极早产儿住院期间死亡的列线图预测模型的建立

Establishment of a predictive nomogram model for predicting the death of very preterm infants during hospitalization

摘要:

目的 构建预测极早产儿住院期间死亡风险的列线图模型。 方法 回顾性分析2015年1月至2019年12月郑州大学第三附属医院新生儿科收治的极早产儿1 714例的临床资料。按7∶3比率将1 714例极早产儿随机分为训练队列（1 179例）和验证队列（535例），通过logistic回归分析筛选独立预测因子并建立列线图模型，并由验证集评估列线图预测模型的可行性。最后，分别采用受试者工作特征曲线下面积（area under curve，AUC）、校准曲线和决策曲线分析对模型的鉴别能力、准确性和临床实用性进行评估。 结果 1 714例极早产儿中，住院期间死亡260例，存活1 454例。对训练集进行多因素logistic回归分析后筛选出胎龄<28周、出生体重<1 000 g、重度窒息、重度脑室内出血（intraventricular hemorrhage，IVH）、Ⅲ~Ⅳ级新生儿呼吸窘迫综合征（respiratory distress syndrome，RDS）、败血症、剖宫产、孕母产前使用糖皮质激素等8个变量建立列线图预测模型。训练队列中列线图模型预测极早产儿住院期间死亡发生的AUC为0.790（95%CI：0.751~0.828），验证队列中列线图模型预测极早产儿住院期间死亡发生的AUC为0.808（95%CI：0.754~0.861）。Hosmer-Lemeshow拟合优度检验显示出较好的拟合度（P>0.05）。决策曲线分析显示当训练队列和验证队列的阈值概率分别为10%~60%和10%~70%时对极早产儿进行临床干预具有较高的净收益。 结论 构建并验证了预测极早产儿住院期间死亡风险的预测模型，可帮助临床医生预测极早产儿住院期间的死亡概率。

Abstract:

Objective To establish a nomogram model for predicting the risk of death of very preterm infants during hospitalization. Methods A retrospective analysis was performed on the medical data of 1 714 very preterm infants who were admitted to the Department of Neonatology, the Third Affiliated Hospital of Zhengzhou University, from January 2015 to December 2019. These infants were randomly divided into a training cohort (1 179 infants) and a validation cohort (535 infants) at a ratio of 7∶3. The logistic regression analysis was used to screen out independent predictive factors and establish a nomogram model, and the feasibility of the nomogram model was assessed by the validation set. The area under the receiver operating characteristic curve (AUC), calibration curve, and decision curve analysis (DCA) were used to assess the discriminatory ability, accuracy, and clinical applicability of the model. Results Among the 1 714 very preterm infants, 260 died and 1 454 survived during hospitalization. By the multivariate logistic regression analysis of the training set, 8 variables including gestational age <28 weeks, birth weight <1 000 g, severe asphyxia, severe intraventricular hemorrhage (IVH), grade III-IV respiratory distress syndrome (RDS), and sepsis, cesarean section, and use of prenatal glucocorticoids were selected and a nomogram model for predicting the risk of death during hospitalization was established. In the training cohort, the nomogram model had an AUC of 0.790 (95%CI: 0.751-0.828) in predicting the death of very preterm infants during hospitalization, while in the validation cohort, it had an AUC of 0.808 (95%CI: 0.754-0.861). The Hosmer-Lemeshow goodness-of-fit test showed a good fit (P>0.05). DCA results showed a high net benefit of clinical intervention in very preterm infants when the threshold probability was 10%-60% for the training cohort and 10%-70% for the validation cohort. Conclusions A nomogram model for predicting the risk of death during hospitalization has been established and validated in very preterm infants, which can help clinicians predict the probability of death during hospitalization in these infants.