Poisson回归在辅助生殖与自然受孕新生儿出生缺陷中的应用
|
摘要
“辅助生殖技术”被誉为20世纪改变人类生活的重大科技发明之一。它的出现为不孕不育患者带来了福音,目前已经成为有效治疗不孕不育的最佳方法。但是,随着科技的逐步发展,辅助生殖技术的应用越来越广泛,其出生子女的安全性问题引起了人们的关注,也成为全球研究者激烈争论的热点。因此,了解辅助生殖技术对新生儿结局的影响,尤其,是否增加新生儿出生缺陷的风险,以及新生儿出生缺陷的相关影响因素,成为国际社会共同关注的目标。
目前用于分析出生缺陷问题的统计方法主要是Logistic回归,但是它对于分析低发生率疾病的影响因素,存在一定的局限性。因此,研究者结合Poisson分布的特点发展了新的回归模型—Poisson回归模型。它是多变量非线性回归分析的扩展,其实质是对数线性方程,其理论基础是基于Poisson分布。一方面,它可以用于单位时间、面积、空间内某事件发生数的影响因素的分析。另一方面,可以用于以人群为基础的稀有疾病、卫生事件资料的影响因素的分析。故本研究通过引入Poisson回归模型对辅助生殖与自然受孕的新生儿出生缺陷及其影响因素的资料进行拟合,分析了各影响因素对新生儿出生缺陷的直接效应和关联,为制定新生儿出生缺陷的对策提供科学依据。
研究目的
本研究的主要目的是:通过对辅助生殖与自然受孕的新生儿出生缺陷Poisson回归模型中参数的优化进行探讨,比较不同模型参数的影响,设定合适的参数建立影响新生儿出生缺陷发生的Poisson回归模型。并通过拟合模型的参数,分析各因素对新生儿出生缺陷的影响大小进行衡量。本研究为Poisson回归模型的方法学研究提供一定的参考依据,并通过确定新生儿出生缺陷的影响因素,探讨各因素之间的作用关系,为新生儿出生缺陷的预防和干预提供科学依据。
资料与方法
本研究资料来源于国家重大科学研究(973项目)“辅助生殖技术显微操作安全性研究”,在浙江大学医学院附属妇产科医院收集2003-2007年间,共计辅助生殖受孕例数1067例,自然受孕2134例。以应用辅助生殖技术后妊娠大于等于28周的全部受孕者为暴露组,以同期的自然受孕者为非暴露组。年龄作为控制混杂因素的条件,通过以辅助生殖受孕的女性年龄(年龄±2岁)作为匹配条件,利用SPSS统计软件随机抽取1:2本院同年的自然受孕者作为非暴露组,两者建立回顾性队列。以新生儿出生时的结构畸形为研究结局。新生儿结构畸形以国际疾病诊断(ICD-10)为标准。
数据分析主要采用SAS9.1和Stata10.0统计软件。主要分析内容包括描述性统计分析、单因素回归分析以及多因素分析。采用的统计学方法为Poisson回归和1:2条件Logistic回归等。多因素分析是在描述性分析和单因素回归分析的基础上,系统地分析和研究影响新生儿出生缺陷的因素,进而评价新生儿出生缺陷的Poisson回归模型优越性,为研究新生儿出生缺陷的模型建立提供参考。
主要结果
1.Poisson回归多因素分析
经过多因素Poisson回归分析后,所得结果如下:a模型表示:与新生儿出生缺陷有关的因素为受孕方式、既往流产史和新生儿性别有关。与自然受孕相比,辅助生殖受孕其分娩新生儿出生缺陷危险度增加,大约增加1.962倍。与母亲既往无流产史的相比,母亲有流产史的其分娩新生儿出生缺陷危险度增加,大约增加1.833倍。与新生儿为女孩相比,新生儿为男孩其出生缺陷危险度增加,大约增加1.650倍。模型拟合度检验Pearsonχ~2值为26.000,v=26,P＞0.05,说明模型拟合良好。b模型表示:与新生儿出生缺陷有关的因素为母亲年龄和新生儿胎数。与母亲分娩年龄为25-30岁相比,年龄过小,其分娩新生儿出生缺陷的危险度明显增加,大约增加3倍。与新生儿单胎相比,随着母亲分娩新生儿胎数的逐渐增加,其分娩新生儿出现出生缺陷的危险度增加,大约增加1.850倍。模型拟合度检验Pearsonχ~2值为21.293,v=20,P＞0.05,说明模型拟合良好。
2.Poisson回归向前法、向后法和逐步回归法的比较
三种回归模型进行比较,可以得出,新生儿出生缺陷采用Poisson回归向前法和向后法,所得结果相同,两种筛选自变量的方法之间不存在差异。而同逐步回归法相比,存在较多不同,主要为以下几方面:
首先,从分析结果来看,前者的两个模型分析结果为与新生儿出生缺陷的主要因素为既往流产史和受孕方式。而后者模型分析结果为既往流产史、新生儿性别和受孕方式。根据文献资料,后者的分析结果可靠性更高。
其次,三者模型虽然拟合均好,P＞0.05,但是,根据模型的自变量筛选标准,“确定系数越大,说明模型拟合越好”,这一原则,可以得出,逐步回归法的确定系数稍高于其它两种方法,因此,对于新生儿出生缺陷的相关问题可以采用逐步回归分析法。
最后,从三种回归模型的标准误来看,三种回归模型相差不多,说明,各自变量对新生儿出生缺陷的解释误差相差不大。
综上所述,在新生儿出生缺陷的资料中,可以采用Poisson回归逐步分析法。
另外,对于连续性变量,拟合了两种回归模型,通过模型拟合度检验评判得出,将连续性变量转化成分类变量的模型拟合较好,其决定系数稍高于连续性变量,得出最终结果。
3.1:2条件Logistic回归多因素分析
经过1:2条件Logistic回归多因素分析后,所得结果如下:a模型表示:与新生儿出生缺陷有关的因素为受孕方式和既往流产史。与自然受孕相比,辅助生殖受孕其分娩新生儿出生缺陷危险度增加,大约增加2倍。与既往母亲无流产史的相比,既往母亲有流产史的其分娩新生儿出生缺陷危险度增加,大约增加2倍。其Pearsonχ~2值为20.908,v=5,P＜0.05,说明模型拟合不够良好。b模型表示:与新生儿出生缺陷有关的因素为新生儿胎数。随着新生儿胎数的增加,母亲分娩新生儿出生缺陷危险度增加,大约增加2倍。其Pearsonχ~2值为8.886,v=2,P＜0.05,说明模型拟合不够良好。
4.Poisson回归与1:2条件Logistic回归比较
首先,在自变量相同的条件下,分别进行Poisson回归与1:2条件Logistic回归,可以看出,前者的两个拟和模型P值均大于0.05,说明模型拟和良好。而后者的两个拟合模型P值均小于0.05,说明模型拟和不够良好。
其次,前者所得的各个自变量的标准误也相应的小于后者,说明前者的各个自变量综合结果误差较小。
最后,与新生儿出生缺陷有关的主要影响因素中,前者模型所得结果为既往流产史、新生儿性别、受孕方式、新生儿胎数和母亲的分娩年龄。而后者模型所得结果仅为既往流产史、新生儿性别、受孕方式和新生儿胎数。由于后者匹配掉了母亲的分娩年龄,因此,其对新生儿出生缺陷的影响不能很好的分析出来,结果不够准确。综合以上分析结果,新生儿出生缺陷的模型拟合可以采用“Poisson回归模型”。
结论
1.通过单因素及多因素Poisson回归分析,影响新生儿出生缺陷的主要因素是:新生儿胎数、母亲年龄、既往流产史、受孕方式等。同时也初步显示了这些影响因素对新生儿出生缺陷的效应值,最终模型拟合较好。
2.比较分析多因素Poisson回归筛选自变量的三种方法(逐步发、向前法、向后法),可以看出,对于新生儿出生缺陷的相关问题可以采用逐步回归法。
3.比较分析多因素Poisson回归分析与1:2条件Logistic回归,可以总结出对于发病率较低的疾病,可以采用Poisson回归模型进行拟合。
4.Poisson回归将数理知识与专业知识有机结合,是处理新生儿出生缺陷研究资料较为满意的方法,解决了方法学上的问题。虽具有许多独特的优越性,但它也有一定的局限和不足之处。
"Assisted reproductive technology" is known as one of the major scientific and technological invention that changed people's lives in the 20th century. As infertility patients, its emergence has brought the gospel, and has become the best way in infertility treatment. However, with the gradual development of science and technology, more and more infertility people choice assisted reproductive technology as treatment, whether it is more prone to birth defects or other pregnancy complications in their offspring born has aroused people's concern, and the safety problem of assisted reproductive technology has become the world's researchers hotly debated issue. Therefore, the understanding of assisted reproductive technology in neonatal outcomes particular in, whether to increase the risk of birth defects, and affecting factors has become common concern in the international community research area.
