目的 构建基于多时间点的动态预测模型，为胎龄<32周早产儿支气管肺发育不良（bronchopulmonary dysplasia, BPD）的早期诊断和个体化干预提供依据。 方法 回顾性分析2016年1月—2020年11月中南大学湘雅二医院收治的472例胎龄<32周的早产儿的临床资料。采用多因素logistic回归分析，分别在生后第1、7、14、21和28天建立5个独立的预测模型。模型性能通过受试者操作特征曲线下面积和Hosmer-Lemeshow检验评估。 结果 BPD组（n=147）与非BPD组（n=325）的胎龄、出生体重等基线指标比较，差异有统计学意义（P<0.05）。多因素分析显示，各时间点BPD发生的预测因子具有动态演变特征：第1天主要包括胎龄、出生体重、新生儿急性生理学评分Ⅱ、有创机械通气和吸入氧浓度>30%；第7天新增禁食时间>2 d、平均加奶速度<8.5 mL/（kg·d）、新生儿呼吸窘迫综合征、早产儿呼吸暂停及痰培养阳性等变量；第14天及以后则新增营养与治疗相关指标。生后第1、7、14、21和28天预测模型的判别效能良好，受试者操作特征曲线下面积依次为0.917、0.927、0.939、0.944和0.968，且Hosmer-Lemeshow检验显示校准良好（P>0.05）；内部验证组受试者操作特征曲线下面积介于0.899~0.958之间，模型具有较好的稳健性。 结论 基于围产期因素及呼吸支持、营养管理、治疗措施等方面的多个指标构建的胎龄<32周早产儿生后多时间点动态预测模型具有较高的预测效能，有助于BPD的动态风险评估。

Objective To develop dynamic prediction models based on multiple postnatal time points to support early diagnosis and individualized intervention for bronchopulmonary dysplasia (BPD) in preterm infants with gestational age < 32 weeks. Methods Clinical data of 472 preterm infants with gestational age <32 weeks admitted to the Second Xiangya Hospital of Central South University between January 2016 and November 2020 were retrospectively analyzed. Multivariable logistic regression was applied to develop five independent prediction models at postnatal days 1, 7, 14, 21, and 28. The performance of the models was assessed using the area under the receiver operating characteristic curve (AUC) and the Hosmer-Lemeshow test. Results Baseline characteristics such as gestational age and birth weight differed significantly between the BPD group (n=147) and the non-BPD group (n=325) (P<0.05). Predictors of BPD evolved across time points: on day 1, key predictors included gestational age, birth weight, Score for Neonatal Acute Physiology II (SNAP-II), invasive mechanical ventilation, and fraction of inspired oxygen >30%; by day 7, additional variables emerged, including fasting duration >2 days, mean feeding advancement rate <8.5 mL/(kg·d), neonatal respiratory distress syndrome, apnea of prematurity, and positive sputum culture; from day 14 onward, nutrition- and treatment-related indicators were incorporated additionally. The models demonstrated good discrimination at postnatal days 1, 7, 14, 21, and 28, with AUCs of 0.917, 0.927, 0.939, 0.944, and 0.968, respectively, and good calibration (Hosmer-Lemeshow P>0.05). Internal validation showed AUCs ranging from 0.899 to 0.958, indicating robust performance. Conclusions Dynamic postnatal prediction models incorporating indicators spanning perinatal factors, respiratory support, nutritional management, and therapeutic interventions demonstrate high predictive performance and facilitate dynamic risk assessment for BPD in preterm infants with gestational age < 32 weeks.