研究目的

分析人工流产女性的人口学特征、孕产史、避孕行为特征等资料，探索重复人工流产的影响因素并构建预测模型，为降低非意愿妊娠所导致的重复人工流产发生率，推动和实现保护女性生育力及提高妇幼健康水平的目标提供理论依据。

研究方法

人工流产生育风险的前瞻性队列研究（中国医学科学院医学与健康创新工程重大协同创新项目，CIFMS，2018-I2M-1-004)经过国家卫生健康委科学技术研究所伦理委员会审批，并在ClinicalTrials.gov（NCT04183829）注册。研究对象为全国8个省市10家临床研究中心自2019年5月起至2022年2月所招募的因非意愿妊娠而实施人工流产的女性，通过纸质调查问卷收集记录妊娠妇女的人口学特征、孕产史、个人史等相关资料信息。采用SPSS 23.0和R 4.1.3软件进行数据整理及统计描述分析。应用卡方检验、二元Logistic回归进行单因素和多因素分析，初步探索重复人工流产的影响因素。分别采用Logistic回归和随机森林算法构建预测模型，输出变量重要性排序并对模型预测效果进行评估及比较。

研究结果

本研究共纳入9245例研究对象，重复人工流产的女性有4031例，占比为43.6%，其中人工流产次数2次的女性有2637例（65.4%），3次及以上的女性有1394例（34.6%）；半年内重复人工流产的女性有258例（6.4%），半年以上的女性有3773例（93.6%）。研究对象的平均年龄为（31.23±5.10）岁，最大年龄为49岁，最小年龄为19岁，其中年龄不足25岁女性有872例，占比为9.5%；未婚女性有2559例，占比为27.7%。

单因素分析结果发现，重复人工流产与年龄、婚姻状况、文化程度、职业、月收入情况、地区GDP水平、吸烟及饮酒频率、子女数、阴道分娩次数、剖宫产次数、生育计划、性生活频率、避孕情况有关。

按人工流产次数及时间间隔进一步进行亚组分析结果发现，人工流产次数与除上述避孕情况外的其他因素均有关，人工流产时间间隔仅与年龄、婚姻状况、子女数、剖宫产次数有关。

多因素分析结果发现高龄、低文化程度、地区较高GDP水平、吸烟、饮酒、育有子女、性生活频率高、离异或丧偶、避孕失败、其他职业以及工农从业者或家务工作者是重复人工流产的危险因素。

采用逐步回归法构建的Logistic回归模型在测试集的预测准确度为65.10%，灵敏度为79.41%，特异度为46.93%，ROC曲线下面积为0.683（95%CI：0.647-0.720）。采用随机森林算法构建的预测模型在测试集的预测准确度为65.33%，灵敏度为79.41%，特异度为47.47%，ROC曲线下面积为0.683（95%CI：0.646-0.719）。变量重要性排序结果提示年龄、文化程度、子女数、阴道分娩次数、地区GDP水平与重复人工流产的发生关系更为密切。

研究结论

人工流产女性仍以育龄的已婚已育女性为主，但年龄不足25岁、未婚以及未育女性的比例仍处于较高的水平，同时因避孕失败所致的非意愿妊娠比例超过半数。2019年至2022年间的重复人工流产率相对保持稳定，整体发生率依然接近半数，其中人工流产次数3次及以上的女性比例为34.6%，半年内再次人工流产的比例为6.4%。

年龄、婚姻状况、文化程度、职业、地区GDP水平、吸烟频率、饮酒频率、子女数、生育计划、性生活频率以及避孕情况是重复人工流产的独立影响因素。此外年龄、婚姻状况、子女数还与重复人工流产的时间间隔有关。

采用Logistic回归构建的预测模型和随机森林算法构建的预测模型的预测效果较为接近，具有较好的一致性，预测模型整体的灵敏度较高，便于尽早发现重复人工流产发生风险较高的女性，从而为进一步开展针对性的干预措施以及科学的避孕宣传教育提供理论依据和支撑。

Objective

The study aims to analyze the data information of women seeking induced abortion , that includes demographic characteristics, history of gestation and contraceptive behavior, and to explore the influencing factors of repeated induced abortion and constructing a  prediction model so as to provide theoretical basis for the goal of reducing the incidence of repeated induced abortion caused by unwanted pregnancy and protecting the female fertility in order to improve the health of women and children.

