Premium
An expert panel‐based study on recognition of gastro‐esophageal reflux in difficult esophageal pH‐impedance tracings
Author(s) -
Smits M. J.,
Loots C. M.,
Wijk M. P.,
Bredenoord A. J.,
Benninga M. A.,
Smout A. J. P. M.
Publication year - 2015
Publication title -
neurogastroenterology and motility
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.489
H-Index - 105
eISSN - 1365-2982
pISSN - 1350-1925
DOI - 10.1111/nmo.12536
Subject(s) - gold standard (test) , gastro , logistic regression , reflux , artificial intelligence , inter rater reliability , statistics , computer science , medicine , machine learning , pathology , mathematics , rating scale , disease
Background Despite existing criteria for scoring gastro‐esophageal reflux ( GER ) in esophageal multichannel pH‐impedance measurement (pH‐I) tracings, inter‐ and intra‐rater variability is large and agreement with automated analysis is poor. To identify parameters of difficult to analyze pH‐I patterns and combine these into a statistical model that can identify GER episodes with an international consensus as gold standard. Methods Twenty‐one experts from 10 countries were asked to mark GER presence for adult and pediatric pH‐I patterns in an online pre‐assessment. During a consensus meeting, experts voted on patterns not reaching majority consensus (>70% agreement). Agreement was calculated between raters, between consensus and individual raters, and between consensus and software generated automated analysis. With eight selected parameters, multiple logistic regression analysis was performed to describe an algorithm sensitive and specific for detection of GER . Key Results Majority consensus was reached for 35/79 episodes in the online pre‐assessment (interrater κ = 0.332). Mean agreement between pre‐assessment scores and final consensus was moderate ( κ = 0.466). Combining eight pH‐I parameters did not result in a statistically significant model able to identify presence of GER . Recognizing a pattern as retrograde is the best indicator of GER , with 100% sensitivity and 81% specificity with expert consensus as gold standard. Conclusions & Inferences A greement between experts scoring difficult impedance patterns for presence or absence of GER is poor. Combining several characteristics into a statistical model did not improve diagnostic accuracy. Only the parameter ‘retrograde propagation pattern’ is an indicator of GER in difficult pH‐I patterns.