Currently it has been mainly used Logistic regression methods to analyze the birth defects, but there are some limitations when it is analysis the influencing factors of low incidence. Therefore, the researchers combined the characteristics of Poisson distribution and developed a new regression model-Poisson regression model. It is a expansion of multi-variable nonlinear regression analysis, and its essence is a log-linear equations, and its theoretical foundation is based on Poisson distribution. On the one hand, it can be used to analyze a few factors affecting in the unit of time, space, space, or an occurred event. On the other hand, it also can be used in population-based rare diseases, health event data rates. Therefore, this study, we will use Poisson regression model to introduce the birth defects and influencing factors in assisted reproduction and natural conception, analyze the impact of various factors on the direct effects and association of birth defects, and provide a scientific basis for the prevention of newborn defects.
Objective
The main purpose of this study are: through Poisson regression model to explore the optimization of parameters ,and then compare the effects of different model parameters, set the appropriate parameters of the establishment of Poisson regression models in assisted reproduction and natural conception of newborn birth defects. Using model parameters to decide which model is fitted. This study use Poisson regression model to provide a reference for research, and confirm the factors affecting birth defects in newborns, explore the role of the relationship between the factors, and provide a scientific basis for birth defects prevention and intervention.
Data and Methods
The data collected from the project of "assisted reproductive technology micro-operational safety study" in Women's Hospital School of Medicine Zhejiang University. The total number of cases of assisted reproductive fertility is 1067 cases; the other natural conception is 2134 cases from2003 to 2007. The full of application of assisted reproductive technology pregnancies are as exposed group, and the natural conception in same period pregnancies are as un-exposed group, they are all the survey objects. Age is controlled conditions as confounding factors, stratified and use SPSS statistical software to random sample of 1:2 in the same conceived year as a non-exposed group, both of them establish the retrospective cohort. The study outcome is the structure birth defects in newborns. Neonatal diagnosis of structural malformations is according to International Classification of Diseases (ICD-10).
Data analysis is mainly used SAS9.1 and Stata10.0 statistical software. The main analysis is included in descriptive statistical, single-factor regression and multi-factor analysis. The statistical methods are used for the Poisson regression and 1:2 conditions Logistic regression. Multivariate analysis is based on descriptive analysis and single-factor regression analysis, then systematically analysis and study of the factors affecting birth defects, and thus evaluation of birth defects in the superiority of the Poisson regression model for the study, while provide a reference for building a birth defects model.
Results
1. Poisson regression analysis
After multivariate Poisson regression analysis, the results are as follows: model a said that: it revealed that relating to the birth defects influence factors are past history of abortion, neonatal sex and exposure factors (assisted reproductive technology). Compared with non-abortion in the past, the mother who has a history of birth abortion, the risk of birth defects in newborn has increased by about 1.833 times. Compared with newborn girls, boys have increased by about 1.650 times. Compared with natural conception, assisted reproductive conception has increased by about 1.962 times. The test of model goodness (Pearson x~2 =26.000, v =26, P > 0.05,) shows that it is a good model. Model b said that: the number of neonatal and maternal age is related to birth defects. Single baby is the reference, it is the more baby, the more risk of birth defects, by about 1.850 times. Compared with the mother's age 25-30 years old, when the age is too small, its birth defects significantly increased risk has increased by about 3 times. Model goodness of fit test (Pearson x~2 =21.293, v= 20, P> 0.05,) shows that it is a good model.
2. Comparison the three methods of Poisson regression
Comparing the three kinds of regression model, you can conclude that using forward method and backward method of Poisson regression to analysis the birth defects, the results are similar, there is no difference in screening independent variables between the two kinds of methods. But compared with the stepwise regression, there are more differences, mainly include the following aspects:
First of all, from the results of the analysis point of view, the results of the former two models which analysis the major factor in birth defects are the past history of abortion and assisted reproductive fertility. The latter model analysis is the past history of abortion, neonatal sex and assisted reproductive fertility. Based on the literature, the latter results of the analysis are more reliable.
Secondly, although the three models are a good fit, P> 0.05, according to the model since the variable selection criteria, "the greater the coefficient of determination, the better model fit", this principle can be drawn that, the coefficient in stepwise regression is slightly higher than the other two methods. Therefore, application of stepwise regression models to analysis the birth defects is a relatively good method.
Finally, from three kinds of standard error of the regression model view, three kinds of regression models very similar, and show that the error in interpretation each variable on birth defects are similar.
In summary, in birth defect data, the application of stepwise Poisson regression analysis is a better choice.
In addition, for continuous variables, through the model goodness of fit test evaluation concludes that the model is better when continuous variables transferred into categorical variables, and then the coefficient is slightly higher.