Methods

The prospective cohort study on reproductive risk of induced abortion (CAMS Innovation Fund for Medical Sciences, CIFMS, 2018-I2M-1-004) approved by the Ethics Committee of National Research Institute for Family Planning and registered at ClinicalTrials.gov (NCT04183829). The subjects seeking induced abortion due to unwanted pregnancy were recruited from ten clinical research centers of eight provinces in China from May 2019 to February 2022. The demographic characteristics, history of gestation and personal histories of pregnant women were collected and recorded through a questionnaire survey. Data processing and statistical description and analysis were performed by using SPSS 23.0 and R 4.1.3 software. Chi-square test and binary Logistic regression were used for univariate and multivariate analysis to explore the influencing factors of repeated induced abortion. The Logistic regression and random forests were used to construct prediction model, and output the importance ranking of variables and to evaluate and compare the prediction performance of the prediction model.

Results

(1) A total of 9245 women were included in this research, and 4031 women had repeated induced abortion, accounting for 43.6%. Among them, 2637 women (65.4%) had two previous induced abortion, and 1394 women (34.6%) had three and more previous induced abortion. There were 258 women (6.4%) of repeated induced abortions within six months, and 3773 women (93.6%) of repeated induced abortions over six months. The mean age of the subjects was (31.23±5.10) years, among which 872 were under 25 years old, accounting for 9.5%. There were 2,559 unmarried women, accounting for 27.7%.

(2) The univariate analysis showed that repeated induced abortion was related to the age, marital status, educational level, occupation, monthly income, regional GDP level, frequency of smoking and drinking, number of children, number of vaginal delivery, number of cesarean section, fertility intentions, frequency of sexual intercourse and contraceptive use.

(3) The subgroup analysis according to the number of induced abortion and interval time further found that the number of induced abortion was related to the above factors except the contraceptive use, and the interval of induced abortion was only related to age, marital status, number of children, and the number of cesarean section.

(4) The risk factors of repeated induced abortion were older age, lower education level, higher GDP level, smoking, drinking, having children, high sexual intercourse frequency, divorce or widower, contraceptive failure, and other occupation or worker, farmer and household worker.

(5) The prediction accuracy of the prediction model constructed by the stepwise logistic regression in the testing set was 65.10%, the sensitivity was 79.41%, the specificity was 46.93%, and the area under the ROC was 0.683 (95%CI:0.647-0.720). The prediction accuracy of the prediction model constructed by the random forests in the testing set was 65.33%, the sensitivity was 79.41%, the specificity was 47.47%, and the area under the ROC was 0.683 (95%CI:0.646-0.719). The importance ranking of variables indicated that age, education level, number of children, number of vaginal deliveries and regional GDP level were more closely related to the occurrence of repeated induced abortion.

Conclusion

(1) The majority of induced abortion women were still married women of childbearing age, but the proportions of unmarried and childless women under 25 years old were still at a high level, and the proportion of unwanted pregnancy caused by contraceptive failure was more than half. The rate of repeated induced abortions remained relatively stable between 2019 and 2022, with the overall incidence remaining nearly half, with 34.6% women having three or more abortions and 6.4% having a second abortion within six months.

(2) The age, marital status, educational level, occupation, regional GDP level, smoking frequency, drinking frequency, number of children, fertility intentions, sexual intercourse frequency and contraceptive use were independent factors of repeated induced abortion. In addition, age, marital status and number of children were also related to the time interval of repeated induced abortion.

(3) The prediction performance of the prediction model constructed by Logistic regression and the random forest was relatively close and had good consistency. The sensitivity of the prediction model was high, which facilitated the early detection of women with high risk of repeated induced abortion, thus providing theoretical basis and support for further targeted intervention measures and scientific contraceptive publicity and education.