Do as AI say: susceptibility in deployment of clinical decision-aids

Abstract

Artificial intelligence (AI) models for decision support have been developed for clinical settings such as radiology, but little work evaluates the potential impact of such systems. In this study, physicians received chest X-rays and diagnostic advice, some of which was inaccurate, and were asked to evaluate advice quality and make diagnoses. All advice was generated by human experts, but some was labeled as coming from an AI system. As a group, radiologists rated advice as lower quality when it appeared to come from an AI system; physicians with less task-expertise did not. Diagnostic accuracy was significantly worse when participants received inaccurate advice, regardless of the purported source. This work raises important considerations for how advice, AI and non-AI, should be deployed in clinical environments.

Evaluating Adoption, Impact, and Factors Driving Adoption for TREWS, a Machine Learning-Based Sepsis Alerting System

“…To our knowledge, this is the first prospective study of a deployed machine learning-based bedside clinical tool that quantitatively studies and achieves high provider adoption. Previous studies of deployed systems have used post-hoc surveys or interviews to gauge provider impressions 25,36,43 ; however, provider impressions can differ from their actual use 4,18,24 . A common alternative is to study real-time provider response using a clinical simulation 23,24 .…”

“…Adoption of automated systems in non-clinical settings depends on several factors including personal characteristics and preferences of the user, characteristics of the automated system (e.g., CDS tool), and the environment in which the technology is used 22 . In clinical simulations in a 'laboratory' setting, where providers are shown simulated CDS recommendations for exemplar patients, studies have found that interface design 23 , provider expertise 24 , and clinical time constraints 25 all play a role in adoption of the tool. However, in the real-world clinical setting, there are additional barriers to system adoption, including unpredictable variations in workflow, changes in personnel, and high-stakes consequences of incorrect decisions, that are difficult to replicate in simulations 26 .…”

Precision strength training: Data-driven artificial intelligence approach to strength and conditioning

“…We will envision strength training and sports management to follow the advances in medicine and society as a whole, with the trend for increased human-machine interplay in human-inthe-loop systems, that are developed augmenting human experts rather than by replacing them with AI [49,258,279,310,615,742,753]. It is not trivial to transfer and quantify the tacit knowledge from athlete-coach to be used for quantitative modelling [81,482,588], and the n = 1 expert knowledge [364,379,716,750];…”

Can deep learning classify stroke subtypes from chest X-rays?

“…Would the advice of ASTRO-X be considered in the diagnosis? These are all questions that the AI Contents lists available at ScienceDirect EBioMedicine journal homepage: www.elsevier.com/locate/ebiom community is dealing with for many newly developed deep learning models [8].…”

“…Would the advice of ASTRO-X be considered in the diagnosis? These are all questions that the AI community is dealing with for many newly developed deep learning models [8] .…”