3.1:2 condition Logistic regression analysis
After 1:2 condition Logistic regression analysis, the results are as follows: model a said that: it is post that relating to the birth defects influence factors are past history of abortion and exposure factors (assisted reproductive technology). Compared with non-abortion in the past, the mother who has a history of birth abortion, the risk of birth defects in newborn has increased by about 2 times. Compared with natural conception, assisted reproductive conception has increased by about 2 times. The test of model goodness (Pearson x~2=20.908, v = 5, P <0.05,) shows that it is not a good model enough. Model b said that: the number of neonatal is related to birth defects. The more baby, the more risk of birth defects, it is by about 2 times risk. The test of model goodness (Pearson x~2 = 8.886, v=2, P <0.05,) shows that also it isn't a good model enough.
4. Poisson regression compared with 1:2 conditions Logistic Regression
First of all, under the same conditions as independent variables, implicate respectively Poisson regression and 1:2 conditions, Logistic regression, we can see that, the former two models P >0.05, indicating that the model are fitting well. The latter two P <0.05, indicating that they are not good enough.
Secondly, the former independent variables standard error is smaller than the latter corresponding, indicating the former is less errors independent variables.
Finally, birth defects of the major factors, the former model results for the past history of abortion, neonatal gender, exposure factors, the number of newborns and mothers of child-bearing age. The latter model is only the results of past history of abortion, neonatal gender, exposure factors and neonatal number. Since the latter match the mother's birth age, and therefore its impact on birth defects are not well analyzed, the result is not accurate enough. Based on the above analysis, birth defects model fitting is best for the "Poisson regression model for birth defects in the multi-factor analysis."
Conclusions
1. By single factor and multivariate Poisson regression analysis, the main influence factors of birth defects are: the number of newborn, maternal age, past history of abortion, exposure factors. At the same time it also reveals the effect value of birth defects. In the final, the model is fit well.
2. A comparative analysis of multivariate Poisson regression of the three methods of screening variables (stepwise, forward and backward method), we can conclude that stepwise method is relatively good in analysis the birth defects.
3. A comparative of multivariate Poisson regression analysis and 1:2 conditions Logistic regression, it can be summed up that using Poisson regression model is a relatively perfect statistical method for the lower incidence of disease.
4. Poisson regression will be a combination of mathematic and expertise knowledge, it is a satisfactory ways to deal with research data, and solved more methodological problems. Although with many unique advantages, it also has some limitations and shortcomings.
目前用于分析出生缺陷问题的统计方法主要是Logistic回归,但是它对于分析低发生率疾病的影响因素,存在一定的局限性。因此,研究者结合Poisson分布的特点发展了新的回归模型—Poisson回归模型。它是多变量非线性回归分析的扩展,其实质是对数线性方程,其理论基础是基于Poisson分布。一方面,它可以用于单位时间、面积、空间内某事件发生数的影响因素的分析。另一方面,可以用于以人群为基础的稀有疾病、卫生事件资料的影响因素的分析。故本研究通过引入Poisson回归模型对辅助生殖与自然受孕的新生儿出生缺陷及其影响因素的资料进行拟合,分析了各影响因素对新生儿出生缺陷的直接效应和关联,为制定新生儿出生缺陷的对策提供科学依据。
研究目的
本研究的主要目的是:通过对辅助生殖与自然受孕的新生儿出生缺陷Poisson回归模型中参数的优化进行探讨,比较不同模型参数的影响,设定合适的参数建立影响新生儿出生缺陷发生的Poisson回归模型。并通过拟合模型的参数,分析各因素对新生儿出生缺陷的影响大小进行衡量。本研究为Poisson回归模型的方法学研究提供一定的参考依据,并通过确定新生儿出生缺陷的影响因素,探讨各因素之间的作用关系,为新生儿出生缺陷的预防和干预提供科学依据。
资料与方法
本研究资料来源于国家重大科学研究(973项目)“辅助生殖技术显微操作安全性研究”,在浙江大学医学院附属妇产科医院收集2003-2007年间,共计辅助生殖受孕例数1067例,自然受孕2134例。以应用辅助生殖技术后妊娠大于等于28周的全部受孕者为暴露组,以同期的自然受孕者为非暴露组。年龄作为控制混杂因素的条件,通过以辅助生殖受孕的女性年龄(年龄±2岁)作为匹配条件,利用SPSS统计软件随机抽取1:2本院同年的自然受孕者作为非暴露组,两者建立回顾性队列。以新生儿出生时的结构畸形为研究结局。新生儿结构畸形以国际疾病诊断(ICD-10)为标准。
数据分析主要采用SAS9.1和Stata10.0统计软件。主要分析内容包括描述性统计分析、单因素回归分析以及多因素分析。采用的统计学方法为Poisson回归和1:2条件Logistic回归等。多因素分析是在描述性分析和单因素回归分析的基础上,系统地分析和研究影响新生儿出生缺陷的因素,进而评价新生儿出生缺陷的Poisson回归模型优越性,为研究新生儿出生缺陷的模型建立提供参考。
主要结果
1.Poisson回归多因素分析
经过多因素Poisson回归分析后,所得结果如下:a模型表示:与新生儿出生缺陷有关的因素为受孕方式、既往流产史和新生儿性别有关。与自然受孕相比,辅助生殖受孕其分娩新生儿出生缺陷危险度增加,大约增加1.962倍。与母亲既往无流产史的相比,母亲有流产史的其分娩新生儿出生缺陷危险度增加,大约增加1.833倍。与新生儿为女孩相比,新生儿为男孩其出生缺陷危险度增加,大约增加1.650倍。模型拟合度检验Pearsonχ~2值为26.000,v=26,P＞0.05,说明模型拟合良好。b模型表示:与新生儿出生缺陷有关的因素为母亲年龄和新生儿胎数。与母亲分娩年龄为25-30岁相比,年龄过小,其分娩新生儿出生缺陷的危险度明显增加,大约增加3倍。与新生儿单胎相比,随着母亲分娩新生儿胎数的逐渐增加,其分娩新生儿出现出生缺陷的危险度增加,大约增加1.850倍。模型拟合度检验Pearsonχ~2值为21.293,v=20,P＞0.05,说明模型拟合良好。
2.Poisson回归向前法、向后法和逐步回归法的比较
三种回归模型进行比较,可以得出,新生儿出生缺陷采用Poisson回归向前法和向后法,所得结果相同,两种筛选自变量的方法之间不存在差异。而同逐步回归法相比,存在较多不同,主要为以下几方面:
首先,从分析结果来看,前者的两个模型分析结果为与新生儿出生缺陷的主要因素为既往流产史和受孕方式。而后者模型分析结果为既往流产史、新生儿性别和受孕方式。根据文献资料,后者的分析结果可靠性更高。
其次,三者模型虽然拟合均好,P＞0.05,但是,根据模型的自变量筛选标准,“确定系数越大,说明模型拟合越好”,这一原则,可以得出,逐步回归法的确定系数稍高于其它两种方法,因此,对于新生儿出生缺陷的相关问题可以采用逐步回归分析法。
最后,从三种回归模型的标准误来看,三种回归模型相差不多,说明,各自变量对新生儿出生缺陷的解释误差相差不大。
综上所述,在新生儿出生缺陷的资料中,可以采用Poisson回归逐步分析法。
另外,对于连续性变量,拟合了两种回归模型,通过模型拟合度检验评判得出,将连续性变量转化成分类变量的模型拟合较好,其决定系数稍高于连续性变量,得出最终结果。
3.1:2条件Logistic回归多因素分析
经过1:2条件Logistic回归多因素分析后,所得结果如下:a模型表示:与新生儿出生缺陷有关的因素为受孕方式和既往流产史。与自然受孕相比,辅助生殖受孕其分娩新生儿出生缺陷危险度增加,大约增加2倍。与既往母亲无流产史的相比,既往母亲有流产史的其分娩新生儿出生缺陷危险度增加,大约增加2倍。其Pearsonχ~2值为20.908,v=5,P＜0.05,说明模型拟合不够良好。b模型表示:与新生儿出生缺陷有关的因素为新生儿胎数。随着新生儿胎数的增加,母亲分娩新生儿出生缺陷危险度增加,大约增加2倍。其Pearsonχ~2值为8.886,v=2,P＜0.05,说明模型拟合不够良好。
4.Poisson回归与1:2条件Logistic回归比较
首先,在自变量相同的条件下,分别进行Poisson回归与1:2条件Logistic回归,可以看出,前者的两个拟和模型P值均大于0.05,说明模型拟和良好。而后者的两个拟合模型P值均小于0.05,说明模型拟和不够良好。
其次,前者所得的各个自变量的标准误也相应的小于后者,说明前者的各个自变量综合结果误差较小。
最后,与新生儿出生缺陷有关的主要影响因素中,前者模型所得结果为既往流产史、新生儿性别、受孕方式、新生儿胎数和母亲的分娩年龄。而后者模型所得结果仅为既往流产史、新生儿性别、受孕方式和新生儿胎数。由于后者匹配掉了母亲的分娩年龄,因此,其对新生儿出生缺陷的影响不能很好的分析出来,结果不够准确。综合以上分析结果,新生儿出生缺陷的模型拟合可以采用“Poisson回归模型”。
