Development and internal validation of a vaginal microecology-based multivariable prediction model for persistent...

Persistent high-risk human papillomavirus (HR-HPV) infection is a key determinant of cervical carcinogenesis, yet early identification of women at increased risk of persistence remains challenging in routine clinical practice. Vaginal microecological disturbances have been associated with HR-HPV persistence, but their integration into individualized risk prediction models has been insufficiently explored.

To develop and internally validate a multivariable prediction model for persistent/recurrent HR-HPV positivity at 12 months using vaginal microecological and clinical characteristics.

This retrospective study included women with baseline HR-HPV positivity, available baseline vaginal microecological examination results, and completed 12-month HR-HPV follow-up testing at a tertiary hospital gynecologic outpatient clinic between January 2019 and December 2024. A total of 1,186 women were included and randomly divided into a training cohort (n = 830) and an internal validation cohort (n = 356). Persistent/recurrent HR-HPV positivity was defined as any HR-HPV positivity at the 12-month follow-up among women who were HR-HPV-positive at baseline, regardless of whether the same HR-HPV genotype was detected at both time points. Candidate predictors were selected in the training cohort using least absolute shrinkage and selection operator (LASSO) regression with ten-fold cross-validation. Variables retained at the one-standard-error criterion (λ_1se) were entered into a multivariable logistic regression model, which was subsequently presented as a nomogram. Model discrimination, calibration, and clinical utility were evaluated using the area under the receiver operating characteristic curve (AUC), calibration curves with the Hosmer-Lemeshow goodness-of-fit test, and decision curve analysis, respectively.

Among the 1,186 women, 354 (29.8%) had persistent/recurrent HR-HPV positivity at 12 months. LASSO regression retained seven predictors for final model development: age, smoking status, HPV16/18 infection, vaginal pH, non-Lactobacillus-dominant microbiota, bacterial vaginosis, and moderate-to-severe local inflammation. In multivariable logistic regression, all seven variables remained independently associated with persistent/recurrent HR-HPV positivity. HPV16/18 infection showed the strongest association (OR 8.564, 95% CI 5.383-13.955, P < 0.001). The final model showed good apparent discrimination in the training cohort (AUC 0.892, 95% CI 0.870-0.914) and maintained acceptable discrimination in the internal validation cohort (AUC 0.821, 95% CI 0.775-0.867). Calibration was acceptable in both cohorts, with Hosmer-Lemeshow P-values of 0.149 and 0.296, respectively. Decision curve analysis suggested potential net clinical benefit across a range of threshold probabilities in both cohorts.

We developed and internally validated a vaginal microecology-based multivariable prediction model for persistent/recurrent HR-HPV positivity that showed promising discrimination, acceptable calibration, and potential clinical usefulness within this single-center cohort. This internally validated model may support individualized risk stratification and help identify women who may benefit from closer follow-up; however, external validation in independent cohorts is needed before broader clinical implementation.