Mind the gap: can artificial intelligence platforms bridge unmet needs in clinical decision making?

Abstract

Artificial intelligence (AI) refers to the creation of computerized algorithms to replicate cognitive function in order to solve real-world problems by building platforms which think and act like humans [1-4]. AI has clocked up many milestones since the concept was first described by John McCarthy in the 1950s [2]. No longer limited to the silver screen, AI is becoming increasingly woven into the digital tapestry of modern healthcare systems with every decade that passes [3]. This timely review from Stenzl et al. spotlights another area where AI is taking on a growing role, namely, in the setting of clinical decision making [1]. The authors show great insight in bringing our attention to this particular field where its momentum arguably is lesser known compared to its application in radiomics and treatment outcome predictions [5]. The authors begin by sharing data on the surging bibliographic trends for uro-oncology trials in recent years. This sets the background for raising the key question: How can time-pressured clinicians keep up to date with the burgeoning volume of new evidence being reported as well as distill the most useful information to implement at the point of care? AI is offered as a potential solution to this processing gap. The application of AI in this setting relies heavily on one of the core elements of AI: natural linguistic programming. These techniques comprise algorithms designed to digitally translate large datasets of language. However, this is no easy task and there are many challenges, including one of semantics. For example, how to instruct a computer to appreciate or correctly interpret nuances and the true intention of a manuscript is difficult. The authors of this review provide a summary of 12 AI-based platforms as well as lead us through a worked example using one called 'Dimensions'. They are to be commended for making the extra effort to include such an exercise as it helps the reader to better understand some of the strengths and weaknesses associated with these tools. Indeed, while the outlook for such technology holds great promise, it is made clear in the article that key ingredients are still missing. To this end, the authors have shared a proverbial shopping list of what is still needed to make the recipe work. This includes accounting for bias, generalizability of results for under-represented populations as well as troubleshooting for issues such as publisher licensing and lack of access to full texts. In doing so, they have a given a very fair overview of the current status of these AI-based platforms. Furthermore, all of these tools require the user to have a degree of knowledge of the topic under question to be able to accommodate for these limitations. However, their work does allow us to speculate on possible future applications such as in assisting guideline panels. In future, machine-learning models will provide a new benchmark for predicting oncological and surgical outcomes, help optimize pre- and postoperative data collection, and highlight opportunities for improving patient care [6]. While use of these models is likely to increase in future, the true potential and day-to-day clinical usage of these models are yet to come. None declared.

Ähnliche Arbeiten