结论
1.通过单因素及多因素Poisson回归分析,影响新生儿出生缺陷的主要因素是:新生儿胎数、母亲年龄、既往流产史、受孕方式等。同时也初步显示了这些影响因素对新生儿出生缺陷的效应值,最终模型拟合较好。
2.比较分析多因素Poisson回归筛选自变量的三种方法(逐步发、向前法、向后法),可以看出,对于新生儿出生缺陷的相关问题可以采用逐步回归法。
3.比较分析多因素Poisson回归分析与1:2条件Logistic回归,可以总结出对于发病率较低的疾病,可以采用Poisson回归模型进行拟合。
4.Poisson回归将数理知识与专业知识有机结合,是处理新生儿出生缺陷研究资料较为满意的方法,解决了方法学上的问题。虽具有许多独特的优越性,但它也有一定的局限和不足之处。
"Assisted reproductive technology" is known as one of the major scientific and technological invention that changed people's lives in the 20th century. As infertility patients, its emergence has brought the gospel, and has become the best way in infertility treatment. However, with the gradual development of science and technology, more and more infertility people choice assisted reproductive technology as treatment, whether it is more prone to birth defects or other pregnancy complications in their offspring born has aroused people's concern, and the safety problem of assisted reproductive technology has become the world's researchers hotly debated issue. Therefore, the understanding of assisted reproductive technology in neonatal outcomes particular in, whether to increase the risk of birth defects, and affecting factors has become common concern in the international community research area.
Currently it has been mainly used Logistic regression methods to analyze the birth defects, but there are some limitations when it is analysis the influencing factors of low incidence. Therefore, the researchers combined the characteristics of Poisson distribution and developed a new regression model-Poisson regression model. It is a expansion of multi-variable nonlinear regression analysis, and its essence is a log-linear equations, and its theoretical foundation is based on Poisson distribution. On the one hand, it can be used to analyze a few factors affecting in the unit of time, space, space, or an occurred event. On the other hand, it also can be used in population-based rare diseases, health event data rates. Therefore, this study, we will use Poisson regression model to introduce the birth defects and influencing factors in assisted reproduction and natural conception, analyze the impact of various factors on the direct effects and association of birth defects, and provide a scientific basis for the prevention of newborn defects.
Objective
The main purpose of this study are: through Poisson regression model to explore the optimization of parameters ,and then compare the effects of different model parameters, set the appropriate parameters of the establishment of Poisson regression models in assisted reproduction and natural conception of newborn birth defects. Using model parameters to decide which model is fitted. This study use Poisson regression model to provide a reference for research, and confirm the factors affecting birth defects in newborns, explore the role of the relationship between the factors, and provide a scientific basis for birth defects prevention and intervention.
Data and Methods
The data collected from the project of "assisted reproductive technology micro-operational safety study" in Women's Hospital School of Medicine Zhejiang University. The total number of cases of assisted reproductive fertility is 1067 cases; the other natural conception is 2134 cases from2003 to 2007. The full of application of assisted reproductive technology pregnancies are as exposed group, and the natural conception in same period pregnancies are as un-exposed group, they are all the survey objects. Age is controlled conditions as confounding factors, stratified and use SPSS statistical software to random sample of 1:2 in the same conceived year as a non-exposed group, both of them establish the retrospective cohort. The study outcome is the structure birth defects in newborns. Neonatal diagnosis of structural malformations is according to International Classification of Diseases (ICD-10).
Data analysis is mainly used SAS9.1 and Stata10.0 statistical software. The main analysis is included in descriptive statistical, single-factor regression and multi-factor analysis. The statistical methods are used for the Poisson regression and 1:2 conditions Logistic regression. Multivariate analysis is based on descriptive analysis and single-factor regression analysis, then systematically analysis and study of the factors affecting birth defects, and thus evaluation of birth defects in the superiority of the Poisson regression model for the study, while provide a reference for building a birth defects model.
Results
1. Poisson regression analysis
After multivariate Poisson regression analysis, the results are as follows: model a said that: it revealed that relating to the birth defects influence factors are past history of abortion, neonatal sex and exposure factors (assisted reproductive technology). Compared with non-abortion in the past, the mother who has a history of birth abortion, the risk of birth defects in newborn has increased by about 1.833 times. Compared with newborn girls, boys have increased by about 1.650 times. Compared with natural conception, assisted reproductive conception has increased by about 1.962 times. The test of model goodness (Pearson x~2 =26.000, v =26, P > 0.05,) shows that it is a good model. Model b said that: the number of neonatal and maternal age is related to birth defects. Single baby is the reference, it is the more baby, the more risk of birth defects, by about 1.850 times. Compared with the mother's age 25-30 years old, when the age is too small, its birth defects significantly increased risk has increased by about 3 times. Model goodness of fit test (Pearson x~2 =21.293, v= 20, P> 0.05,) shows that it is a good model.
