宫腔粘连术后复发风险预测模型构建与验证

Development and Validation of a Prediction Model for Postoperative Recurrence Risk of Intrauterine Adhesions

摘要: 目的：探讨宫腔粘连术后复发的相关影响因素，构建并评价基于临床资料及术前超声特征的术后复发风险预测模型。方法：回顾性纳入2023年2月至2025年6月在郑州大学第五附属医院接受宫腔粘连分离术且术后6个月复发结局可判定的患者348例，其中复发143例、未复发205例。收集临床资料及术前超声特征，对缺失比例较低的候选预测变量采用多重插补处理，采用单因素和多因素Logistic回归分析构建预测模型，进行1000次bootstrap内部验证，并进一步比较完整模型与去除术前子宫内膜厚度后的简化模型性能。结果：复发组在粘连持续时间、宫腔粘连分离术次数、美国生育学会(American Fertility Society, AFS)评分、术前子宫内膜厚度及子宫内膜连续性方面与未复发组差异有统计学意义。多因素分析显示，AFS评分、术前子宫内膜连续性差和宫腔粘连分离术次数为独立预测因素，术前子宫内膜厚度与复发风险呈负向关联趋势。Bootstrap内部验证显示，模型AUC为0.777 (95%CI: 0.774~0.780)，敏感度为0.648，特异度为0.829，校准斜率为0.930。完整模型与简化模型比较显示，完整模型在AUC和AIC方面略优。结论：基于AFS评分、术前子宫内膜连续性差、宫腔粘连分离术次数和术前子宫内膜厚度构建的预测模型经内部验证显示在本中心回顾性数据中显示出一定区分度和校准度，可为宫腔粘连术后复发风险评估及个体化随访管理提供初步参考。与简化模型相比，完整模型在区分度和拟合优度方面略占优势，且术前子宫内膜厚度具有一定临床可解释性，故最终予以保留。

Abstract: Objective: To investigate factors associated with postoperative recurrence of intrauterine adhesions and to construct and evaluate a postoperative recurrence risk prediction model based on clinical data and preoperative ultrasound features. Methods: A total of 348 patients who underwent transcervical resection of adhesions between February 2023 and June 2025 and had evaluable 6-month postoperative recurrence outcomes were retrospectively included, comprising 143 patients with recurrence and 205 without recurrence. Clinical data and preoperative ultrasound features were collected. Candidate predictors with a low proportion of missing data were handled using multiple imputation. Univariate and multivariate Logistic regression analyses were performed to construct the prediction model, followed by internal validation using 1000 bootstrap resamples. The performance of the full model was further compared with that of a simplified model excluding preoperative endometrial thickness. Results: Significant differences were observed between the recurrence and non-recurrence groups in adhesion duration, number of TCRA procedures, American Fertility Society (AFS) score, preoperative endometrial thickness, and endometrial continuity. Multivariate analysis showed that AFS score, poor preoperative endometrial continuity, and number of TCRA procedures were independent predictors, while preoperative endometrial thickness showed a negative association with recurrence risk. Bootstrap internal validation yielded an AUC of 0.777 (95% CI: 0.774~0.780), a sensitivity of 0.648, a specificity of 0.829, and a calibration slope of 0.930. Comparison between the full and simplified models showed that the full model performed slightly better in terms of AUC and AIC. Conclusion: The prediction model incorporating AFS score, poor preoperative endometrial continuity, number of TCRA procedures, and preoperative endometrial thickness demonstrated a certain degree of discrimination and calibration after internal validation using retrospective data from this center and may provide a preliminary reference for postoperative recurrence risk assessment and individualized follow-up management in patients with intrauterine adhesions. Compared with the simplified model, the full model showed slightly better discrimination and goodness of fit. Preoperative endometrial thickness was therefore retained because of its clinical interpretability.