If you’re in healthcare or data science I highly recommend reading “Predicting the Future – Big Data, Machine Learning, and Clinical Medicine” by Drs. Obermeyer and Emanuel in the latest issue of the New England Journal of Medicine. The question of whether machine learning can contribute to making more accurate predictions, whether diagnostic or prognostic has long been answered in hundreds of clinical studies across nearly all clinical sub-domains of medicine. The remaining barriers to these techniques having widespread impact are now, in my opinion: 1) education and 2) accessibility.
Drs. Obermeyer and Emanuel do a fantastic job with #1, distilling the nature of this set of tools and the characteristics that make them line up well with the nature of healthcare and healthcare data. They clearly draw distinctions between current risk assessment approaches that rely on human intuition and a few data points versus the ability of machine learning technologies to “learn” from millions of data points in combinations impossible for a human to consider. They point to the impact of these methods in other sciences such as astronomy – a far more relatable comparison for healthcare than the references to retail that I’ve come to rely upon too heavily.
They even offer a map of where impact is most likely to be experienced first and most impactfully. Assessments of risk (prognosis) are prime for improvement whereas reliable diagnosis is likely to take longer because the definitions of diagnosis are rarely binary. Most interesting is their choice to call out the obvious but controversial impact on the role of certain subspecialties: “machine learning will displace much of the work of radiologists and anatomical pathologists.” If you’re med student considering one or the other, you’ve been warned.
As someone focused on #2 (making these tools accessible and effective for healthcare), I’m always grateful for articles, talks, and presentations that I can share with others in healthcare to help them see through the fog of hype and confusion surrounding these techniques.
Great work guys
Leonard D’Avolio, PhD