Machine Learning and Evidence-Based Medicine

Annals of Internal Medicine
3 July 2018 Vol: 169, Issue 1

Ideas and Opinions |3 July 2018
Machine Learning and Evidence-Based Medicine
Ian A. Scott, MBBS, MHA, MEd
Machine learning (ML), which converts complex data into algorithms, challenges the traditional epidemiologic approach of evidence-based medicine (EBM). Here I outline the differences, strengths, and limitations of these 2 approaches and suggest areas of reconciliation.
Beginning in the 1970s, scientists extolled the virtues of EBM’s hypothesis-driven, protocolized experiments involving well-defined populations and preselected exposure and outcome variables. Inferences were made using traditional biostatistics. In the early 1990s, ML emerged, whereby advanced computing programs (machines) processed huge data sets (big data) from many sources and discerned patterns among multiple unselected variables. Such patterns were undiscoverable using traditional biostatistics (1) and were used to iteratively refine (learn) layered mathematical models (algorithms). The Table lists key differences between EBM and ML.