摘要
目的：目前，国内外在进行临床疾病以及疾病类型的计量诊断中，越来越多的运用了多元分析的统计学方法，但在特定疾病类型的计量诊断中，大多只运用了一种多元分析方法，本次研究应用三种多元分析方法对卵巢囊肿进行早期鉴别诊断,并对三种多元分析方法在卵巢囊肿鉴别诊断上的应用实施程序开发，以提高疾病鉴别诊断的准确性和效率。
方法：1、最大似然法：根据卵巢囊肿临床上的多项检查指标，利用最大似然法求出各指标征象的条件概率及其相应的评分，据此建立卵巢囊肿早期鉴别的定量诊断表。2、信息分析：利用信息分析法的原理以及信息分析法在临床疾病诊断价值上的应用理论，求出两种卵巢囊肿类型的信源熵和各症候传递的总信息量，据此求出各症候的贡献率，并根据各症候贡献率的大小剔除贡献率相对较小的症候，然后求出所保留症候表现的信息熵并据此建立卵巢囊肿早期鉴别的定量诊断表。3、logistic回归分析：首先对logistic回归模型进行共线性诊断,然后选用主成分logistic回归进行分析。经主成分logistic回归分析后，筛选出对鉴别卵巢囊肿有意义的检查指标并建立logistic回归模型，在对logistic回归模型拟合情况分析后进一步在logistic回归模型的基础上建立用于logistic判别分析的logistic线性回归方程以实现卵巢囊肿的早期鉴别。4、软件开发：软件开发应用美国
Borland International Inc出品的Delphi7.0。程序开发过程主要包括鉴别诊断用表的数据库建立、应用程序界面的设计与建立、鉴别诊断功能的程序编写和应用程序的发布等步骤。
结果：1、最大似然法：回顾性判别符合率为83.57%，赘生性判别符合率为85.03%，非赘生判别符合率为82.42%；前瞻性判别合率为82.43%，赘生性判别符合率为80%，非赘生判别符合率为84.09%。2、信息分析：回顾性判别符合率为89.05%，赘生性判别符合率为86.39%，非赘生判别符合率为90.48%；前瞻性判别合率为87.84%，赘生性判别符合率为86.67%，非赘生判别符合率为88.64%。3、logistic回归分析：经主成分logistic回归分析得到logistic线性回归方程为：=0.228年龄+0.698个人史-0.950痛经史+0.405产次+0.883囊肿表面+1.265内部回声-1.287压迫症状+1.244壁厚-0.734囊肿房数-0.423流产次数+0.906囊肿大小-3.977；ROC曲线下面积为0.868，标准误等于0.018，p=0.000,p<0.05,表明该模型预报能力中等；回顾性判别符合率为87.86%，赘生性判别符合率为85.71%，非赘生判别符合率为89.01%；前瞻性判别合率为85.14%，赘生性判别符合率为83.33%，非赘生判别符合率为86.36%。4、一致性检验：最大似然法和logistic回归分析的kappa指数k=0.699, p=0.000,p<0.01，说明两种方法存在一致性，又因为0.4.75，说明两者的一致性好［1］。5、
软件开发：开发完成的三种多元分析软件所得结果与按照三种方法的原理逐步计算所得结果完全一致，在新病例的鉴别诊断上，只需在软件中选取其相应的征象，即可快速的做出鉴别诊断从而大大提高了鉴别诊断的效率。
结论：1、分别应用最大似然法、信息分析和logistic回归分析三种多元分析方法实现了卵巢囊肿的早期鉴别诊断，并且三种多元分析方法的判别效果的一致性较好。2、开发完成的三种多元分析软件所得结果与按照三种方法的原理逐步计算所得结果完全一致，开发的软件可以纳入医院门诊作为卵巢囊肿早期鉴别的辅助诊断。
Objective: At the moment, the multivariate analysis statistics means is more and more used in the quantitative diagnosis of clinical disease and its type home and abroad, Yet in the quantitative diagnosis of specific disease type, only one kind of multivariate analysis means was mostly used .In this research, in order to improve the accuracy and effectiveness of discrimination diagnosis of disease, three kinds of multivariate analysis means were used to discriminate early stage ovarian cyst and programs were developed for the application of the three kinds of multivariate analysis means that were used in the ovarian cyst early stage discrimination.
