Machine Learning and Evidence-Based Medicine

Ideas and Opinions3 July 2018Machine Learning and Evidence-Based MedicineIan A. Scott, MBBS, MHA, MEdIan A. Scott, MBBS, MHA, MEdUniversity of Queensland, Brisbane, Queensland, Australia (I.A.S.)Search for more papers by this authorAuthor, Article, and Disclosure Informationhttps://doi.org/10.7326/M18-0115 SectionsAboutFull TextPDF ToolsAdd to favoritesDownload CitationsTrack CitationsPermissions ShareFacebookTwitterLinkedInRedditEmail 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.A Historical ContextBeginning 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 ...References1. Alanazi HO, Abdullah AH, Qureshi KN. A critical review for developing accurate and dynamic predictive models using machine learning methods in medicine and health care. J Med Syst. 2017;41:69. [PMID: 28285459] CrossrefMedlineGoogle Scholar2. Hashimoto DA, Rosman G, Rus D, Meireles OR. Artificial intelligence in surgery: promises and perils. Ann Surg. 2018. [PMID: 29389679] CrossrefMedlineGoogle Scholar3. Bruland P, McGilchrist M, Zapletal E, Acosta D, Proeve J, Askin S, et al. Common data elements for secondary use of electronic health record data for clinical trial execution and serious adverse event reporting. BMC Med Res Methodol. 2016;16:159. [PMID: 27875988] CrossrefMedlineGoogle Scholar4. Frizzell JD, Liang L, Schulte PJ, Yancy CW, Heidenreich PA, Hernandez AF, et al. Prediction of 30-day all-cause readmissions in patients hospitalized for heart failure: comparison of machine learning and other statistical approaches. JAMA Cardiol. 2017;2:204-9. [PMID: 27784047] doi:10.1001/jamacardio.2016.3956 CrossrefMedlineGoogle Scholar5. Deo RC. Machine learning in medicine. Circulation. 2015;132:1920-30. [PMID: 26572668] doi:10.1161/CIRCULATIONAHA.115.001593 CrossrefMedlineGoogle Scholar6. Hwan Bae M, Hoon Lee J, Heon Yang D, Sik Park H, Cho Y, Chull Chae S, et al. Erroneous computer electrocardiogram interpretation of atrial fibrillation and its clinical consequences. Clin Cardiol. 2012;35:348-53. [PMID: 22644921] doi:10.1002/clc.22000 CrossrefMedlineGoogle Scholar7. Derman E. Models.Behaving.Badly. Why Confusing Illusion With Reality Can Lead to Disaster, on Wall Street and in Life. New York: Free Pr; 2012. Google Scholar8. Cabitza F, Rasoini R, Gensini GF. Unintended consequences of machine learning in medicine. JAMA. 2017;318:517-8. [PMID: 28727867] doi:10.1001/jama.2017.7797 CrossrefMedlineGoogle Scholar9. Linden A, Yarnold PR. Combining machine learning and matching techniques to improve causal inference in program evaluation. J Eval Clin Pract. 2016;22:864-70. [PMID: 27353301] doi:10.1111/jep.12592 CrossrefMedlineGoogle Scholar10. Karim ME, Pang M, Platt RW. Can we train machine learning methods to outperform the high-dimensional propensity score algorithm? Epidemiology. 2018;29:191-8. [PMID: 29166301] doi:10.1097/EDE.0000000000000787 CrossrefMedlineGoogle Scholar Author, Article, and Disclosure InformationAffiliations: University of Queensland, Brisbane, Queensland, Australia (I.A.S.)Acknowledgment: The author thanks Prof. Paul Glasziou, Director, Centre for Research in Evidence-Based Practice, Bond University, Gold Coast, Australia, and Prof. Adam Elshaug, Co-Director, Menzies Centre for Health Policy, University of Sydney, Sydney, Australia, for helpful comments on previous drafts of the manuscript.Disclosures: The author has disclosed no conflicts of interest. The form can be viewed at www.acponline.org/authors/icmje/ConflictOfInterestForms.do?msNum=M18-0115.Corresponding Author: Ian A. Scott, MBBS, MHA, MEd, University of Queensland, Translational Research Institute, 37 Kent Street, Brisbane, Queensland 4102, Australia; e-mail, ian.[email protected]qld.gov.au.