected measurement devices, I ask myself - can we now truly and granularly measure patients in their lives? How do we turn data into meaningful information and ultimately knowledge?

SL: Great points. Help us understand how we can contextualize data effectively.

DK: This one above is the concepts of modern analytical techniques. Start from huge or dense data sets, use a certain domain knowledge to understand what features you might be interested in, extract those features, and use them to model the clinical phenomenon you’re interested in.

The ultimate goal is to understand how people feel, which is an incredibly complex problem to solve.

SL: What do you think is the additional potential for chronic monitoring and developing better therapeutics in the future?

DK: I’d refer to mostly three aspects.

First. Even illnesses such as Alzheimers and Parkinsons - which are chronic, progressive debilitating illnesses, which means that their course gets worse over time - have much more complicated patterns than we realise. Continuous monitoring can help us gather more realistic and accurate data.

Second. We need to monitor subjects for enough time both on and off drugs to have meaningful data from which to draw conclusions.

Third. Monitoring subjects in a more natural context would be better because it would lead to measurements that have more relationships to their lives and normal tasks versus what we do now where, for example with Parkinson's patients, who may have substantial difficulty with movement, must come into clinic visits, which is both a hardship and a potential source of measurement bias.

SL: What do you think is the space for combining wearable devices with intervention (e.g. deep brain stimulators chronically implanted)?

DK: It’s early days. In neuroscience we’ve seen the first novel modern digital device and drug combination approval but despite a 3-year development effort, the sponsor was unable to get any new label claims and instead go there new warnings related to device reliability. So yes, we will see more combinations emerge but it’s not clear when this will be or how much efficacy will be gained through combinations.

SL: Finally, how do you manage complexity when you’re trying to assemble a big picture of multiple datasets from a collection of wearables?

DK: The time alignment and metadata tagging aspect is difficult, data storage not so much. Integrating different sensor streams across different sensing modalities is key. And as I was saying at the beginning, to make sense of the data by understanding the context, look at correlations, and ultimately knowledge will emerge.

Share this post online