Introduction
Implementation of European Society of Gastrointestinal Endoscopy (ESGE) Resect-and-Discard and Diagnose-and-Leave strategies requires accurate optical characterization of colorectal polyps and appropriate translation of optical diagnosis into clinical management. Colo-ID is a novel gamified mobile application, developed with the institutional support of three French gastroenterology societies, designed to train colorectal polyp characterization and ESGE guideline-based management through repeated image-based cases. We evaluated whether training with Colo-ID improves optical diagnosis and management decisions in routine clinical practice.
Aims & Methods
COLO-PREDICT was a prospective, single-center study involving 12 experienced gastroenterologists organized into two endoscopist teams. Colo-ID (v1.2) training was sequentially introduced in the two teams, providing a temporary concurrent untrained control group. For each prospectively recorded colorectal polyp, endoscopists recorded an optical histological prediction and proposed management before histopathology was available. An optical diagnosis was required for every polyp irrespective of diagnostic confidence. Histopathology served as the reference standard. Performance before and after Colo-ID training was compared.
Results
Overall, 1,107 polyps from 567 consecutive patients were analyzed (543 pre-training; 564 post-training). Correct histological prediction increased from 61.0% (331/543; 95% CI 56.8–65.0) before training to 71.8% (405/564; 95% CI 68.0–75.4) after training, an absolute improvement of 10.9 percentage points (95% CI 5.4–16.6; P<0.001).
Correct management decisions increased from 51.7% (268/518; 95% CI 47.4–56.0) to 81.6% (439/538; 95% CI 78.1–84.6), an absolute improvement of 29.9 percentage points (95% CI 24.6–35.5; P<0.001).
The proportion of polyps with both correct histological prediction and correct management increased from 35.4% (191/540) to 63.9% (357/559), an absolute improvement of 28.5 percentage points (95% CI 22.6–34.3; P<0.001).
These improvements remained significant after adjustment for endoscopist team (adjusted OR 1.67 [95% CI 1.29–2.17], 3.77 [2.87–5.00], and 3.14 [2.44–4.05] for correct histological prediction, correct management, and both outcomes combined, respectively).
Both endoscopist teams had similar baseline optical characterization accuracy (61.0% vs 60.9%; P=0.973). After Colo-ID training, accuracy increased to 70.3% in Team 1 (P=0.019 vs pre-training) and 74.9% in Team 2 (P=0.001 vs pre-training). During staggered implementation, correct management decisions were significantly more frequent in the trained than in the concurrently untrained team (74.3% vs 59.3%; P=0.028).
Conclusion
Following training with the gamified Colo-ID mobile application, both real-world optical characterization of colorectal polyps and guideline-based management significantly improved among experienced endoscopists. Colo-ID may represent a practical tool to facilitate implementation of Resect-and-Discard and Diagnose-and-Leave strategies, with potential integration into structured endoscopy training and competency programs across different countries. Even in the era of AI-assisted optical diagnosis, maintaining human expertise will remain essential for informed and safe clinical implementation.
Disclosure
D.K.: Co-founder and CEO of Colo-ID; consultancy and/or speaker fees from COVIDIEN and NORGINE; support for attending meetings from ALFASIGMA, COOK, and FUJIFILM.
F.V.: Speaker fees from ABBVIE, TAKEDA, AMGEN, and JANSSEN-CILAG; support for attending meetings from ABBVIE; advisory board participation for JANSSEN-CILAG and BIOGEN.
M.C.: Consultancy fees from JOHNSON & JOHNSON, ABBVIE, and MSD; speaker fees from MSD, ABBVIE, AMGEN, PFIZER, MAUNEA KEA, and TILLOTS; support for attending meetings from FERRING, and MSD; advisory board participation for ABBVIE, PFIZER, and LILLY.
A.G.: Support for attending meetings from ABBVIE.
N.B.: Support for attending meetings from NORGINE.
N.M.: Grants or contracts from Groupe ADENE and NOMICS; Consultancy fees from SANOFI, NOMICS, BRUXLESS, ID SOLUTIONS, ARCHEAN TECHNOLOGIES, MSD, BMS; advisory Board participation for AB SCIENCE and SIMEOX.
All other authors declare no conflicts of interest.
