硬膜外分娩镇痛产妇产时发热风险预测模型的构建与内部验证

Development and Internal Validation of a Risk Prediction Model for Intrapartum Fever in Parturients Receiving Epidural Labor Analgesia

摘要: 目的：构建并内部验证硬膜外分娩镇痛产妇产时发热风险预测候选模型，为产程中发热高风险产妇的早期识别、动态监测和分层管理提供依据，并进一步明确模型在不同产程时点中的适用边界。方法：选取2025年1~12月合肥市第二人民医院接受硬膜外分娩镇痛的足月单胎头位产妇为研究对象，开展单中心回顾性队列研究。研究对象按7:3比例分层随机分为建模集与内部验证集，主要结局为镇痛启动至胎儿娩出期间体温 ≥ 38.0˚C。仅纳入风险评估前可获取的临床、产科、镇痛、炎症及产程干预相关指标，排除结局相关后续变量。通过LASSO回归筛选变量，结合多因素Logistic回归构建预测模型，采用AUC、校准曲线、Brier评分及决策曲线评估模型效能。按预测变量可获得时间，将入院及镇痛前可获得指标用于早期风险识别，将第一产程进展、阴道检查次数和镇痛持续时间等产程中获得指标用于相应节点后的动态风险更新。结果：本研究共纳入480例产妇，产时发热发生率为16.2% (78/480)，其中建模集336例(发热55例)、验证集144例(发热23例)。模型最终纳入入院体温 ≥ 37.2˚C、胎膜早破、缩宫素使用、阴道检查 ≥ 4次、第一产程 ≥ 8 h、镇痛时长 ≥ 6 h、SII ≥ 850、NLR ≥ 4.0八项预测因子。建模集与验证集AUC分别为0.860、0.817，Brier评分分别为0.095、0.099，且产时发热发生率随模型预测风险等级升高逐步递增。结论：基于常规临床资料、产程变量和炎症指标构建的联合模型对硬膜外分娩镇痛后产时发热显示出可接受的内部验证性能，可为体温监测强化、感染风险再评估和多学科分层管理提供量化参考；该模型目前可作为候选风险分层工具，暂不单独用于临床决策，需经多中心外部验证、模型校准及前瞻性研究后再推广应用。

Abstract: Objective: To develop and internally validate a risk prediction model for intrapartum fever among parturients receiving epidural labor analgesia, and to support early identification, dynamic surveillance and stratified intrapartum management. The model is positioned as a candidate tool with explicitly defined timing of predictor availability. Methods: This single-center retrospective cohort study was conducted, enrolling term singleton cephalic presentation parturients who received epidural labor analgesia at Hefei Second People’s Hospital from January to December 2025. Subjects were divided into a development set and an internal validation set via stratified random sampling at a 7:3 ratio. The primary outcome was intrapartum fever defined as core temperature ≥ 38.0˚C from epidural analgesia initiation to fetal delivery. Only clinical, obstetric, analgesic, inflammatory and labor intervention indicators available prior to risk assessment were included, while post-outcome variables were excluded. LASSO regression was adopted for variable screening, followed by multivariate Logistic regression to construct the prediction model. Model performance was evaluated via the area under the receiver operating characteristic curve (AUC), calibration curves, Brier score and decision curve analysis (DCA). According to the availability timeline of predictive variables, indicators accessible upon admission and before epidural analgesia were used for early risk stratification; parameters obtained mid-labor, including progression of the first stage of labor, frequency of vaginal examinations and duration of epidural analgesia, were applied to dynamically update risk assessment after corresponding labor timepoints. Results: A total of 480 parturients were enrolled, with an overall intrapartum fever incidence of 16.2% (78/480). The development set included 336 women (55 febrile cases), and the validation set included 144 women (23 febrile cases). Eight predictors were finally incorporated into the model: admission body temperature ≥ 37.2˚C, premature rupture of membranes (PROM), oxytocin administration, ≥4 vaginal examinations, first stage of labor ≥ 8 hours, epidural analgesia duration ≥ 6 hours, systemic immune-inflammation index (SII) ≥ 850, and neutrophil-to-lymphocyte ratio (NLR) ≥ 4.0. The AUC values of the development set and validation set were 0.860 and 0.817 respectively, with corresponding Brier scores of 0.095 and 0.099. The incidence of intrapartum fever increased progressively with ascending predicted risk strata generated by the model. Conclusions: The combined model constructed based on routine clinical data, labor-related variables and inflammatory indicators exhibited acceptable internal validation performance for predicting intrapartum fever after epidural labor analgesia. It can provide quantitative references for intensified temperature surveillance, re-evaluation of infectious risks and multidisciplinary stratified management. As a candidate risk stratification tool, this model shall not be independently applied to guide clinical decision-making at present. Multicenter external validation, model recalibration and prospective clinical trials are required prior to its widespread clinical implementation.