基于机器学习算法的小于胎龄儿风险预测

张瑞敏; 王科科; 李金波; 陈转转; 杨海澜; 邬惟为; 冯永亮; 王素萍; 张新日

doi:10.16462/j.cnki.zhjbkz.2023.08.009

基于机器学习算法的小于胎龄儿风险预测

doi: 10.16462/j.cnki.zhjbkz.2023.08.009

1.
山西医科大学公共卫生学院流行病学教研室，太原 030001
2.
山西医科大学第一医院健康管理中心，太原 030001
3.
山西医科大学第一医院妇产科，太原 030001
4.
山西医科大学第一医院呼吸与危重症医学科，太原 030001

基金项目:

山西省基础研究计划青年科学研究项目 20210302124581

详细信息

通讯作者:
张新日，E-mail: ykdzxr61@163.com

WANG Suping, E-mail: supingwang@sxmu.edu.cn

中图分类号: R173
计量
- 文章访问数: 112
- HTML全文浏览量: 47
- PDF下载量: 21
- 被引次数: 0
出版历程
- 收稿日期: 2022-12-16
- 修回日期: 2023-01-26
- 网络出版日期: 2023-09-02
- 刊出日期: 2023-08-10

Risk prediction of small for gestational age birth based on machine learning algorithms

1.
Department of Epidemiology, School of Public Health, Shanxi Medical University, Taiyuan 030001, China
2.
Health Management Center, the First Hospital of Shanxi Medical University, Taiyuan 030001, China
3.
Obstetrics and Gynecology, the First Hospital of Shanxi Medical University, Taiyuan 030001, China
4.
Pulmonary and Critical Care Medicine, the First Hospital of Shanxi Medical University, Taiyuan 030001, China

Funds:

Youth Scientific Research Project of Fundamental Research Program in Shanxi Province 20210302124581

More Information

Corresponding author: ZHANG Xinri, E-mail: ykdzxr61@163.com

摘要

摘要: 目的评价极端梯度提升(extreme gradient boosting, XGBoost)、支持向量机(support vector machine, SVM)和朴素贝叶斯等6种机器学习模型与传统logistic回归分析模型对小于胎龄儿(small for gestational age, SGA)的预测效能。方法选取2012年3月―2016年9月在山西医科大学第一医院产科住院分娩的9 972例孕妇作为研究对象，采用问卷调查及从医院信息系统收集数据。依据分娩结局分为SGA组(n=1 124)与非SGA组(n=8 848)，按7.50∶2.50比例划分训练集与测试集。采用多因素logistic回归模型筛选危险因素，基于XGBoost、SVM、朴素贝叶斯、梯度提升决策树(gradient boosting decision tree, GBDT)、K最近邻(k-nearest neighbor, KNN)算法及传统logistic回归分析模型方法分别建立预测模型，使用受试者工作特征曲线的曲线下面积(area under the curve, AUC)、准确率和精确度等指标比较预测性能。结果 Logistic回归模型结果显示，妊娠期高血压和子痫等7项变量是SGA的影响因素。将以上因素纳入预测模型，SVM算法构建的预测模型效能最佳，AUC达0.72，模型准确率为71%。传统logistic回归分析模型表现欠佳，AUC为0.71，准确率为66%。结论基于机器学习算法尤其是SVM算法建立的SGA风险预测模型具有较好的效能，能够有效预测山西省SGA的发生，为实现SGA的一级预防提供参考。
- 小于胎龄儿 /
- 机器学习 /
- 风险预测模型 /
- logistic回归模型
Abstract: Objective To evaluate the performance of risk prediction of five machine learning models and traditional logistic regression models, such as, extreme gradient boosting (XGBoost), support vector machine (SVM), and Naive Bayes, aimed at small for gestational age (SGA). Methods A total of 9 972 women who gave birth in the First Hospital of Shanxi Medical University from March 2012 to September 2016 were selected as the research subjects in this study. Their data was collected from the hospital information system and through questionnaire surveys. Based on delivery outcomes, each case was put into one of two categories: an SGA group (n=1 124) and a non-SGA group (n=8 848), with the trial set and test set according to the ratio of 7.50∶2.50. Multivariate logistic regression model were used to screen the influencing factors. To establish predictive models, XGBoost, SVM, Naive Bayes, gradient boosting decision tree (GBDT) and k-nearest neighbor (KNN) algorithms were used. Furthermore, their predictive performance was measured with metrics such as the area under the curve (AUC), accuracy, and precision. Results Logistic regression analysis showed that gestational hypertension and eclampsia were among the seven variables related to the occurrence of SGA. By incorporating such variables into the machine learning algorithms and traditional logistic regression, the SVM model achieved the best performance with the highest AUC of 0.72 and 71% accuracy. Comparatively, compared to the SVM model, the logistic regression-based model was under performing, with an AUC of 0.71 and 66% accuracy. Conclusions Machine learning models, especially SVM, are capable of more accurately evaluating the risk of the occurrence of SGA in Shanxi Province, and can provide a reference for the primary prevention of SGA.
- Small for gestational age /
- Machine learning /
- Risk prediction model /
- Logistic regression model

