An interobserver reproducibility study on invasiveness of bladder cancer using virtual microscopy and heatmaps

The distinction between non-invasive (pTa) and invasive (pT1) non-muscle invasive bladder cancer (NMIBC) is subject to considerable interobserver variation. We aimed to generate a teaching set of images based on the diagnostic opinions of a panel of expert genitourinary pathologists.

Twenty-five transurethral resection specimens initially reported as pT1 NMIBC from two university hospitals were selected on the basis of potential uncertainty of stromal invasion. Digitized slides were reviewed independently by a panel of eight genitourinary pathologists, who annotated any invasive area if present. Annotations were reviewed by the lead panel, and heatmaps of annotated areas were constructed. Reasons for discrepancies were analysed, and kappa scores were calculated to determine agreement among the eight panellists. Full agreement by the eight panellists was obtained in 11 of 25 cases (44%), with a multi-rater (Fleiss) kappa of 0.47 (P < 0.0001). After joint review of the seven discordant (agreement <75% of panellists) cases, consensus was obtained for six cases, and a teaching set of images was generated.

Interobserver agreement among the panellists in the selected cases was moderate, but consensus could be reached in almost all cases. Heatmaps proved to be instrumental in generating a teaching set of images for standardization of histological criteria for NMIBC invasion.