Incorporating Helicobacter pylori Infection: A Novel Model for Predicting Advanced Colorectal Neoplasia.

BACKGROUND AND

aimsEarly detection of advanced colorectal neoplasia (ACN) is critical for reducing colorectal cancer (CRC) incidence and mortality. Existing prediction models exhibit limited discriminative power. This study aimed to develop and internally validate a novel risk prediction model for ACN, with particular emphasis on incorporating gastric-related indicators to enable integrated gastrointestinal tumor screening.

methodsConsecutive patients who underwent concurrent gastroscopy and colonoscopy for screening, symptomatic assessment, or surveillance purposes were recruited from the endoscopy center between January 1, 2021, and December 31, 2023. Comprehensive data including demographic characteristics, family history, Helicobacter pylori (H. pylori) infection status, and gastric histopathological results were collected. Patients with incomplete data were excluded a priori with no imputation. Candidate variables were selected using Elastic Net and LASSO regression to develop the prediction model. The model was trained on 80% of the data (n = 2045) and internally validated on the remaining 20% (n = 495).

resultsAmong 2540 participants [mean age 59.0 years; 1269 males (50.0%)], the overall prevalence of ACN was 13.1% (333/2540). The final Elastic Net model incorporated eight variables, with H. pylori infection as the strongest predictor. In the internal validation cohort, the model achieved an area under the curve (AUC) of 0.837 (95% CI: 0.781-0.892), with 75.8% sensitivity, 82.2% specificity, and 81.4% accuracy. Calibration metrics showed good agreement between predicted and observed ACN risks (Brier score = 0.103; calibration plots showed non-significant deviation from perfect calibration). It significantly outperformed the Asia-Pacific Colorectal Screening (APCS) score (AUC = 0.622), its revised edition (AUC = 0.589), and the Colorectal Tumor Prediction Score (AUC = 0.570) (all p < 0.001, DeLong's test). Using the model's optimal cutoff, 25.1% (124/495) of the internal validation cohort were stratified as high-risk, with an ACN detection rate of 37.9%. The number of participants needed to screen (PNS) to detect one ACN case was only 3 in the high-risk group, compared with 25 in the low-risk group, thus demonstrating markedly improved screening efficiency.

conclusionsThis study develops and internally validates a risk prediction model for ACN that integrates gastric-related indicators, with H. pylori infection as the key predictor. The model exhibits promising retrospective risk stratification performance in patients undergoing concurrent gastroscopy and colonoscopy for screening, symptomatic assessment, or surveillance. It enables retrospective risk assessment during routine endoscopy, but its clinical utility for guiding prospective colonoscopy referral decisions requires further validation in prospective cohorts.