The AHRQ PSNet Collection comprises an extensive selection of resources relevant to the patient safety community. These resources come in a variety of formats, including literature, research, tools, and Web sites. Resources are identified using the National Library of Medicine’s Medline database, various news and content aggregators, and the expertise of the AHRQ PSNet editorial and technical teams.
Nagendran M, Chen Y, Lovejoy CA, et al. BMJ. 2020;368:m689.
This systematic review assessed randomized and non-randomized trials comparing the performance of artificial intelligence (AI; specifically deep learning algorithms) in medical imaging versus expert clinicians in order to characterize the state of the evidence and suggest future research directions which encourage innovation while protecting patients. The review identified 10 registered trials and 81 published non-randomized trials. Although 61 of 81 published studies reported that AI performance was comparable or better than that of clinicians, the authors identified few prospective studies or studies conducted in real-world settings; additionally, overall risk of bias was high and adherence to reporting standards was poor. Future studies examining the impact of AI in medicine must decrease risk of bias, increase relevance to real world clinical settings, and improve reporting and transparency.