The reproducibility crisis in the age of digital medicine

By Aaron Stupple, David Singerman, Leo Anthony Celi. 2019

138 citations recorded by the source corpus.

INTRODUCTION If anyone doubts the explosive growth of interest in digital medicine, consider a recent conference and workshop in Beijing, jointly organized by the People’s Liberation Army General Hospital and MIT Critical Data to showcase the opportunities and challenges of applying machine learning to the kind of data routinely collected during the provision of care. In person, 500 attendees heard a keynote and panels and participated in a health data hackathon. Online, however, the event was streamed to more than one million unique viewers. As databases of medical information are growing, the cost of analyzing data is falling, and computer scientists, engineers, and investment are flooding into the field, digital medicine is subject to increasingly hyperbolic claims. Every week brings news of advances: superior algorithms that can predict clinical events and disease trajectory, classify images better than humans, translate clinical texts, and generate sensational discoveries around new risk factors and treatment effects. Yet the excitement about digital medicine—along with the technologies like the ones that enable a million people to watch a major event—poses risks for its robustness. How many of those new findings, in other words, are likely to be reproducible? Digital medicine must take steps to avoid a reproducibility “crisis” of the kind that has engulfed other areas of biomedicine and human science in the last decade and shaken public confidence in the validity of scientific work. The goal of this paper is to use a historical perspective on reproducibility and its current crisis to suggest how digital medicine can avoid a reproducibility crisis of its own.

Subfields: Health Informatics, Information Systems and Management, Public Health, Environmental and Occupational Health

Citations in the corpus