Microsoft Research’s Principal Research Manager, Danielle Belgrave, believes understanding heterogeneity is the key to unlocking personalized medicine

 

Danielle is Principal Research Manager at the Healthcare Intelligence group at Microsoft Research, in Cambridge (UK) where she works on Project Talia. Her research focuses on integrating medical domain knowledge, probabilistic graphical modelling and causal modelling frameworks to help develop personalized treatment and intervention strategies for mental health. Danielle obtained a BSc in Mathematics and Statistics from London School of Economics, an MSc in Statistics from University College London and a PhD in the area of machine learning in health applications from the University of Manchester. Prior to joining Microsoft, she was a tenured Research Fellow at Imperial College London.

 

“A lot of my research focused on understanding heterogeneity,” says Danielle Belgrave, Principal Research Manager of the Healthcare Intelligence Team at Microsoft Research UK. “Heterogeneity is a huge challenge when we are looking at effects of interventions on different diseases.” For example, as Belgrave cited, a group of patients with similar manifestations of symptoms might be given the same diagnosis and prescriptions. However, because of heterogeneity, these patients will not respond to the treatments in the same way and thus, progress differently over time.

“For some people, the drug might be toxic or not at all beneficial,” Belgrave explains. “Some might suffer from adverse side effects, while others are fine with having such a drug. As such, I want to know, what’s the distinctive underlying mechanism that’s responsible for heterogeneity? How can we understand subgroups of complex diseases and drug responses? All these are foundations for seeking better, more targeted medical interventions so that we can get the right treatments to the right people at the right time.”

Belgrave began her career as a statistician and has a background in probabilistic graphical modeling which allows her to study what’s happening to patients at a particular time based on their history of events. Interestingly, mathematics wasn’t her strength in her formative years in Trinidad and Tobago. “I was always failing the subject so when I was 14, my mom decided to bring me to a teacher,” Belgrave recalls. “I always give credit to this gentleman because he transformed my life and got me to love numbers. He was so inspiring. He thought I was brilliant at mathematics, so I became brilliant at mathematics. It was like a self-fulfilling prophecy. I think having good teachers and a teacher who believes in you is so precious and important. He was the driving force behind my choice of becoming a statistician in the pharmaceutical industry.”

Eventually, Belgrave received a scholarship from Microsoft Research and completed her doctoral study at the University of Manchester. “That was when I learned about AI and leverage machine learning to explore patient heterogeneity,” she continues. “This also marked the start of my journey into machine learning in healthcare. The combination of computer science and statistics was fascinating but I count my blessings being in the 2010s, when there was increasing computational power and an explosion of data, including the types and varieties of data, availability of genomic and metagenomic data that allowed me to venture into this area.

“There were lots of promises and treasures to understand people and patients in ways that we couldn’t have done before. It was a time of exciting development. Since then, my ambition has been looking at how we can create generalized machine learning frameworks for personalized healthcare through understanding the structure in disease heterogeneity.”

Fast forward to the present day and Belgrave is now heading Project TALIA, a collaboration between Microsoft Research UK and digital mental health platform, SilverCloud, that aims to create trustworthy AI systems to improve mental health. “We are using many different machine learning tools to disentangle heterogeneity and uncover the contributing factors towards making different subgroups of patients better,” Belgrave explains. “Specifically, we are examining the depression and anxiety scores of SilverCloud users and their patterns of engaging with the cognitive behavioral therapy (CBT) offered on the platform.”

Belgrave highlights the fact that the more one uses CBT, the better the outcomes. Nevertheless, what Belgrave and her team found, in a study published last August, is that it’s not just about how much people are using the SilverCloud platform but also the way they are interacting with the platform too. “We managed to identify five distinct subtypes of patients and engagement in this study which facilitates personalization of therapies,” she explains.

“We modeled our new interventions based on the characteristics and engagement patterns of these five groups of patients for 14 weeks. Overall, all of them improved. We also realized those with the best engagement aren’t necessarily those with the best clinical outcomes. It could be people are finding the interventions more effective and they are improving rapidly that they decided to stop further engagement.”

Looking ahead, Belgrave believes her next challenge lies in scaling interventions for heterogeneous groups of patients. “For example, regardless of interventions, there’d often be a small proportion of individuals who do not respond to anything,” she says. “I don’t have an answer now, but I hope we can create something they find useful.”

Unsurprisingly, Belgrave is also interested in increasing the uptake of AI in healthcare. “We can develop a great algorithm with the end-user in mind, helping them to overcome various obstacles,” she adds. “But if nobody is using it, it will remain, an algorithm at the end of the day. I want the algorithm to also be a solution. While we may not be able to change the world with one algorithm, we can make small changes and those little efforts will add up. So, I am truly excited about what AI can do for healthcare, particularly, mental health in the long run.”