2. Comparison the three methods of Poisson regression
Comparing the three kinds of regression model, you can conclude that using forward method and backward method of Poisson regression to analysis the birth defects, the results are similar, there is no difference in screening independent variables between the two kinds of methods. But compared with the stepwise regression, there are more differences, mainly include the following aspects:
First of all, from the results of the analysis point of view, the results of the former two models which analysis the major factor in birth defects are the past history of abortion and assisted reproductive fertility. The latter model analysis is the past history of abortion, neonatal sex and assisted reproductive fertility. Based on the literature, the latter results of the analysis are more reliable.
Secondly, although the three models are a good fit, P> 0.05, according to the model since the variable selection criteria, "the greater the coefficient of determination, the better model fit", this principle can be drawn that, the coefficient in stepwise regression is slightly higher than the other two methods. Therefore, application of stepwise regression models to analysis the birth defects is a relatively good method.
Finally, from three kinds of standard error of the regression model view, three kinds of regression models very similar, and show that the error in interpretation each variable on birth defects are similar.
In summary, in birth defect data, the application of stepwise Poisson regression analysis is a better choice.
In addition, for continuous variables, through the model goodness of fit test evaluation concludes that the model is better when continuous variables transferred into categorical variables, and then the coefficient is slightly higher.
3.1:2 condition Logistic regression analysis
After 1:2 condition Logistic regression analysis, the results are as follows: model a said that: it is post that relating to the birth defects influence factors are past history of abortion and exposure factors (assisted reproductive technology). Compared with non-abortion in the past, the mother who has a history of birth abortion, the risk of birth defects in newborn has increased by about 2 times. Compared with natural conception, assisted reproductive conception has increased by about 2 times. The test of model goodness (Pearson x~2=20.908, v = 5, P <0.05,) shows that it is not a good model enough. Model b said that: the number of neonatal is related to birth defects. The more baby, the more risk of birth defects, it is by about 2 times risk. The test of model goodness (Pearson x~2 = 8.886, v=2, P <0.05,) shows that also it isn't a good model enough.
4. Poisson regression compared with 1:2 conditions Logistic Regression
First of all, under the same conditions as independent variables, implicate respectively Poisson regression and 1:2 conditions, Logistic regression, we can see that, the former two models P >0.05, indicating that the model are fitting well. The latter two P <0.05, indicating that they are not good enough.
Secondly, the former independent variables standard error is smaller than the latter corresponding, indicating the former is less errors independent variables.
Finally, birth defects of the major factors, the former model results for the past history of abortion, neonatal gender, exposure factors, the number of newborns and mothers of child-bearing age. The latter model is only the results of past history of abortion, neonatal gender, exposure factors and neonatal number. Since the latter match the mother's birth age, and therefore its impact on birth defects are not well analyzed, the result is not accurate enough. Based on the above analysis, birth defects model fitting is best for the "Poisson regression model for birth defects in the multi-factor analysis."
Conclusions
1. By single factor and multivariate Poisson regression analysis, the main influence factors of birth defects are: the number of newborn, maternal age, past history of abortion, exposure factors. At the same time it also reveals the effect value of birth defects. In the final, the model is fit well.
2. A comparative analysis of multivariate Poisson regression of the three methods of screening variables (stepwise, forward and backward method), we can conclude that stepwise method is relatively good in analysis the birth defects.
3. A comparative of multivariate Poisson regression analysis and 1:2 conditions Logistic regression, it can be summed up that using Poisson regression model is a relatively perfect statistical method for the lower incidence of disease.
4. Poisson regression will be a combination of mathematic and expertise knowledge, it is a satisfactory ways to deal with research data, and solved more methodological problems. Although with many unique advantages, it also has some limitations and shortcomings.
引文
[1]任娟,曹缵孙.辅助生殖技术与出生缺陷.国外医学计划生育/生殖健康分册,2007,26(2):79-82.
[2]乐杰.妇产科学.6版.北京:人民卫生出版社,2004:384.
[3]Paul Merlob,Onit Sapir,Jaqueline Sulkes,et al.The prevalence of major congenital malformations during two periods of time,1986-1994 and 1995-2002 in newborns conceived by assisted reproduction technology.European Journal of Medical Genetics,2005,48:5-11.
[4]Michele H.,Jennifer J.,Carol B.,et al.The risk of major birth defects after intracytoplasmic sperm injection and in vitro fertilization.The New England Journal of Medicine.2002,346(10):725-730.
[5]Christine K.,Kim M.,bPaul A.,etal.In vitro fertilization is associated with an increase in major birth defects.Fertility and Sterility.2005,84,(5):1308-1315.
[6]沈其君.SAS统计分析.第一版.北京:高等教育出版社,2005:169.
[7]郭志刚.社会统计分析方法-SPSS软件应用.第一版.北京:中国人民大学出版社,1999:.
[8]方积乾.卫生统计学.第五版.北京:人民卫生出版社,2005:362.
[9]Hwang IK.Superiority,Noninferiority and Equivalence trials.北京:新药临床研究生物统计学国际学术研讨会,2003:01.
[10]童峰,陈坤.常见结局事件的前瞻性研究中修正Poisson回归模型的应用[J].中国卫生统计,2006,23(5):410-412.
[11]方积乾.卫生统计学.第五版.北京:人民卫生出版社,2005:52-53.
[12]杨珉,李晓松.医学和公共卫生研究常用多水平统计模型.第一版.北京:北京大学医学出版社,2007:104.
[13]陈峰.医用多元统计分析方法.第二版.北京:中国统计出版社,2007:125.
[14]吴喜之主编.统计学:从数据到理论.第一版.北京:中国统计出版社, 2004:42-43.
[15]宇传华,余松林,向惠云等.Poisson回归在职业队列研究中的应用[J].中国卫生统计,1996,13(1):6-9.
[16]Samuel k.Encyclopedia of Statistical Sciences,Plackett Family of Distributions to Regression,Wrong.Vol7.New York:John Wiley & Sons,Inc,1986:18-43.
[17]Frome EL.The analysis of rates using Poisson regression models [J].Biometrics,1983:39(3):665.
[18]Frome EL.,Checkoway H.Use of Poisson regression models in estimating incidence rates and ratios[J].American Journal of Epidemiology,1985,121:309-323.
[19]Joshi AY.,Iyer VN.,Hagan JB.,et al.Incidence and Temporal Trends of Primary Immunodeficiency:A Population-Based Cohort Study[J].Mayo Clin.Proc.,2009:84(1):16-22.
[20]Villar E.,Remontet L.,Labeeuw M.,et al.Effect of Age,Gender,and Diabetes on Excess Death in End-Stage Renal Failure[J].J.Am.Soc.Nephrol,2007:18(7):2125-2134.
[21]沈其君.SAS统计分析.第一版.北京:高等教育出版社,2005:219.
[22]Selmer R.A comparison of Poisson regression model fitted to multiway summary tables and Cox's survival model using data from a blood pressure screening in the city of Bergen,Norway[J].StatMed,1990,9(10):1157-1165.
[23]Breslow NE.,Day NE.Statistical methods in cancer research.VolⅡ.Lyon:International Agency for Research on Cancer,1987:82.
