[Development and validation of a prediction model for postoperative recurrence risk of ovarian endometriotic cysts based on preoperative imaging and clinical biomarkers].

Objective

To develop and validate a prediction model for the postoperative recurrence of ovarian endometriotic cysts based on preoperative imaging and clinical serum biomarkers.

Methods

A retrospective study was conducted on 300 patients who underwent laparoscopic cystectomy at Shanxi Children's Hospital (Shanxi Maternal and Child Health Hospital) between June 2021 and June 2024. Patients were randomly assigned in a 7∶3 ratio to a training set (n=210) and an internal validation set (n=90) using a computer-generated random number method. Variables were selected using least absolute shrinkage and selection operator (LASSO) regression; multivariable logistic regression, random forest, and extreme gradient boosting (XGBoost) models were constructed; the predictive ability, accuracy, and clinical applicability of the model were evaluated using the area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis (DCA).

Results

The 300 patients had a mean age of (34.2±5.8) years. The median follow-up duration was 28 months (interquartile range: 24-38 months; range: 12-48 months), and recurrence occurred in 75 patients (25.0%) during the follow-up period. LASSO regression identified 10 predictive variables. Subsequent multivariate logistic regression analysis revealed that age<35 years (OR=3.24, 95%CI: 1.74-6.03), revised American Fertility Society (r-AFS) stage Ⅲ-Ⅳ (OR=3.42, 95%CI: 1.81-6.47), maximum cyst diameter≥6 cm (OR=2.89, 95%CI: 1.61-5.19), bilateral involvement (OR=2.56, 95%CI: 1.41-4.65), apparent diffusion coefficient<1.2×10-3 mm²/s (OR=2.31, 95%CI: 1.31-4.08), CA-125>70 U/ml (OR=2.78, 95%CI: 1.56-4.96), and absence of postoperative pharmacological treatment (OR=4.15, 95%CI: 2.18-7.89) were risk factors for postoperative recurrence, while the duration of pharmacological treatment was a protective factor (OR=0.88, 95%CI: 0.79-0.97) (all P<0.05). The AUC for the three models in the internal validation set were 0.839 (95%CI: 0.759-0.919), 0.876 (95%CI: 0.805-0.947), and 0.894 (95%CI:0.829-0.959), respectively; the XGBoost model outperformed the multivariate logistic regression model (P=0.008), whereas the difference compared with the Random Forest model was not statistically significant (P=0.162). All three models demonstrated good calibration (Hosmer-Lemeshow test χ² values: 6.83, 5.92, and 5.14; P-values: 0.555, 0.656, and 0.743, respectively). Within the threshold probability range of 10% to 60%, the extreme gradient boosting model yielded the highest net benefit.

Conclusion

The extreme gradient boosting model exhibits good predictive performance for the postoperative recurrence of ovarian endometriotic cysts and can serve as a reference for risk stratification and personalized follow-up.