Artificial intelligence for improved patient outcomes: The pragmatic randomized controlled trial is the secret sauce
Presenting author: Daniel W. Byrne, Department of Biostatistics, Vanderbilt University Medical Center
- Henry J. Domenico, Department of Biostatistics, Vanderbilt University Medical Center
- Shannon C. Walker, Department of Pathology, Microbiology and Immunology, Vanderbilt University Medical Center
- Ryan P. Moore, Department of Biostatistics, Vanderbilt University Medical Center
Artificial Intelligence (AI) has exploded in the media for both its astonishing power and concerning limitations. Yet, to date, we have almost no rigorous evidence that AI has been used to improve patient health outcomes. What needs to change? In one word: "Randomization." The excuses for why randomization is not an option are plentiful and just roll off the tongue, yet in our experience the excuses almost all turn out to be unjustified. In 2016, the "godfather of AI," Geoffrey Hinton, said "People should stop training radiologists now. It is just completely obvious that within 5 years deep learning is going to do better than radiologists." Perhaps this should be updated to "People should start training physicians to create, use and evaluate AI tools in a modern and rigorous manner." This training needs to include evaluation skills, such as: regression to the mean, reverse causation, the issues with overall accuracy, residual confounding, pragmatic patient-level randomized controlled trials, and adaptive platform trials. AI has enormous potential to reduce the workload and burnout among physicians. Once we embrace pragmatic randomized controlled trials, Artificial Intelligence will improve patient outcomes. Randomization is the secret sauce—but only if scientists lead these projects.