Our expert answers 3 Questions
I am thinking about how to combine health data across multiple sources and levels of analysis, including biomarker, psychological and social data. These data integration issues involve merging data across multiple time scales and multiple systems.
I have always been interested in issues that now constitute the field of data science – how one moves from research questions, to working with data, to understanding. Now that we have many new ways of collecting data, we have opportunities to rethink how to work with data.
One implication is that we can refine the concept of "health effect" into a multivariate concept, meaning that we can evaluate more than one effect or outcome of a treatment at a time. Take cancer treatment, for example. One effect of, say, chemotherapy, involves shrinking the size of the tumor. But there are other "effects" of chemotherapy, including side effects such as nausea, fatigue, loss of appetite, effects on family members, etc. We will be able to provide better healthcare (and be able to quantify metrics to assess success) if we take a multivariate approach to quantifying a treatment effect and can specify how to trade off competing effects of treatment. This creates an opportunity to extend the current interest in modeling variations on one treatment outcome to modeling variations of multiple treatment effects. Such methodological advances will permit new treatment optimizations for risk and benefits for individual patients.