Introduction
Advanced colorectal neoplasia (ACN) has a high potential for malignancy, and its early detection and removal can effectively reduce the incidence and mortality of colorectal cancer (CRC). A predictive model to estimate the risk of ACN could help direct patients and providers to improve screening efficiency.
Aims & Methods
We aimed to develop a comprehensive risk prediction model specifically for ACN.
A retrospective cross-sectional study was conducted on patients who underwent endoscopy at the Affiliated Wuxi People's Hospital of Nanjing Medical University from January 1, 2021, to December 31, 2023. The researchers collected clinical data, including histopathological results, demographic information, lifestyle features, and medical history. Univariate and multivariable analyses were used to derive a risk prediction model based on the training cohort of the sample, and points were assigned to each variable to create a risk score. Scores with comparable risks were grouped into different risk categories. The model and its scores were tested using the validation cohort.
Results
Among the 2054 participants in the training cohort [mean age, 60 (52, 68) years; 1022 (50.0%) female], the prevalence of ACN was 271 (13.3%). The eight-variable model (sex, age, diabetes, family history of CRC in a first-degree relative, history of colonoscopy history, smoking, gastric histopathological results, and Helicobacter pylori infection history) resulted in two risk categories, with an area under the receiver operating characteristic curve (AUC) of 0.710 (95% CI: 0.679-0.742). In the validation cohort of 495 participants with an ACN prevalence of 62 (12.5%), model performance was comparable (AUC=0.715, 95% CI: 0.649-0.781). The ACN risk was 6.9% and 23.9% in the low- and high-risk groups, respectively, and 14 and 4 participants needed to be screened in the low- and high-risk groups, respectively. Further analysis revealed that three other prediction methods, including the Asia-Pacific Colorectal Screening (APCS) score [0.622 (0.556-0.688), p=0.049], the APCS score-revised edition [0.589 (0.524-0.653), p=0.007], and the colorectal tumor prediction score [0.570 (0.504-0.636), p=0.002], displayed significantly lower AUCs than our prediction ACN model.
Conclusion
The developed and validated prediction model performed well in identifying individuals with ACN at high risk in asymptomatic Chinese populations. Colonoscopy examinations in these patients were necessary, as they not only facilitated early detection of ACN but also enabled timely intervention to prevent CRC.