[24]Diabetes Epidemiology Research International Group.Secular trends in the incidence of childhood IDDM in 10 countries[J].Diabetes,1990;39(7):858-864.
[25]Nystrom L,Dahlquist G,REWERS M.,et al.The Swedish Childhood diabetes study:an analysis of the temporal variation in diabetes incidence 1978-1987[J].International Journal Epidemiology,1990,19(1):141-146.
[26]陈峰.现代医学统计方法与Stata应用.第一版.北京:中国统计出版社,2000:197.
[27]陈峰.医学多元统计分析方法.第一版.北京:中国统计出版社,2001:125-128.
[28]张颖,倪宗瓒,姚树祥,等.广义Poisson回归模型及其应用[J].中国卫生统计,2000,17(13):130-133.
[29]李克,俞顺章.Poisson模型与队列研究-原理及应用[J].中国卫生统计,1997,14(1):46-48.
[30]Stefano Parodi,Ezio Bottarelli.Poisson regression modelin Epidemiology-an Introduction[J].Ann.Fac.Medic.Vet.DiParma,2006,(26):25-44.
[31]Michele H.,Jennifer J.,Carol B.,et al.The risk of major birth defects after intracytoplasmic sperm injection and in vitro fertilization.The New England Journal of Medicine.2002,346(10):725-730.
[32]Bergh T.,Ericson A.,Hillensj(o|¨) T.,et al.Deliveries and children born after in-vitro fertilisation in Sweden 1982-95:a retrospective cohort study.The Lancet,1999,354:1579-1585.
[33]修晓燕,张荣莲,陈起燕.出生缺陷的研究进展[J].中国妇幼健康研究,2007,18(5):400-402.
[34]郑晓瑛,宋新明,陈功等.中国出生缺陷高发地区出生缺陷的发生水平和流行病学特征[J].中华流行病学杂志,2007,28(1):5-9.
[35]Anthony S.,Buitendijk SE.,Dorrepaal CA.,et al.Congenital malformations in 4224 children conceived after IVF.Hum Reprod,2002,17(8):2089-2095
[36]胡荣华,杜玉开.武汉市围产儿先天畸形流行状况及影响因素研究[J].医学与社会,2005,18(5):7-11.
[37].郑晓英,宋新明,陈功等.中国出生缺陷高发地区出生缺陷的发生水平和流行病学特征[J].中华流行病学杂志,2007,28(1):5-9.
[38]陈雪芹,周纯先,姬梅田等.安徽省蚌埠地区0-5岁儿童出生缺陷调查及对策研究[J].中国计划生育学杂志,2008(12):736-738.
[39]任娟,曹缵孙.辅助生殖技术与出生缺陷[J].国外医学计划生育/生殖健康分册,2007,26,(2):79-82.
[40]高韦,沈毅.体外受精-胚胎移植与自然受孕新生儿畸形的Meta分析.中华预防医学杂志,2008,43(6).
[41]薛富波,张文彤,田晓燕.SAS8.2统计应用教程.第一版.北京:北京希望电子出版社,2004,212.
[42]Gardner W.,Mulvey EP.,Shaw EC.Regression analyses of counts and rates:Poisson overdispersed Poisson and negative binomial models.Psychological Bulletin [J],1995,118(3):392-404.
[43]Travier N.,Gridley G.,Blair A.,Dosemeci M.,et al.Cancer incidence among male Swedish veterinarians and other workers of the veterinary industry:a record-linkage study.Cancer Causes & Control[J],2003,14(6):587-593.
[44]Parodi S.,Bottarelli E.Poisson regression model in epidemiology-an introduction[J].Ann.Fac.Medic.Vet.di Parma,2006,26(1):25-44.
[1]沈其君.SAS统计分析.第一版.北京:高等教育出版社,2005:169.
[2]郭志刚.社会统计分析方法-SPSS软件应用.第一版.北京:中国人民大学出版社,1999:.
[3]方积乾.卫生统计学.第五版.北京:人民卫生出版社,2005:362.
[4]Hwang IK.Superiority,Noninferiority and Equivalence trials.北京:新药临床研究生物统计学国际学术研讨会,2003:01.
[5]童峰,陈坤.常见结局事件的前瞻性研究中修正Poisson回归模型的应用[J].中国卫生统计,2006,23(5):410-412.
[6]方积乾.卫生统计学.第五版.北京:人民卫生出版社,2005:52-53.
[7]杨珉,李晓松.医学和公共卫生研究常用多水平统计模型.第一版.北京:北京大学医学出版社,2007:104.
[8]陈峰.医用多元统计分析方法.第二版.北京:中国统计出版社,2007:125
[9]吴喜之主编.统计学:从数据到理论.第一版.北京:中国统计出版社,2004:42-43.
[10]宇传华,余松林,向惠云等.Poisson回归在职业队列研究中的应用[J].中国卫生统计,1996,13(1):6-9.
[11]Samuel k.Encyclopedia of Statistical Sciences,Plackett Family of Distributions to Regression,Wrong.Vol7.New York:John Wiley & Sons,Inc,1986:18-43.
[12]Frome EL.The analysis of rates using Poisson regression models [J].Biometrics,1983:39(3):665.
[13]Frome EL.,Checkoway H.Use of Poisson regression models in estimating incidence rates and ratios[J].American Journal of Epidemiology,1985,121:309-323.
[14]Joshi AY.,Iyer VN.,Hagan JB.,et al.Incidence and Temporal Trends of Primary Immunodeficiency:A Population-Based Cohort Study[J].Mayo Clin.Proc.,2009:84(1):16-22.
[15]Villar E.,Remontet L.,Labeeuw M.,et al.Effect of Age,Gender,and Diabetes on Excess Death in End-Stage Renal Failure[J].J.Am.Soc.Nephrol.,2007:18(7):2125-2134.
[16]沈其君.SAS统计分析.第一版.北京:高等教育出版社,2005:219.
[17]Selmer R.A comparison of Poisson regression model fitted to multiway summary tables and Cox's survival model using data from a blood pressure screening in the city of Bergen,Norway[J].StatMed,1990,9(10):1157-1165.
[18]Breslow NE.,Day NE.Statistical methods in cancer research.VolⅡ.Lyon:International Agency for Research on Cancer,1987:82.
[19]Diabetes Epidemiology Research International Group.Secular trends in the incidence of childhood IDDM in 10 countries[J].Diabetes,1990;39(7):858-864.
[20]Nystrom L,Dahlquist G,REWERS M.,et al.The Swedish Childhood diabetes study:an analysis of the temporal variation in diabetes incidence 1978-1987[J].International Journal Epidemiology,1990,19(1):141-146.
[21]陈峰.现代医学统计方法与Stata应用.第一版.北京:中国统计出版社,2000:197.
[22]陈峰.医学多元统计分析方法.第一版.北京:中国统计出版社,2001:125-128.
[23]张颖,倪宗瓒,姚树祥,等.广义Poisson回归模型及其应用[J].中国卫生统计,2000,17(13):130-133
[24]于浩,石卫,戴胜利.Poisson回归模型的应用[J].江苏预防医学,1996,3:15-16.
