At the University of Michigan, Jenna Wiens, PhD, an assistant professor of computer science and engineering, draws on her computational expertise to break down enormous data sets using machine learning. Ultimately, she hopes to improve disease modeling and to predict patient outcomes.
She will be speaking at the Big Data in Precision Health conference on May 24. I recently emailed with her to learn more about her perspective on the convergence of computation, big data and health.
In your lab, Machine Learning for Data-Driven Decisions, how do you approach the integration of big data and health care?
Much of our work lies at the intersection of machine learning and health, where we focus on developing novel methods for transforming health data (e.g., electronic health records, microbiome data and waveforms from wearables) into actionable knowledge that improves patient care. More specifically, we work on leveraging existing datasets to build predictions for adverse clinical outcomes.
What are you planning to discuss at the conference?
I’ll present a brief overview of a number of ongoing projects in my group. One project in particular, in which we’ve been able to make considerable progress, is in building models for identifying which patients are at greatest risk of acquiring hospital-born infections with Clostridium difficile.
Machine learning approaches to support clinical decisions typically focus on creating accurate models. Through our work with clinicians we’ve come to realize that this alone is not enough. In addition to being accurate, models deployed in a clinical setting must be credible, robust and actionable. In my talk, I'll discuss what the difference is between 'accurate' and 'credible,' and I'll speak about new work that aims to build risk stratification tools (which help separate patients into high-risk or low-risk categories) with these qualities.
How do you think big data can change the current health care landscape?
Big data and machine learning will improve the current health care landscape, not through replacement, but through the augmentation of clinicians and health care staff. Today’s clinicians are spending ever more time studying data about their patients, while still ignoring the vast majority of it — like if, for example, monitors continuously record data but clinicians only look at a small fraction, or they don't collect certain types of data (such as genetics or microbiome samples).
Algorithms that can effectively leverage these data can help physicians deliver the right treatment to the right patient at the right time.
Photo by Chris Liverani