Reliable predictors of clinical complete response (CR) are lacking for colon cancer (CC) patients treated with immunotherapy. Optical biomarkers have been used in other contexts for staging and response assessment, presenting a promising strategy for this purpose. This study aims to externally validate an optical evaluation framework for predicting CR.

We conducted a retrospective, multi-reader, multi-case study involving 28 international endoscopists and 75 cases of stage I-III CC, treated with a single dose of pembrolizumab. Endoscopists assessed response using a published optical framework and provided confidence ratings for each case. Evaluations were based on still images from re-staging colonoscopies, with accuracy compared to biopsy and surgical pathologic specimens.

The optical evaluation performed with comparable accuracy to routine biopsy (72.7% vs. 74% OR=0.94, p=0.80) with moderate inter-observer agreement (κ=0.522). Reader confidence significantly correlated with accuracy. High confidence (HC) evaluations demonstrated a marked increase in diagnostic accuracy and AUROC compared to low-confidence (LC) evaluations (0.799 vs 0.582, p=<0.0001). The addition of biopsy data doubled the odds of accurate LC-optical predictions (OR 2.01, p=<0.0001), but significantly decreased accuracy in HC cases (OR 0.53, p=<0.0001).

Confidence-weighted optical assessment accurately predicts CR in CC. HC-optical evaluations matched biopsy accuracy and were not improved by addition of biopsy data, supporting a selective biopsy approach. LC-optical predictions of CR improved with biopsy confirmation, highlighting its role in uncertain cases. This framework provides an accurate, reliable and efficient strategy for disease restaging.