Derivation and validation of lifestyle-based and microbiota-based models for colorectal adenoma risk evaluation and self-prediction.

objectiveEarly warning and screening of colorectal adenoma (CRA) is important for colorectal cancer (CRC) prevention. This study aimed to construct a non-invasive prediction model to improve CRA screening efficacy.

methodsThis study incorporated three cohorts, comprising 9747 participants who underwent colonoscopy. In cohort 1, 683 participants were prospectively recruited with comprehensive lifestyle information and faecal samples. CRA-associated bacteria were identified through 16S rRNA sequencing and quantitative real-time PCR. CRA prediction models were established using lifestyle and gut microbiota information. Cohort 2 prospectively enrolled 1529 participants to validate the lifestyle-based model, while cohort 3 retrospectively analysed 7535 individuals to determine the recommended initial colonoscopy screening ages for different risk groups based on age-specific CRA incidence rates.

resultsMultivariable logistic regression yielded a prediction model incorporating 14 variables, demonstrating robust discrimination (c-statistic=0.79, 95% CI 0.75, 0.82). Other machine learning approaches showed comparable performance (random forest: 0.78, 95% CI 0.73, 0.81; gradient boosting: 0.78, 95% CI 0.76, 0.83). The ages for starting colonoscopy screening were established at 42 years for the high-risk group vs 53 years for the low-risk group. The inclusion of

conclusionIntegrated mathematical modelling incorporating lifestyle parameters and gut microbial signatures provides an effective non-invasive strategy for CRA risk stratification, while the accompanying machine learning-assisted prediction application enables cost-effective, population-level screening implementation to optimise CRC prevention protocols.