HTML全文

图 1 SGA风险预测模型ROC曲线图

1. SGA: 小于胎龄儿；2. ROC：受试者工作特征；3. AUC: 受试者工作特征曲线的曲线下面积；4. XGBoost：极端梯度提升；5. SVM：支持向量机；6. GBDT：梯度提升决策树；7. KNN：K最近邻；8. Naive Bayes：朴素贝叶斯。

Figure 1. ROC curve of risk prediction model of SGA

1. SGA: small for gestational age; 2. ROC: receiver operating characteristic; 3. AUC, area under the receiver operating characteristic curve; 4. XGBoostC: extreme gradient boosting; 5. SVM: support vector machine; 6. GBDT: gradient boosting decision tree; 7. KNN: k-nearest neighbor; 8. Naive Bayes: Naive Bayes.

下载: 全尺寸图片幻灯片

表 1 SGA的一般情况及影响因素分析

Table 1. Analysis of general characteristics and influencing factors of SGA

变量Variable	合计[人数(占比/%)] Total[Number of people (proportion/%)](n=9 972)	SGA[人数(占比/%)] [Number of people (proportion/%)](n=1 124)	非SGA[人数(占比/%)] Non-SGA[Number of people (proportion/%)](n=8 848)	χ²值 value	P值 value
年龄组/岁Age group/years				33.915	＜0.001
＜25	1 085(10.88)	177(15.75)	908(10.26)
25~＜30	4 468(44.81)	478(42.53)	3 990(45.10)
30~＜35	3 003(30.11)	305(27.13)	2 698(30.49)
≥35	1 416(14.20)	164(14.59)	1 252(14.15)
民族Nationality				2.664	0.103
汉族Han nationality	9 897(99.25)	1 120(99.64)	8 777(99.20)
少数民族Minority nationality	75(0.75)	4(0.36)	71(0.80)
文化程度Education level				141.997	＜0.001
初中及以下Junior high school and below	1 725(17.30)	321(28.56)	1 404(15.87)
高中High school	1 199(12.02)	165(14.68)	1 034(11.69)
大专及以上Junior college and above	7 048(70.68)	638(56.76)	6 410(72.44)
家庭人均月收入/元Household per capita monthly income/yuan				87.580	＜0.001
＜2 000	1 229(12.33)	224(19.93)	1 005(11.36)
2 000~＜4 000	6 097(61.14)	692(61.57)	5 405(61.09)
≥4 000	2 646(26.53)	208(18.50)	2 438(27.55)
孕前BMI/(kg·m^-2) Pre-pregnancy BMI /(kg·m^-2)				3.014	0.389
消瘦Emaciated	1 200(12.03)	146(12.99)	1 054(11.91)
正常Normal	6 691(67.10)	733(65.21)	5 958(67.34)
超重Overweight	1 521(15.25)	185(16.46)	1 336(15.10)
肥胖Obese	560(5.62)	60(5.34)	500(5.65)
妊娠高血压综合征家族史Family history of gestational hypertensive syndrome				2.466	0.116
有Yes	84(0.84)	14(1.25)	70(0.79)
无No	9 888(99.16)	1 110(98.75)	8 778(99.21)
产检次数/次Number of prenatal examinations/time				88.808	＜0.001
＜7	2 641(26.48)	429(38.17)	2 212(25.00)
≥7	7 331(73.52)	695(61.83)	6 636(75.00)
接受羊水穿刺检查Accepted amniocentesis				0.438	0.508
是Yes	57(0.57)	8(0.71)	49(0.55)
否No	9 915(99.43)	1 116(99.29)	8 799(99.45)
