Endoscopy International Open (Feb 2021)

Interobserver agreement of the Paris and simplified classifications of superficial colonic lesions: a Western study

  • Francesco Cocomazzi,
  • Marco Gentile,
  • Francesco Perri,
  • Antonio Merla,
  • Fabrizio Bossa,
  • Mariano Piazzolla,
  • Antonio Ippolito,
  • Fulvia Terracciano,
  • Arcangela Patrizia Giuliani,
  • Rossella Cubisino,
  • Antonella Marra,
  • Sonia Carparelli,
  • Alessia Mileti,
  • Rosa Paolillo,
  • Andrea Fontana,
  • Massimiliano Copetti,
  • Alfredo Di Leo,
  • Angelo Andriulli

DOI
https://doi.org/10.1055/a-1352-3437
Journal volume & issue
Vol. 09, no. 03
pp. E388 – E394

Abstract

Read online

Background and study aims The Paris classification of superficial colonic lesions has been widely adopted, but a simplified description that subgroups the shape into pedunculated, sessile/flat and depressed lesions has been proposed recently. The aim of this study was to evaluate the accuracy and inter-rater agreement among 13 Western endoscopists for the two classification systems. Methods Seventy video clips of superficial colonic lesions were classified according to the two classifications, and their size estimated. The interobserver agreement for each classification was assessed using both Cohen k and AC1 statistics. Accuracy was taken as the concordance between the standard morphology definition and that made by participants. Sensitivity analyses investigated agreement between trainees (T) and staff members (SM), simple or mixed lesions, distinct lesion phenotypes, and for laterally spreading tumors (LSTs). Results Overall, the interobserver agreement for the Paris classification was substantial (κ = 0.61; AC1 = 0.66), with 79.3 % accuracy. Between SM and T, the values were superimposable. For size estimation, the agreement was 0.48 by the κ-value, and 0.50 by AC1. For single or mixed lesions, κ-values were 0.60 and 0.43, respectively; corresponding AC1 values were 0.68 and 0.57. Evaluating the several different polyp subtypes separately, agreement differed significantly when analyzed by the k-statistics (0.08–0.12) or the AC1 statistics (0.59–0.71). Analyses of LSTs provided a κ-value of 0.50 and an AC1 score of 0.62, with 77.6 % accuracy. The simplified classification outperformed the Paris classification: κ = 0.68, AC1 = 0.82, accuracy = 91.6 %. Conclusions Agreement is often measured with Cohen’s κ, but we documented higher levels of agreement when analyzed with the AC1 statistic. The level of agreement was substantial for the Paris classification, and almost perfect for the simplified system.