Objective To explore the expression characteristics and risk stratification value of P16 protein （P16） and ferroptosis markers in the progression of cervical lesions, construct a clinical prediction model based on the P16-ferroptosis pathway, and evaluate its potential in precise intervention for cervical cancer with the help of machine learning methods.

Methods A total of 438 patients with cervical biopsy were selected, covering four pathological stages of cervical intraepithelial neoplasia （CIN） 1, CIN2, CIN3 and cervical cancer. The clinical data of patients and seven ferroptos-related scores （GPX4, SLC7A11, ACSL4, LPO, Ferritin, TFRC and P16 scores） were collected to evaluate the distribution characteristics of different scores in pathological grade and progression status. The progress risk stratification was achieved through k-means clustering analysis, and an XGBoost prediction model based on P16 and ferroptosis scores, and logistic regression model were constructed for comparison. SHAP analysis was used to clarify the contribution degree of features. The nomogram was further constructed to evaluate its visualization application effects in risk assessment.

Results The differences of P16 and various ferroptosis scores in the patients with different pathological types were statistically significant （P ＜ 0.01）. The P16 score was significantly negatively correlated with the SLC7A11 score （r = –0.78, P ＜ 0.01）, and the ferroptosis load was significantly positively correlated with the inflammation score （r = 0.811, P ＜ 0.01）. The results of logistic regression analysis showed that the highest quartile of the P16 score was an independent risk factor of disease progression OR = 7.006, 95%CI: 1.382–39.764, P = 0.02）. The patients were divided into six subtypes by k-means clustering, among which the progression rates of subtypes C5 and C6 were significantly higher （84.1% and 76.3%, respectively）. The predictive efficacy of the XGBoost model was comparable to that of the logistic model （AUC was 0.685 and 0.700, respectively）, and the results of SHAP analysis indicated that the contribution of the P16 score was the highest （SHAP = 2.075）. The nomogram constructed based on the above features had good intuitiveness and clinical practicability.

Conclusions P16 and ferroptosis pathways are closely associated with the progression of cervical lesions. The combined modeling of the two can effectively achieve risk stratification of progression. The XGBoost prediction model and nomogram can provide the reliable tools for accurately identifying high-risk populations. This model shows stable performance in multi-center samples, and has further clinical transformation and application value.