[25]刘敏,肖求文,杨纲等.煤矿矿工尘肺与肺癌死CPoisson回归分析[J].中国职业医学,2001,28(6):21-22.
[26]顾海雁,黄萍萍,刘锋等.上海市徐汇区1995-2004年20岁以下人群白血病死亡率Poisson回归分析[J].上海预防医学杂志,2006,18(3):106-108.
[27]TRON ANDERS MOGER,ODD O.AALEN.Regression models for infant mortality data in Norwegian siblings,using a compound Poisson frailty distribution with random scale[J].Biostatistics,2008,9(3),577-591.
[28]易玲,潘晓平,赵梓伶.Poisson回归在脑卒中死亡危险因素筛选中的应用[J].现代预防医学,2008,35(8):1405-1407.
[29]罗艳,栾荣生,朱军等.尿道下裂与环境污染的关系-Poisson回归模型与GM(1,N)模型结合分析[J].现代预防医学,2005,32(1):1-2.
[30]Gardner W.,Mulvey EP.,Shaw EC.Regression analyses of counts and rates:Poisson overdispersed Poisson and negative binomial models.Psychological Bulletin [J],1995,118(3):392-404.
[31]Travier N.,Gridley G.,Blair A.,Dosemeci M.,et al.Cancer incidence among male Swedish veterinarians and other workers of the veterinary industry:a record-linkage study.Cancer Causes & Control[J],2003,14(6):587-593.
[32]Parodi S.,Bottarelli E.Poisson regression model in epidemiology-an introduction[J].Ann.Fac.Medic.Vet.Di Parma,2006,26(1):25-44.
[2]乐杰.妇产科学.6版.北京:人民卫生出版社,2004:384.
[3]Paul Merlob,Onit Sapir,Jaqueline Sulkes,et al.The prevalence of major congenital malformations during two periods of time,1986-1994 and 1995-2002 in newborns conceived by assisted reproduction technology.European Journal of Medical Genetics,2005,48:5-11.
[4]Michele H.,Jennifer J.,Carol B.,et al.The risk of major birth defects after intracytoplasmic sperm injection and in vitro fertilization.The New England Journal of Medicine.2002,346(10):725-730.
[5]Christine K.,Kim M.,bPaul A.,etal.In vitro fertilization is associated with an increase in major birth defects.Fertility and Sterility.2005,84,(5):1308-1315.
[6]沈其君.SAS统计分析.第一版.北京:高等教育出版社,2005:169.
[7]郭志刚.社会统计分析方法-SPSS软件应用.第一版.北京:中国人民大学出版社,1999:.
[8]方积乾.卫生统计学.第五版.北京:人民卫生出版社,2005:362.
[9]Hwang IK.Superiority,Noninferiority and Equivalence trials.北京:新药临床研究生物统计学国际学术研讨会,2003:01.
[10]童峰,陈坤.常见结局事件的前瞻性研究中修正Poisson回归模型的应用[J].中国卫生统计,2006,23(5):410-412.
[11]方积乾.卫生统计学.第五版.北京:人民卫生出版社,2005:52-53.
[12]杨珉,李晓松.医学和公共卫生研究常用多水平统计模型.第一版.北京:北京大学医学出版社,2007:104.
[13]陈峰.医用多元统计分析方法.第二版.北京:中国统计出版社,2007:125.
[14]吴喜之主编.统计学:从数据到理论.第一版.北京:中国统计出版社, 2004:42-43.
[15]宇传华,余松林,向惠云等.Poisson回归在职业队列研究中的应用[J].中国卫生统计,1996,13(1):6-9.
[16]Samuel k.Encyclopedia of Statistical Sciences,Plackett Family of Distributions to Regression,Wrong.Vol7.New York:John Wiley & Sons,Inc,1986:18-43.
[17]Frome EL.The analysis of rates using Poisson regression models [J].Biometrics,1983:39(3):665.
[18]Frome EL.,Checkoway H.Use of Poisson regression models in estimating incidence rates and ratios[J].American Journal of Epidemiology,1985,121:309-323.
[19]Joshi AY.,Iyer VN.,Hagan JB.,et al.Incidence and Temporal Trends of Primary Immunodeficiency:A Population-Based Cohort Study[J].Mayo Clin.Proc.,2009:84(1):16-22.
[20]Villar E.,Remontet L.,Labeeuw M.,et al.Effect of Age,Gender,and Diabetes on Excess Death in End-Stage Renal Failure[J].J.Am.Soc.Nephrol,2007:18(7):2125-2134.
[21]沈其君.SAS统计分析.第一版.北京:高等教育出版社,2005:219.
[22]Selmer R.A comparison of Poisson regression model fitted to multiway summary tables and Cox's survival model using data from a blood pressure screening in the city of Bergen,Norway[J].StatMed,1990,9(10):1157-1165.
[23]Breslow NE.,Day NE.Statistical methods in cancer research.VolⅡ.Lyon:International Agency for Research on Cancer,1987:82.
[24]Diabetes Epidemiology Research International Group.Secular trends in the incidence of childhood IDDM in 10 countries[J].Diabetes,1990;39(7):858-864.
[25]Nystrom L,Dahlquist G,REWERS M.,et al.The Swedish Childhood diabetes study:an analysis of the temporal variation in diabetes incidence 1978-1987[J].International Journal Epidemiology,1990,19(1):141-146.
[26]陈峰.现代医学统计方法与Stata应用.第一版.北京:中国统计出版社,2000:197.
[27]陈峰.医学多元统计分析方法.第一版.北京:中国统计出版社,2001:125-128.
[28]张颖,倪宗瓒,姚树祥,等.广义Poisson回归模型及其应用[J].中国卫生统计,2000,17(13):130-133.
[29]李克,俞顺章.Poisson模型与队列研究-原理及应用[J].中国卫生统计,1997,14(1):46-48.
[30]Stefano Parodi,Ezio Bottarelli.Poisson regression modelin Epidemiology-an Introduction[J].Ann.Fac.Medic.Vet.DiParma,2006,(26):25-44.
[31]Michele H.,Jennifer J.,Carol B.,et al.The risk of major birth defects after intracytoplasmic sperm injection and in vitro fertilization.The New England Journal of Medicine.2002,346(10):725-730.
[32]Bergh T.,Ericson A.,Hillensj(o|¨) T.,et al.Deliveries and children born after in-vitro fertilisation in Sweden 1982-95:a retrospective cohort study.The Lancet,1999,354:1579-1585.
[33]修晓燕,张荣莲,陈起燕.出生缺陷的研究进展[J].中国妇幼健康研究,2007,18(5):400-402.
[34]郑晓瑛,宋新明,陈功等.中国出生缺陷高发地区出生缺陷的发生水平和流行病学特征[J].中华流行病学杂志,2007,28(1):5-9.