孕期吸烟Smoking during pregnancy				0.350	0.554
是Yes	12(0.12)	2(0.18)	10(0.11)
否No	9 960(99.88)	1 122(99.82)	8838(99.89)
自然妊娠Natural pregnancy				0.376	0.540
是Yes	8 146(81.72)	926(82.38)	7 223(81.63)
否No	1 823(18.28)	198(17.62)	1 625(18.37)
妊娠期高血压Gestational hypertension				434.368	＜0.001
是Yes	777(7.79)	264(23.49)	513(5.80)
否No	9 195(92.21)	860(76.51)	8 335(94.20)
子痫Eclampsia				75.359	＜0.001
是Yes	82(0.82)	34(3.02)	48(0.54)
否No	9 890(99.18)	1 090(96.98)	8 800(99.46)
妊娠期胆汁淤积Intrahepatic cholestasis of pregnancy				6.050	0.014
是Yes	32(0.32)	8(0.71)	24(0.27)
否No	9 940(99.68)	1 116(99.29)	8 824(99.73)
被动吸烟Passive smoking				7.249	0.007
是Yes	1 006(10.09)	139(12.37)	867(9.80)
否No	8 966(89.91)	985(87.63)	7 981(90.20)
孕早期发热Fever in the first trimester				9.140	0.003
是Yes	180(1.81)	33(2.94)	147(1.66)
否No	9 792(98.19)	1 091(97.06)	8 701(98.34)
孕期消化不良Indigestion during pregnancy				0.526	0.468
是Yes	11(0.11)	2(0.18)	9(0.10)
否No	9 961(99.89)	1 122(99.82)	8 839(99.90)
注：SGA, 小于胎龄儿。 Note: SGA, small for gestational age.

下载: 导出CSV

表 2 SGA发生的多因素logistic回归分析模型

Table 2. Results of multivariate logistic regression of SGA

变量Variable	β值 value	Wald值 value	OR值(95% CI) OR value (95% CI)	P值 value
年龄组/岁Age group/years
＜25			1.000
25~＜30	-0.161	2.407	0.851(0.694~1.043)	0.236
30~＜35	-0.205	3.442	0.815(0.656~1.012)	0.064
≥35	-0.223	3.288	0.800(0.629~1.018)	0.070
文化程度Education level
初中及以下Junior high school and below			1.000
高中High school	-0.245	5.010	0.783(0.631~0.970)	0.025
大专及以上Junior college and above	-0.426	22.764	0.653(0.548~0.778)	＜0.001
家庭人均月收入/元Household per capita monthly income /yuan
＜2 000			1.000
2 000~＜4 000	-0.168	3.159	0.846(0.703~1.017)	0.076
≥4 000	-0.403	11.823	0.668(0.531~0.841)	0.001
产检次数/次Number of prenatal examinations /time
＜7			1.000
≥7	-0.261	12.722	0.771(0.668~0.889)	＜0.001
妊娠期高血压Gestational hypertension
否No			1.000
是Yes	1.348	228.299	3.849(3.232~4.584)	＜0.001
子痫Eclampsia
否No			1.000
是Yes	0.908	13.540	2.479(1.528~4.020)	＜0.001
妊娠期胆汁淤积Intrahepatic cholestasis of pregnancy
否No			1.000
是Yes	0.944	4.909	2.570(1.115~5.922)	0.027
被动吸烟Passive smoking
否No			1.000
是Yes	-0.069	0.433	0.933(0.760~1.147)	0.511
孕早期发热Fever in the first trimester
否No			1.000
是Yes	0.536	6.871	1.710(1.145~2.553)	0.009
注：SGA, 小于胎龄儿。 Note: SGA, small for gestational age.

