AI Imaging Analysis Needs Evaluation Before Implementation

Abstract

To date, surgeons have yet to operationalize a decade of research showing that frailty correlates with worse surgical outcomes. Despite the development of many scoring systems, there is no universally accepted and easy-to-calculate pointof-care assessment tool to accurately identify patients with frailty who are at risk for poor outcomes, nor are there proven programs to improve the fitness of the patients to prevent complications. Sarcopenia, progressive and generalized loss of muscle mass and function, is a risk factor for adverse outcomes, functional decline, and frailty in aging populations, but its assessment and management have faced similar challenges for integration into routine clinical practice.In this issue, Fumagalli et al 1 propose that artificial intelligence (AI) software can analyze muscle quantity and quality on preoperative computed tomography scans to provide automated and objective assessments of frailty to streamline clinical risk stratification. They retrospectively evaluated scans from 48 444 patients who underwent abdominal operations, using principal component analysis to develop weighted body composition scores. They found that muscle quantity and quality scores correlated with a validated frailty assessment 2 and were associated with 30-day morbidity, mortality, and readmission after adjusting for relevant factors (reduced risk of morbidity for highest vs lowest quartile: relative risk, 0.59 [95% CI, 0.52-0.67]; and reduced risk of readmission or mor-