PREDICTIVE RULES FOR OPTICAL DIAGNOSIS OF ≤10 MM COLORECTAL POLYP: BASIC VALIDATION

Aims:

To in vivo predict colorectal polyp histology, the BASIC classification incorporates both morphological and pit/vessel descriptors domains. To define the new BASIC classes for adenomatous and hyperplastic polyps, we analytically assessed the strength of predictions of these descriptors individually and in combination.

Methods:

A video-library consisting of 102≤10 mm polyps with white-light (WLI) and blue-light imaging (BLI) still images/videos was reviewed by 7 expert-endoscopists by an user-friendly software application. Each polyp was rated according to the individual descriptors of the three (surface/pit/vessel) domains of the BASIC without (I phase) and with (II phase) a final diagnosis. Adjusted probability for adenomatous/hyperplastic histology and overall accuracy were calculated. By a modified Delphi agreement, the new BASIC classes for adenomatous and hyperplastic histologies were defined, and corresponding accuracies assessed.

Results:

By pooling 2,175 observations, strength of prediction for adenomatous histology of the two surface descriptors was poor-to-fair (AUC/OR 95% CI for mucus: 0.5/0.2, 0.14 – 0.26; AUC/OR 95% CI for irregular surface: 0.68/5.88, 4.55 – 7.69), while it was stronger for most of pit (AUC/OR 95% CI: featureless: 0.81/0.13, 0.10 – 0.16; AUC/OR 95% CI: not round pits: 0.87/10.1, 7.6 – 13.4) and vessel (AUC/OR 95% CI for pericryptal vessels 0.87/0.13, 0.01 – 0.19) descriptors. By combining the surface with pit/vessel domains in the multilevel model, a good-to-excellent prediction power was shown (AUC: 0.89; 95% CI: 0.81 – 0.96). After definition of BASIC classes for adenomatous (A) and hyperplastic (H) polyps, endoscopist-based accuracy for high-confidence BLI prediction was 90.3% (86.3%-93.2%), and it was superior to both high confidence WL predictions (OR: 1.88, 95% CI: 1.42 – 2.49) and low-confidence BLI (OR 2.33, 95% CI: 1.13 – 4.98) diagnosis.

Conclusions:

Based on analytical quantification in strength of prediction among morphological and pit/vessel descriptors, new BASIC classes for adenomatous and hyperplastic histologies were created with favourable effects on accuracy and confidence level.