下载: 导出CSV

表 3 机器学习模型的预测效能指标

Table 3. Predictive performance indicators of the machine learning prediction model

模型 Model	AUC	精确度 Accuracy	准确率 Precision	灵敏度 Sensitivity	f1分数 f1-score
SVM	0.72	0.67	0.71	0.40	0.52
GBDT	0.70	0.67	0.65	0.56	0.60
XGBoost	0.70	0.67	0.66	0.57	0.61
朴素贝叶斯 Naive Bayes	0.69	0.66	0.76	0.36	0.49
KNN	0.62	0.60	0.58	0.48	0.53
logistic	0.71	0.64	0.66	0.34	0.44
注：SVM, 支持向量机；GBDT, 梯度提升决策树；XGBoost, 极端梯度提升；KNN, K最近邻；AUC, 受试者工作特征曲线的曲线下面积。 Note: SVM, support vector machine; GBDT, gradient boosting decision tree; XGBoost, extreme gradient boosting; KNN, k-nearest neighbor; AUC, area under the receiver operating characteristic curve.

下载: 导出CSV

参考文献(20)

[1]	Physical status: the use and interpretation of anthropometry. Report of a WHO Expert Committee[J]. World Health Organ Tech Rep Ser, 1995, 854: 1-452.
[2]	Lee AC, Katz J, Blencowe H, et al. National and regional estimates of term and preterm babies born small for gestational age in 138 low-income and middle-income countries in 2010[J]. Lancet Glob Health, 2013, 1(1): e26-e36. DOI: 10.1016/S2214-109X(13)70006-8.
[3]	沈忠周, 王雅文, 马帅, 等. 新生儿早产、低出生体重及小于胎龄的危险因素[J]. 中华流行病学杂志, 2019, 40(9): 1125-1129. DOI: 10.3760/cma.j.issn.0254-6450.2019.09.020. Shen ZZ, Wang YW, Ma S, et al. Risk factors for preterm birth, low birth weight and small for gestational age: a prospective cohort study[J]. Chin J Epidemiol, 2019, 40(9): 1125-1129. DOI: 10.3760/cma.j.issn.0254-6450.2019.09.020.
[4]	von Beckerath AK, Kollmann M, Rotky-Fast C, et al. Perinatal complications and long-term neurodevelopmental outcome of infants with intrauterine growth restriction[J]. Am J Obstet Gynecol, 2013, 208(2): 130. e1-130. e6. DOI: 10.1016/j.ajog.2012.11.014.
[5]	Eves R, Mendonça M, Bartmann P, et al. Small for gestational age-cognitive performance from infancy to adulthood: an observational study[J]. BJOG, 2020, 127(13): 1598-1606. DOI: 10.1111/1471-0528.16341.
[6]	Lindqvist PG, Molin J. Does antenatal identification of small-for-gestational age fetuses significantly improve their outcome?[J]. Ultrasound Obstet Gynecol, 2005, 25(3): 258-264. DOI: 10.1002/uog.1806.
[7]	D'Ascenzo F, De Filippo O, Gallone G, et al. Machine learning-based prediction of adverse events following an acute coronary syndrome (PRAISE): a modelling study of pooled datasets[J]. Lancet, 2021, 397(10270): 199-207. DOI: 10.1016/S0140-6736(20)32519-8.
[8]	欧阳平, 李小溪, 冷芬, 等. 机器学习算法在体检人群糖尿病风险预测中的应用[J]. 中华疾病控制杂志, 2021, 25(7): 849-853, 868. DOI: 10.16462/j.cnki.zhjbkz.2021.07.020. Ouyang P, Li XX, Leng F, et al. Application of machine learning algorithm in diabetes risk prediction of physical examination population[J]. Chin J Dis Control Prev, 2021, 25(7): 849-853, 868. DOI: 10.16462/j.cnki.zhjbkz.2021.07.020.
