口服普萘洛尔治疗婴幼儿血管瘤的差效预测模型建立及外部验证

摘要：目的　建立口服普萘洛尔治疗婴幼儿血管瘤（IHs）的疗效预测模型并对其进行验证。方法　收集2011年6月—2018年12月分别于天津市儿童医院及新疆乌鲁木齐儿童医院接受口服普萘洛尔治疗IHs的患儿为建模组，通过对其临床资料的单因素及多因素Logistic回归分析筛选出影响普萘洛尔治疗效果的因素并建立差效预测模型。然后收集2019年1—10月两院收治的口服普萘洛尔治疗IHs的113例患儿作为验证组（好效83例，差效30例），利用受试者工作特征（ROC）曲线外部验证差效预测模型的区分效度。结果　585例患儿作为建模组，根据疗效分为好效组（371例）和差效组（214例）。建模组患儿胎龄＜37周、出生体质量＜2 500 g、病灶大小（病灶5~10 cm、＞10 cm相对于＜5 cm）及服药年龄＞3个月是影响口服普萘洛尔治疗IHs差效的独立危险因素。据此建立的差效预测模型：P=ex/（1+ex），其中e为自然对数，X=-1.082+0.680×（胎龄）+0.665×（出生体质量）-0.920×（服药年龄）+0.375×（病灶5~10 cm）或［+1.327×（病灶＞10 cm）］。拟合度检测提示预测模型的拟合度良好（P=0.766），预测模型的ROC曲线下面积为0.779（95%CI：0.740~0.818）。预测模型外部验证的ROC曲线下面积为0.772（95%CI：0.639～0.784），能较好地将差效患儿区分出来。结论　模型预测对口服普萘洛尔治疗IHs为差效的准确性高，有助于提高对此类患儿的早期识别和筛选能力。

Establishment of poor effect model and external validation of oral propranolol in the treatment of infantile hemangioma

Abstract: Objective To establish and validate the model for predicting the efficacy of oral propranolol for infantile hemangiomas (IHs). Methods The children who received oral propranolol for IHs in Tianjin Children's Hospital and Xinjiang Urumqi Children's Hospital were collected from June 2011 to December 2018 as the model group. Univariate and multivariate Logistic regression analysis were used to screen out the factors affecting the therapeutic effect of propranolol and establish the poor-effect prediction model. A total of 113 IHs cases treated with oral propranolol in two hospitals from January to October 2019 were collected as the validation group (83 cases with good effect and 30 cases with poor effect). The receiver operating characteristics (ROC) curve was used to verify the discriminant validity of the poor-effect prediction model. Results A total of 585 children were used as the model group and were divided into good effect group (n=371) and poor effect group (n=214) according to the efficacy. In the modeling group, gestational age less than 37 weeks, the birth weight less than 2 500 g, the sizes of the lesions (the lesion size was 5-10 cm, >10 cm relative to <5 cm), and the age of medication more than 3 months were independent risk factors for poor effect of oral propranolol for IHs. The poor-effect prediction model was established as P=ex/(1+ex), ‘e’ is the natural logarithm, X=-1.082+0.680×(gestational age) + 0.665×(birth weight)－0.920×(medicine age)+0.375×(lesion size 5-10 cm) or (+1.327×lesion> 10 cm). The detection of the fit degree indicated that the fit degree of the prediction model was good (P=0.766), and the area under the ROC curve of the prediction model was 0.779 (95%CI: 0.740-0.818). The area under the ROC curve of external validation for the prediction model was 0.772 (95%CI: 0.639-0.784), which can better distinguish the children with poor effect. Conclusion The accuracy of the model in predicting the probability of oral propranolol for IHs with poor effect as a difference is high, which is helpful to improve the ability of early identification and screening of children with poor effect.