Estimating Probability for Esophageal Obstruction: A Diagnostic Decision Support Tool Applying Machine Learning to Functional Lumen Imaging Probe Panometry

Abstract

Background/AimsThis study aimed to develop a diagnostic tool using machine learning to apply functional luminal imaging probe (FLIP) panometry data to determine the probability of esophagogastric junction (EGJ) obstruction as determined using the Chicago Classification version 4.0 (CCv4.0) and high-resolution manometry (HRM). MethodsFive hundred and fifty-seven adult patients that completed FLIP and HRM (with a conclusive CCv4.0 assessment of EGJ outflow) and 35 asymptomatic volunteers ("controls") were included. EGJ opening was evaluated with 16-cm FLIP performed during sedated endoscopy via EGJ-distensibility index and maximum EGJ diameter. HRM was classified according to the CCv4.0 as conclusive disorders of EGJ outflow or normal EGJ outflow (timed barium esophagram applied when required and available). The probability tool utilized Bayesian additive regression treesBART, which were evaluated using a leave-one-out approach and a holdout test set. ResultsPer HRM and CCv4.0, 243 patients had a conclusive disorder of EGJ outflow while 314 patients (and all 35 controls) had normal EGJ outflow. The model accuracy to predict EGJ obstruction (based on leave-one-out/holdout test set, respectively) was 89%/90%, with 87%/85% sensitivity, 92%/97% specificity, and an area under the receiver operating characteristic curve of 0.95/0.97. A free, opensource tool to calculate probability for EGJ obstruction using FLIP metrics is available at https://www.wklytics.com/nmgi/prob_flip. html. ConclusionsApplication of FLIP metrics utilizing a probabilistic approach incorporates the diagnostic confidence (or uncertainty) into the clinical interpretation of EGJ obstruction. This tool can provide clinical decision support during application of FLIP Panometry for evaluation of esophageal motility disorders.

Discriminating “impaired” from “disordered” contractile response on FLIP panometry by utilizing pressure measurement

“…However, for low‐pressure IDCR, pursuing HRM to further evaluate hypomotility could be considered on a case by case basis as to whether or not further characterization of hypomotility on HRM would represent an actionable finding. For example, with low‐pressure IDCR and normal EGJ opening on FLIP (i.e., EGJ‐distensibility index ≥2.0 mm 2 /mmHg and maximum EGJ diameter ≥16 mm; Data S1), achalasia is essentially ruled out, 4,23 hence HRM will likely yield IEM or absent contractility. As clinical management between these two HRM diagnoses may be similar, HRM could potentially be deferred (whereas HRM could be pursued in a case where this differentiation would alter management, particularly if achalasia is ruled out based on normal EGJ opening on FLIP) 24 .…”

The Clinical Usefulness of Functional Luminal Imaging Probe in Esophageal Dysmotility Disorder

“…26 The recently published manuscript in this issue of Journal of Neurogastroenterology and Motility used the machine learning technique in the identification of EGJ obstruction. 27 With the incorporation of this machine learning technique in the near future, we can effectively diagnose and expand our understanding of esophageal dysmotility disorders. The illustration of measurement of manometry and functional luminal image probe (FLIP) in the gastrointestinal (GI) tract of humans.…”