[Development of a risk prediction model for colorectal adenoma in non-smoking women].

Objective

To explore risk factors for colorectal adenoma (CRA) in non-smoking women, develop a simplified and efficient predictive model, and evaluate its performances with existing risk evaluation tools and in different time periods.

Methods

Clinical data were collected from non-smoking women between November 2021 and April 2023. The positive case group included patients with colorectal polyps confirmed as CRA by pathology. Candidate variables were identified through single-factor logistic analysis (P<0.2), and prelimiarily screened using multivariate logistic regression. To obtain the optimal model, a stepwise regression method based on the AIC was employed for predictor selection, and the final prediction model was constructed with the selected predictors. A nomogram and model list were developed to visually demonstrate the contribution of each factor. The model's discrimination and calibration were evaluated and compared with existing risk assessment tools, including the Asia-Pacific Colorectal Screening (APCS) score, its modified version (MAPCS), and the "Colorectal Cancer Screening and Early Diagnosis and Treatment Program (2024 Edition)" of China. A temporal external test set was used to further evaluate the model's stability and predictive performance in a real-world clinical setting.

Results

After analyzing data from 1 155 non-smoking women, the final model based on age (5 age groups) and BMI (≥24.0 kg/m²) as the main predictive factors was constructed. The model achieved area under the curve (AUC) values of 0.705 (95%CI: 0.672-0.738) in the training set (n=927) and 0.695 (95%CI: 0.629-0.762) in the validation set (n=228), with calibration curves and Hosmer-Lemeshow tests showing good fitness (P>0.05). A risk threshold of 0.400 was applied, with predicted probabilities ≥0.400 indicating high-risk and <0.400 indicating non-high-risk. The model achieved stable performance in the temporal external test set (n=272) with an AUC of 0.783 (95%CI: 0.730-0.836), further confirming the model's temporal stability and clinical utility. Compared with the existing risk evaluation tools, the values of the model in terms of discrimination are slightly higher than those of the high-risk groups in APCS, MAPCS and "Colorectal Cancer Screening and Early Diagnosis and Treatment Program (2024 Edition)", and the values in terms of specificity and accuracy are also higher.

Conclusions

The simplified prediction model based on age and BMI can effectively evaluate CRA risk in non-smoking women, demonstrating high discriminatory power and temporal stability. It can provide more precise risk stratification guidance for early CRA screening with improved efficiency.