[35]Anthony S.,Buitendijk SE.,Dorrepaal CA.,et al.Congenital malformations in 4224 children conceived after IVF.Hum Reprod,2002,17(8):2089-2095
[36]胡荣华,杜玉开.武汉市围产儿先天畸形流行状况及影响因素研究[J].医学与社会,2005,18(5):7-11.
[37].郑晓英,宋新明,陈功等.中国出生缺陷高发地区出生缺陷的发生水平和流行病学特征[J].中华流行病学杂志,2007,28(1):5-9.
[38]陈雪芹,周纯先,姬梅田等.安徽省蚌埠地区0-5岁儿童出生缺陷调查及对策研究[J].中国计划生育学杂志,2008(12):736-738.
[39]任娟,曹缵孙.辅助生殖技术与出生缺陷[J].国外医学计划生育/生殖健康分册,2007,26,(2):79-82.
[40]高韦,沈毅.体外受精-胚胎移植与自然受孕新生儿畸形的Meta分析.中华预防医学杂志,2008,43(6).
[41]薛富波,张文彤,田晓燕.SAS8.2统计应用教程.第一版.北京:北京希望电子出版社,2004,212.
[42]Gardner W.,Mulvey EP.,Shaw EC.Regression analyses of counts and rates:Poisson overdispersed Poisson and negative binomial models.Psychological Bulletin [J],1995,118(3):392-404.
[43]Travier N.,Gridley G.,Blair A.,Dosemeci M.,et al.Cancer incidence among male Swedish veterinarians and other workers of the veterinary industry:a record-linkage study.Cancer Causes & Control[J],2003,14(6):587-593.
[44]Parodi S.,Bottarelli E.Poisson regression model in epidemiology-an introduction[J].Ann.Fac.Medic.Vet.di Parma,2006,26(1):25-44.
[1]沈其君.SAS统计分析.第一版.北京:高等教育出版社,2005:169.
[2]郭志刚.社会统计分析方法-SPSS软件应用.第一版.北京:中国人民大学出版社,1999:.
[3]方积乾.卫生统计学.第五版.北京:人民卫生出版社,2005:362.
[4]Hwang IK.Superiority,Noninferiority and Equivalence trials.北京:新药临床研究生物统计学国际学术研讨会,2003:01.
[5]童峰,陈坤.常见结局事件的前瞻性研究中修正Poisson回归模型的应用[J].中国卫生统计,2006,23(5):410-412.
[6]方积乾.卫生统计学.第五版.北京:人民卫生出版社,2005:52-53.
[7]杨珉,李晓松.医学和公共卫生研究常用多水平统计模型.第一版.北京:北京大学医学出版社,2007:104.
[8]陈峰.医用多元统计分析方法.第二版.北京:中国统计出版社,2007:125
[9]吴喜之主编.统计学:从数据到理论.第一版.北京:中国统计出版社,2004:42-43.
[10]宇传华,余松林,向惠云等.Poisson回归在职业队列研究中的应用[J].中国卫生统计,1996,13(1):6-9.
[11]Samuel k.Encyclopedia of Statistical Sciences,Plackett Family of Distributions to Regression,Wrong.Vol7.New York:John Wiley & Sons,Inc,1986:18-43.
[12]Frome EL.The analysis of rates using Poisson regression models [J].Biometrics,1983:39(3):665.
[13]Frome EL.,Checkoway H.Use of Poisson regression models in estimating incidence rates and ratios[J].American Journal of Epidemiology,1985,121:309-323.
[14]Joshi AY.,Iyer VN.,Hagan JB.,et al.Incidence and Temporal Trends of Primary Immunodeficiency:A Population-Based Cohort Study[J].Mayo Clin.Proc.,2009:84(1):16-22.
[15]Villar E.,Remontet L.,Labeeuw M.,et al.Effect of Age,Gender,and Diabetes on Excess Death in End-Stage Renal Failure[J].J.Am.Soc.Nephrol.,2007:18(7):2125-2134.
[16]沈其君.SAS统计分析.第一版.北京:高等教育出版社,2005:219.
[17]Selmer R.A comparison of Poisson regression model fitted to multiway summary tables and Cox's survival model using data from a blood pressure screening in the city of Bergen,Norway[J].StatMed,1990,9(10):1157-1165.
[18]Breslow NE.,Day NE.Statistical methods in cancer research.VolⅡ.Lyon:International Agency for Research on Cancer,1987:82.
[19]Diabetes Epidemiology Research International Group.Secular trends in the incidence of childhood IDDM in 10 countries[J].Diabetes,1990;39(7):858-864.
[20]Nystrom L,Dahlquist G,REWERS M.,et al.The Swedish Childhood diabetes study:an analysis of the temporal variation in diabetes incidence 1978-1987[J].International Journal Epidemiology,1990,19(1):141-146.
[21]陈峰.现代医学统计方法与Stata应用.第一版.北京:中国统计出版社,2000:197.
[22]陈峰.医学多元统计分析方法.第一版.北京:中国统计出版社,2001:125-128.
[23]张颖,倪宗瓒,姚树祥,等.广义Poisson回归模型及其应用[J].中国卫生统计,2000,17(13):130-133
[24]于浩,石卫,戴胜利.Poisson回归模型的应用[J].江苏预防医学,1996,3:15-16.
[25]刘敏,肖求文,杨纲等.煤矿矿工尘肺与肺癌死CPoisson回归分析[J].中国职业医学,2001,28(6):21-22.
[26]顾海雁,黄萍萍,刘锋等.上海市徐汇区1995-2004年20岁以下人群白血病死亡率Poisson回归分析[J].上海预防医学杂志,2006,18(3):106-108.
[27]TRON ANDERS MOGER,ODD O.AALEN.Regression models for infant mortality data in Norwegian siblings,using a compound Poisson frailty distribution with random scale[J].Biostatistics,2008,9(3),577-591.
[28]易玲,潘晓平,赵梓伶.Poisson回归在脑卒中死亡危险因素筛选中的应用[J].现代预防医学,2008,35(8):1405-1407.
[29]罗艳,栾荣生,朱军等.尿道下裂与环境污染的关系-Poisson回归模型与GM(1,N)模型结合分析[J].现代预防医学,2005,32(1):1-2.
[30]Gardner W.,Mulvey EP.,Shaw EC.Regression analyses of counts and rates:Poisson overdispersed Poisson and negative binomial models.Psychological Bulletin [J],1995,118(3):392-404.
[31]Travier N.,Gridley G.,Blair A.,Dosemeci M.,et al.Cancer incidence among male Swedish veterinarians and other workers of the veterinary industry:a record-linkage study.Cancer Causes & Control[J],2003,14(6):587-593.
[32]Parodi S.,Bottarelli E.Poisson regression model in epidemiology-an introduction[J].Ann.Fac.Medic.Vet.Di Parma,2006,26(1):25-44.
© 2004-2018 中国地质图书馆版权所有 京ICP备05064691号 京公网安备11010802017129号 地址：北京市海淀区学院路29号 邮编：100083 电话：办公室：(+86 10)66554848；文献借阅、咨询服务、科技查新：66554700 |