Performance of an algorithm for diagnosing acute cholecystitis using clinical and sonographic parameters

Authors:
Publisher website
Google Scholar
Abstract
Purpose Identify an algorithm using clinical and ultrasound (US) parameters with high diagnostic performance for acute cholecystitis. Methods Consecutive emergency department (ED) patients from 4/1/2019 to 12/31/2019 were retrospectively reviewed to record non-US parameters and make US observations. Outcomes were categorized as either: (1) acute cholecystitis; or (2) negative acute cholecystitis. Pivot tables identified parameter combinations either not found with acute cholecystitis or with predictive value for acute cholecystitis to establish the algorithm. US Division radiologists finalized an US report prior to ED disposition without use of the algorithm. Radiologist impression and algorithm prediction for acute cholecystitis were categorized as either (1) acute cholecystitis; (2) negative acute cholecystitis; or (3) inconclusive. Results Three hundred and sixty-six studies on 357 patients (mean age, 51 yrs ± 20 yrs; 215 women) met the inclusion criteria. 10.9% (40/366) of US studies had acute cholecystitis, 12.6% (46/366) had pathologically identified chronic cholecystitis without acute cholecystitis, and 76.5% (280/366) were negative acute cholecystitis. Algorithm compared to radiologist diagnostic performance was as follows: (1) sensitivity: 90.0% vs. 55.0%, p < 0.001; (2) augmented sensitivity (defined as when inconclusive categorization is considered consistent with acute cholecystitis): 100% vs. 85.0%, p < 0.001; (3) specificity: 93.6% vs. 94.8%, p = 0.50; (4) diagnostic rate (opposite of inconclusive rate): 96.4% vs. 93.2%, p = 0.04; (5) adverse outcome rate: 0.0% vs. 1.6%, p undefined. Conclusion For acute cholecystitis, an algorithm using non-binary ultrasound and clinical assessments had higher sensitivity, higher diagnostic rate, and fewer adverse outcomes, than subspecialty radiologist impressions. Graphical abstract

This publication has 19 references indexed in Scilit: