S1290 Development and Validation of a Simplified Tool to Predict Endoscopist-Rated Utility of Tattoo Placement for Colorectal Polyp Localization

Introduction: Tattoo placement is recommended for large colorectal polyps (CRPs) referred for endoscopic resection (ER). However, it is unclear whether endoscopists performing ER find tattoo placement helpful. We aimed to develop a clinical model to predict endoscopist-rated tattoo usefulness based on prospectively collected data. Methods: This was a prospective, multicenter study. Adults referred for ER of a previously tattooed CRP were enrolled at 7 sites from February 2024 to December 2024. After ER, endoscopists completed a 7-item, 5-point Likert survey evaluating tattoo usefulness. The prediction model was developed using Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression with 5-fold cross-validation to select from >50 variables. A simplified logistic regression model was constructed using the most predictive clinical variables. Model discrimination was assessed using cross-validated and bootstrapped area under the curve (AUCs). Calibration was evaluated using the GiViTI calibration belt. Results: A total of 103 CRPs from 101 patients were tattooed and referred for ER. Tattoo placement was rated as helpful in 43% (44/103). LASSO logistic regression identified 7 key predictors and demonstrated strong discrimination (cross-validated AUC 0.88; 95% CI, 0.77–0.94) but poor calibration (GiViTI P = 0.003). A simplified multivariable logistic regression model was constructed using LASSO-selected variables. The final model included the following variables: colonoscopy indication, polyp size, location, ileocecal valve involvement, polyp fibrosis/scarring, morphology, and suspected histology. The model demonstrated excellent performance (AUC 0.83; bootstrap 95% CI, 0.77–0.93) with good calibration (GiViTI P = 0.124). Using an optimal probability threshold (≥0.467), the model achieved 80% sensitivity, 78% specificity, 71.4% PPV, 83% NPV, and 77.5% overall accuracy. To facilitate clinical implementation, a web-based calculator (https://bit.ly/PolypTattoo) that predicts the probability of tattoo usefulness was created using variables from the final prediction model. Conclusion: Using prospectively collected data, we developed a 7-variable model using routinely available colonoscopy data that can accurately predict when tattoo placement will be helpful for lesion localization for future ER. The model demonstrated excellent performance, though external validation is needed.