Berlin Startups Discuss Machine Learning to Support Healthier Outcomes
“The whole is greater than the sum of its parts.”- Aristotle
Knowledge Sharing is Hot
Berlin’s thriving tech scene is actively becoming a unique counterpoint to Silicon Valley. There’s an undeniable energy here to solve meaningful problems and impact people’s lives through technology. This is particularly true in the digital health sector. Despite the typical race to succeed across a competitive market, knowledge sharing helps everyone get ahead. Collective brainstorming can lead to greater advances in science and health. This is clear from the many Meetups and Fireside Chats happening around the city.
In this spirit, on a recent Friday evening in Kreuzberg, two Berlin-based health startups teamed up to discuss the role of machine learning in digital health. The Meetup’s lengthy wait list alone indicates how interested locals are in this tech topic.
Clue, makers of a popular menstrual cycle tracking app by the same name, graciously hosted the event to engage the Berlin community in shared learnings. Clue invited two guest speakers: Adrin Jalali from Ancud IT and Eden Duthie, Head of Data Science for Ada Health, just down the street from Clue.
Clue’s Data Scientist Marija Vlajic opened the discussion explaining that Clue uses scientific data to inform, and connect with, their app users about menstrual health. They’ve been collecting powerful data based on tracking menstrual cycles that, through partnerships with medical researchers, provides valuable insights into female health.
Adrin and Eden then each illuminated how machine learning can positively impact medical treatment, improve access to medical information, and promote healthier outcomes. Despite different angles, each speaker jumpstarted discussion on how collecting and analyzing data will support health-related decision-making in the long run.
Here’s a look at how.
Cancer: Sorting out Data in a Race Against Time
Adrin’s expertise stems from collaborations with oncologists and cell biologists in handling patient data and corresponding diagnoses, and interactions with computer scientists at the Max Planck Institute for Computer Science in Germany.
He presented stark numbers: In 2015, circa 90.5 million people had cancer and 8.8 million people died of cancer, accounting for about 15.7% of all human deaths. Every day doctors and patients face hard decisions about cancer treatment and care. Adrin looks at how big data can help alleviate this situation by increasing knowledge about cancer and ultimately using that to inform the diagnostic process.
He broke down the life cycle of a cell in an understandable way, explaining how cancer cells grow uncontrollably, pushing away normal cells and demanding body resources such as blood vessels. When he describes cancer cells as ‘smart, agile, and resilient’, it becomes clear we need an even smarter response to them. Although he says DNA mutations happen all the time, in cancer they occur much more.
He’s looking for changes that persistently affect cancer cells, but not normal cells.
One of the leading ways to study cancer cells is by measuring ‘cell surface markers’, which helps to distinguish between normal and cancer cells, and to identify subtypes of cancer. This type of research quickly generates big data, as scientists sequence DNA and study hundreds of patient’s cells.
So he’s asking: what can be learned from all this data? Are there clustering of treatments that work? Adrin says the most pressing need are methods that can be reproduced on new data.
If using big data can provide knowledge about how cancer ticks, perhaps more informed decisions regarding cancer care can emerge. Timing here matters; since cancer can spread to different body regions, early detection and treatment is crucial.
Patterns in the Sky, the Race Track, and now Health
Back in Australia, Eden Duthie used to think a lot about clouds and horses.
He led teams in applying data science to weather forecasts, sports betting, and even drug discovery. Eden is interested in pattern recognition — how to automate it and scale it. He came to Berlin to apply his passion for machine learning to health at Ada through our personal health companion app. Ada’s mission to bring quality, personalized healthcare to the world and use tech to improve people’s lives resonates with him.
Eden explained how the Ada app builds a personalized health assessment based on a user’s symptoms. This smart assessment, formed by answering clearly-worded questions, can be shared with a user’s doctor if desired. This potentially saves doctors time in gathering information and enables more face time to explore treatment or any complex medical solutions.
The Ada app draws on a medical knowledge base that doctors and software engineers have developed over the past 6 years, recognizing even rare conditions. It uses a probabilistic model to determine a user’s likely conditions, or closest matches.
The accuracy has so far been rigorously tested by comparing the app’s results to how a doctor would assess the same set of symptoms. So far, it has proven that Ada can keep pace with a doctor’s analysis and support them in the diagnostic process.
With the completion of one million health assessments so far, Ada is actively learning from patterns in user data whilst enabling helpful analysis of how symptoms and diagnoses relate.
Enter machine learning for improved app performance: One of the first areas where machine learning has led to advances in Ada’s capabilities is triage — figuring out the level of care required, and providing information on appropriate next steps. Ada’s triage takes into account current best practice and evidence based guidelines. However, through machine learning Ada is also being trained to weigh up the nuance of each individual case much like an experienced doctor.
Eden says that as it gains more data, the app will continue to grow in confidence. He says machine learning is a way to improve accuracy and perhaps predict what factors influence health. Through various techniques, including supervised learning aided by a large pool of experienced physicians, Ada is developing steadily.
Tending to Data Privacy, Security, and Collection
Wrapping up, attendees asked probing questions about the future of machine learning. They wanted to know, as does everyone, where is it heading? But the answer is still forming. Curiosity about protecting data privacy and security as well as maintaining high standards of data collection was understandably high. Here is a breakdown of what these important terms mean in digital health.
With 70 people attending this Meetup on a warm Spring weekend night, it’s fair to say people care about the direction and possibilities of machine learning for digital health. The Berlin community is alive and growing and at Ada, we are happy to be a part of it. Check back for more on machine learning or leave a comment with any advances or insights you want to share.
Thanks again to Clue for organizing this forward-thinking event!