Jonathan H. Chen, MD, PhD
Stanford Department of Medicine

Tuesday, September 6, 2016, 1:30 pm to 2:30 pm
MSOB Conference Room X-275

ABSTRACT:
Medical decision making is fraught with both uncertainty and undesirable variability. The vast majority of our clinical decisions lack adequate evidence to determine their efficacy and inconsistent implementation compromises quality and efficiency.

The current standards in clinical decision support reinforce best-practices but are limited in scalability by manual production. “Grand challenges” thus include mining clinical data sources to automatically generate decision support content.

Statistical approaches allow us to learn patterns that reflect real-world standards of care vs. outliers. This can range from my evaluation of the national distribution of opioid prescriptions to my current NIH Big Data 2 Knowledge K01 Career Development Award directed to empower individual clinicians with the collective experience of the many.

I will review my efforts developing a collaborative filtering machine-learning approach to clinical order entry, analogous to Netflix or Amazon.com’s “Customers who bought A also bought B” algorithm. This automatically generated decision support content can reproduce, and even optimize, manual constructs like order sets while remaining largely concordant with guidelines and avoiding inappropriate recommendations. This has even more important implications for prevalent cases where well-defined guidelines do not exist. The same methodology is predictive of clinical outcomes comparable to state-of-the-art risk prediction models. Embedded randomization of such decision support interventions could then allow us to explicitly build knowledge for the future, even as we enhance care today, in a closed-loop learning health system.