Newly Published
Editorial  |   March 2020
Machine Learning Comes of Age: Local Impact versus National Generalizability
Author Notes
  • From the Department of Anesthesiology, University of Michigan Medical School, Ann Arbor, Michigan.
  • Accepted for publication January 13, 2020.
    Accepted for publication January 13, 2020.×
  • Correspondence: Address correspondence to Dr. Kheterpal: sachinkh@med.umich.edu
Article Information
Editorial / Technology / Equipment / Monitoring / Quality Improvement
Editorial   |   March 2020
Machine Learning Comes of Age: Local Impact versus National Generalizability
Anesthesiology Newly Published on March 23, 2020. doi:https://doi.org/10.1097/ALN.0000000000003223
Anesthesiology Newly Published on March 23, 2020. doi:https://doi.org/10.1097/ALN.0000000000003223
Machine learning, a subfield of artificial intelligence, is an increasingly popular topic within medicine. Evangelists of machine learning hope that it will revolutionize health care. While machine learning may still be in the “hype” phase of excitement, we are beginning to see applications within perioperative medicine with potential perioperative clinical impact.1  Addressing meaningful problems that may decrease patient harm, improve quality of life, or reduce administrative burden is an important goal when implementing machine learning in health care.
In this issue of Anesthesiology, Mišić et al.2  evaluate various machine learning techniques for predicting 30-day postoperative readmissions. Hospital readmissions are costly and common events that are the target of healthcare improvement and policy change initiatives, but there are broader implications in the article by Mišić et al. All anesthesiologists should note that this work calls into question the purported value of: (1) advanced model diagnostics that are difficult to interpret; (2) using thousands of data elements to predict outcomes versus parsimonious approaches; (3) focusing on multicenter “generalizability” of prediction models rather than just optimizing for future predictions at a given hospital; and (4) advanced machine learning algorithms versus classic techniques.