Introduction

Guidelines indicate that optical diagnosis can be employed during endoscopy for a leave-in-situ or resect-and-discard strategy of diminutive polyps. Additionally, it aids in determining the appropriate techniques for polyp resection.

Aims & Methods

Our objective was to assess the accuracy and the clinical impact of real-life optical diagnosis during real-time colonoscopy conducted in the context of a CADe trial.
We retrospectively compared optical diagnosis with histology results of 1608 polyps that were prospectively included in the CAD-ARTIPOD trial (NCT04442607). During CAD-ARTIPOD endoscopists (ADR 20-50%) provided optical diagnosis, polyps were resected, and histological examination was performed by two independent GI histopathologists.
Mismatch was defined as a difference in polyp type (hyperplastic (HP) vs tubular adenoma (TA) vs tubulovillous adenoma (TVA) vs sessile serrated lesion (SSL) vs pseudopolyp vs carcinoma vs no polyp) or a difference in grade of dysplasia (high grade vs low grade vs no dysplasia). Clinical impact on surveillance interval per patient was consecutively evaluated according to ESGE standards.

Results

On a total of 1608 resected and histologically assessed specimens of 946 patients, 1026 (63.8%) showed a match and 582 (36.2%) a mismatch between optical diagnosis and histology. The optical diagnosis was 552 (34.3%) HP, 904 (56.2%) TA, 3 (0.2%) TVA, 119 (7.4%) SSL, 11 (0.7%) carcinoma, 6 (0.4%) inflammatory pseudopolyp and 13 (0.8%) unknown histology. Mismatch was present in 254 (43.6%) HP, 225 (24.9%) TA, 77 (64.7%) SSL, 7 (63.6%) carcinoma and 5 (83.3%) pseudopolyps.
Clinical consequences occurred for 48 (8.5%) polyps in as many patients (5.3%), with 12 polyps leading to a prolongation of the surveillance interval and 7 polyps overestimated as carcinomas not truly requiring further treatment. Twenty-nine (4.9%) of the optically misdiagnosed polyps led to a shortened surveillance interval (12 polyps with HGD and 17 as fifth TA) and two (0.3%) required further surgical therapy. Of the 115 misdiagnosed hyperplastic polyps, 74 (64.4%) were left-sided and may have been left in situ when applying SODA criteria with correct optical diagnosis applied.

Conclusion

Even in the setting of a clinical trial with high detectors, where the focus was detection, rather than optical diagnosis, discordance between optical diagnosis and histology occured in approximately one out of three polyps. The clinical impact was significant with alterations in clinical management for 8.5% of affected polyps, in 5.3% of patients. This underscores the importance of training in optical diagnoses and the potential role of artificial intelligence to assist in a resect and discard strategy.