Histopathologist Features Predictive of Diagnostic Concordance at Expert Level Amongst a Large International Sample of Pathologists Diagnosing Barrett’s Dysplasia Using Digital Pathology

Helen Coleman, Myrtle van der Wel, Marnix Jansen, Jacques Bergman, Sybren Meijer

Research output: Contribution to journalArticle

89 Downloads (Pure)

Abstract

Introduction
Histopathological diagnosis of dysplasia in Barrett’s oesophagus (BO) is the gold standard for patient risk stratification, but is subject to significant interobserver variation. We investigated histopathologist features that predict diagnostic performance amongst a large international cohort of gastrointestinal (GI) pathologists.

Methods
An online scoring environment was developed for GI-pathologists (n=55) from over 20 countries to grade a case set of 55 digitalised BO biopsies encompassing the complete spectrum from non-dysplastic Barrett’s oesophagus (NDBO), indefinite, low and high-grade dysplasia (IND/LGD/HGD). Case interpretations were recorded before and after revealing P53 immunohistochemistry. Detailed histopathologist demographic data (experience, centre volume, fellowship training etc.) was obtained through an online questionnaire. A consensus gold standard diagnosis was obtained for the entire case set through a reference panel of four expert pathologists. Multivariate regression analyses was conducted to identify pathologist predictors of concordance.

Results
We recorded over 6000 case diagnoses. Of 2,805 hour and E diagnoses, we found excellent concordance for NDBO (643 of 816 diagnoses; 79%) and HGD (544 of 765 diagnoses; 71%) and intermediate concordance for LGD (382 of 918; 42%) and IND (70 of 306; 23%), replicating known glass slide test characteristics. Major over or under-interpretations (i.e. NDBO overstaged as LGD/HGD, or LGD/HGD understaged as NDBO) were reported in 248 diagnoses (8.8%). Addition of p53 staining further improved diagnostic consensus, but had limited impact on major over or under-interpretations. Multivariate regression analyses revealed independent histopathologist predictors of expert level diagnostic performance, including; at least 5 years of experience, working within a teaching hospital, viewing 5–20 Barrett’s cases per week, adherence to major guidelines, and an interest in digital pathology.

Conclusions
Using this rich dataset representing a heterogeneous group of gastrointestinal pathologists working globally, we have quantified diagnostic performance for BO dysplasia diagnosis using digital case review. Our results reveal predictors of diagnostic performance at expert level, and will aid formulation of quality assurance criteria for guideline development.
Original languageEnglish
JournalGut
DOIs
Publication statusPublished - 08 Jun 2019

Fingerprint Dive into the research topics of 'Histopathologist Features Predictive of Diagnostic Concordance at Expert Level Amongst a Large International Sample of Pathologists Diagnosing Barrett’s Dysplasia Using Digital Pathology'. Together they form a unique fingerprint.

  • Cite this