[9]	朱丽, 张蓉, 张淑莲, 等. 中国不同胎龄新生儿出生体重曲线研制[J]. 中华儿科杂志, 2015, 53(2): 97-103. DOI: 10.3760/cma.j.issn.0578-1310.2015.02.007. Zhu L, Zhang R, Zhang SL, et al. Chinese neonatal birth weight curve for different gestational age[J]. Chin J Pediatr, 2015, 53(2): 97-103. DOI: 10.3760/cma.j.issn.0578-1310.2015.02.007.
[10]	中国肥胖问题工作组数据汇总分析协作组. 我国成人体重指数和腰围对相关疾病危险因素异常的预测价值: 适宜体重指数和腰围切点的研究[J]. 中华流行病学杂志, 2002, 23(1): 5-10. DOI: 10.3760/j.issn:0254-6450.2002.01.003. Coorperative Meta-analysis Group of China Obesity Task Force. Predictive values of body mass index and waist circumference to risk factors of related diseases in Chinese adult population[J]. Chin J Epidemiol, 2002, 23(1): 5-10. DOI: 10.3760/j.issn:0254-6450.2002.01.003.
[11]	Li Y, Guo H, Xiao L, et al. Adapted ensemble classification algorithm based on multiple classifier system and feature selection for classifying multi-class imbalanced data[J]. Knowledge-Based Systems, 2016, 94: 88-104. DOI: 10.1016/j.knosys.2015.11.013.
[12]	Royal College of Obstetricians and Gynaecologists. The investigation and management of the small for gestational age fetus. Green-top Guideline No. 31[EB/OL]. (2013-03-22)[2023-02-16]. https://www.rcog.org.uk/en/guidelines-research-services/guidelines/gtg31/.
[13]	Papastefanou I, Wright D, Lolos M, et al. Competing-risks model for prediction of small-for-gestational-age neonate from maternal characteristics, serum pregnancy-associated plasma protein-A and placental growth factor at 11-13 weeks' gestation[J]. Ultrasound Obstet Gynecol, 2021, 57(3): 392-400. DOI: 10.1002/uog.23118.
[14]	Gürgen F, Zengin Z, Varol F. Intrauterine growth restriction (IUGR) risk decision based on support vector machines[J]. Expert Syst Appl, 2012, 39(3): 2872-2876. DOI: 10.1016/j.eswa.2011.08.147.
[15]	Gardosi J, Madurasinghe V, Williams M, et al. Maternal and fetal risk factors for stillbirth: population based study[J]. BMJ, 2013, 346: f108. DOI: 10.1136/bmj.f108.
[16]	Gurung S, Tong HH, Bryce E, et al. A systematic review on estimating population attributable fraction for risk factors for small-for-gestational-age births in 81 low-and middle-income countries[J]. J Glob Health, 2022, 12: 04024. DOI: 10.7189/jogh.12.04024.
[17]	Parihar S, Singh S. Perinatal outcomes and intrahepatic cholestasis of pregnancy: a prospective study[J]. Int J Reprod Contracept Obstet Gynecol, 2019, 8(3): 1177-1182. DOI: 10.18203/2320-1770.ijrcog20190901.
[18]	Natarajan V, Singh P, Vigneshwar NKV, et al. Maternal and placental risk factors for small gestational age and fetal malnutrition[J]. Curr Pediatr Rev, 2023, 19(2): 187-196. DOI: 10.2174/1573396318666220705154424.
[19]	Dreiseitl S, Ohno-Machado L. Logistic regression and artificial neural network classification models: a methodology review[J]. J Biomed Inform, 2002, 35(5-6): 352-359. DOI: 10.1016/S1532-0464(03)00034-0.
[20]	Vapnik VN, Kotz S. Estimation of dependences based on empirical data[M]. New York: Springer Science & Business Media, 2006: 232-457.