Technology-induced bias in the theory of evidence-based medicine.

Abstract

Designing trials and studies to minimize confounding and bias is central to evidence-based medicine (EBM). The widespread use of recent technologies such as machine learning, smartphones, and the World Wide Web to collect, analyse, and disseminate information can improve the efficiency, reliability, and availability of medical research. However, it also has the potential to introduce new sources of significant, technology-induced evidential bias. This paper assesses the extent of the impact by reviewing some of the methods by and principles according to which evidence is collected, analysed, and disseminated in EBM, supported by specific examples. It considers the effect of personal health tracking via smartphones, the current proliferation of research data and the influence of search engine "filter bubbles", the possibility of machine learning-driven study design, and the implications of using machine learning to seek patterns in large quantities of data, for example from observational studies and medical record databases. It concludes that new technology may introduce profound new sources of bias that current EBM frameworks do not accommodate. It also proposes new approaches that could be incorporated in to EBM theory to mitigate the most obvious risks, and suggests where further assessment of the practical implications is needed.