Development of a risk assessment model for the recurrence of high-grade squamous intraepithelial lesions and differentiated vulvar intraepithelial neoplasia in Chinese cohort

Objective To construct a sophisticated risk prediction model for the recurrence of high-grade squamous intraepithelial lesions (HSIL) and differentiated vulvar intraepithelial neoplasia (dVIN) after treatment. This model will help the early detection and focused screening of individuals at increased risk of dVIN. Methods The clinical data from 257 patients diagnosed with dVIN or HSIL were retrospectively reviewed. Patients were divided into two distinct cohorts: relapse ( n = 60) and non-recurrence ( n = 197). For robust model development, the dataset was methodically divided into two subsets: training (70% of cases) and validation (30% of cases). Logistic regression analysis was applied to identify key predictors. Subsequently, they were combined to construct a risk prediction model for post-treatment recurrence of HSIL and dVIN. Results Univariate logistic regression analysis revealed that age, menopause, immunosuppression, HPV16 infection, histopathological characteristics, and positive surgical margins were positively correlated with recurrence risk. In contrast, low-risk HPV types exhibited a negative correlation with recurrence risk (all P < 0.05). Multivariable logistic regression analysis revealed that age, smoking history, immunosuppression, HPV16 infection, and histopathology were robustly associated with an increased recurrence risk (all P < 0.05). In the training dataset, the area under the receiver operating characteristic curve (AUC) was 0.793 (95% CI 0.77–0.880), accompanied by a median prediction success probability of 0.810. In the internal validation dataset, the AUC improved to 0.831 (95% CI 0.726–0.937). The Hosmer–Lemeshow goodness-of-fit test revealed acceptable model calibration for the training ( P = 0.069) and internal validation ( P = 0.086) sets. Calibration curves revealed trends remarkably consistent with the ideal curve, indicating commendable calibration. Furthermore, clinical decision curve analysis substantiated the net benefit of the model. Conclusion A prediction model including age, smoking history, immunosuppression, HPV16 infection, and histopathology exhibits good predictive performance for post-treatment recurrence of HSIL and dVIN. Our model can help clinicians assess recurrence risk, providing guidance for clinical consultations and enabling targeted follow-up and treatment plans for individuals at high risk.