Artificial intelligence–based skeletal muscle estimates and outcomes of EUS-guided treatment of pancreatic fluid collections

Background and Aims: Skeletal muscle status may affect clinical outcomes of a variety of pancreatic diseases. Skeletal muscle quality and quantity have not been fully examined in relation to the outcomes of EUS-guided treatment of pancreatic fluid collections (PFCs). Methods: Using a multi-institutional cohort of 372 patients receiving EUS-guided treatment of PFCs in 2010 to 2020, we examined the association of skeletal muscle status with adverse outcomes, including clinical treatment failure and in-hospital mortality. We used an in-house deep learning–based platform for preprocedural CT images, and skeletal muscle density (SMD) and skeletal muscle index (SMI; height-adjusted muscle area) were calculated as surrogates for muscular quality and quantity, respectively. Multivariable logistic regression analysis was conducted to calculate odds ratios (ORs) for adverse outcomes. Results: Lower-level SMD was associated with higher risks of clinical failure and in-hospital mortality (P_trend < .001). The adjusted OR for clinical failure comparing the extreme quartiles was 3.64 (95% confidence interval, 1.52-8.72). Compared with patients in the top 2 quartiles, patients in the lowest quartile had an adjusted OR for in-hospital mortality of 12.4 (95% confidence interval, 3.43-44.8). No effect modification according to the PFC types on the SMD–outcome relationship (P_interaction > .16) was observed. SMD was not associated with the risk of procedure-related adverse events or PFC recurrence. SMI was not associated with adverse outcomes (P_trend > .39). Conclusions: In patients with endoscopically managed PFCs, SMD (but not SMI) was associated with the risks of clinical failure and in-hospital mortality, supporting the prognostic role of skeletal muscle quality.