Abstract
Life is filled with puzzles and mysteries, and we often fail to recognize the difference. As described by Gregory Treverton and Malcolm Gladwell, puzzles are solved by gathering and assimilating all relevant data in a logical, linear fashion, as in deciding which antibiotic to prescribe for an infection. In contrast, mysteries remain unsolved until all relevant data are analyzed and interpreted in a way that appreciates their depth and complexity, as in determining how to best modulate the host immune response to infection. When investigating mysteries, we often fail to appreciate their depth and complexity. Instead, we gather and assimilate more data, treating the mystery like a puzzle. This strategy is often unsuccessful. Traditional approaches to predictive analytics and phenotyping in surgery use this strategy.
Highlights
Specialty section: This article was submitted to Medicine and Public Health, a section of the journal Frontiers in Artificial Intelligence
Most patients recover along a clinical trajectory that can be predicted by their physiologic reserve, the severity of the underlying disease process, and the physiologic insult associated with the planned operation. These predictions augment the decision to offer an operation and inform discussions with patients and their caregivers regarding treatment options and prognosis. This process often relies on biased, error-prone individual judgement, especially when decisions are made under time constraints and uncertainty, leading to preventable harm
Parametric regression models make predictions with logical, linear rules expressed as algorithms; machine and deep learning artificial
Summary
Most patients recover along a clinical trajectory that can be predicted by their physiologic reserve, the severity of the underlying disease process, and the physiologic insult associated with the planned operation These predictions augment the decision to offer an operation and inform discussions with patients and their caregivers regarding treatment options and prognosis. In applying six different regression-based prediction models to 1,380 patients undergoing colorectal surgery, Bagnall et al (2018) found that all six models performed poorly with area under the receiver operating characteristic curve (AUROC) 0.46–0.61 (Bagnall et al, 2018) In these cases, poor model accuracy is often attributed to stochastic, or random, risk
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