Methods: 1.the method of maximum likelihood：with the method of maximum likelihood, the conditional probability of every index symptom and its score were computed according to many clinical examination indexes of ovarian cyst, in view of which to establish the quantitative diagnosis table of ovarian cyst early stage discrimination.2.Information analysis ：Based on the principle of information analysis and the theorem of information analysis’ application on the value of clinical disease diagnosis, the information source entropy of the two types of
ovarian cyst and the total information amount of every symptom were computed, in view of which the contribution rate of every symptom were computed and the symptoms which contribution rate were relatively smaller were removed, then the information entropy of symptom reserved were computed, in view of which the quantitative diagnosis table of ovarian cyst early stage discrimination were established.3. Logistic regression analysis：Firstly, collinearity diagnosis was made on logistic regression model, if there was collinearity between the independent variable; logistic regression based on principal component analysis was selected. After logistic regression based on principal component analysis, the significant examination indexes for ovarian cyst discrimination was screened and the logistic regression equation was established, after analysis of the fit of logistic regression model, based on which establishing the logistic linearity regression equation which used on logistic discrimination analysis was established to realize the ovarian cyst early stage discrimination. 4 Software development ： Software development applies Delphi7.0 that was produced by American Borland International Inc.The process of program development mainly was consisted of the establishment of DB on discrimination table, the design and establishment of application program interface, the programming the function of discrimination and the open of application programmed.
Results:1.the method of maximum likelihood :The retrospective discrimination coincidence rate is 83.57%,
redundant discrimination coincidence rate is 85.03%, non-redundant discrimination coincidence rate is 82.42%;the prospective discrimination coincidence rate is 82.43%,redundant discrimination coincidence rate is 80%,non-redundant discrimination coincidence rate is 84.09%.2.Information analysis:the retrospective discrimination coincidence rate is 89.05%,redundant discrimination coincidence rate is 86.39 %,non-redundant discrimination coincidence rate is 90.48%; the prospective discrimination coincidence rate is 87.84%,redundant discrimination coincidence rate is 86.67%,non-redundant discrimination coincidence rate is 88.64%.3.Logistic regression analysis：After logistic regression based on principal component analysis, the logistic linearity regression equation is =0.228 age+0.698 personal history-0.950 menorrhalgia history+0.405 parity+0.883 cyst surface+1.265 inside echo-1.287 compression symptom+1.244 thick of wall-0.734 cyst chamber-0.423 abortion number+0.906 cyst size-3.977. The area under ROC curve is 0.868, the std.error is 0.018, p=0.000,p<0.05,which show that the model has the moderate forecast ability. The retrospective discrimination coincidence rate is 87.86%, redundant discrimination coincidence rate is 85.71%, non-redundant discrimination coincidence rate i
引文
1 方积乾 主编。医学统计学与电脑实验，第二版，上海：上海科技出版社，2001，245，443
陈雄飞，董晓梅，汪宁等。多因子共线性的主成分logistic回归分析，中国卫生统计，2003 20（4）：213～214
陈忠年主编，杜心谷副主编。妇产科病理学，第一版，上海：上海科学技术出版社，1982，151～152
张文彤主编。SPSS11统计分析教程（高级篇），第一版，北京：北京希望电子出版社，2002，83，92，179
孙振球 主编。医学统计学，第一版，北京：人民卫生出版社，2002，259，297
刘守君。卵巢非赘生囊肿的超声诊断，实用妇科与产科杂志，1991，7(3)：122
钟显玉。赘生性卵巢囊肿声像图与病例对照分析，中国医刊，1999，34(8)：41
郭祖超主编。医用数理统计方法，第一版，北京：人民卫生出版社，1988，464
周怀梧 主编。医用生物数学，第一版，北京：人民卫生出版社，1996，85～86，89
赵宇东、肖峰、张扬等。多元logistic回归的共线性分析，中国卫生统计，2003，17（5）： 259～261
余松林 主编。医学统计学，第一版，北京：人民卫生出版社，2002，200，206，209～210,303
朱军 著。线性模型分析原理，第一版，北京：科学出版
社，1999，49
刘泽田、袁星 主编。线性代数，第一版，北京：学苑出版社，1999，166～167
陈彬，苏祖兰，罗祖明等。logistic判别分析及其在医学中的应用，中国卫生统计，1991；8：5
王炼，刘正明，陈彬等。甲状腺滤泡性肿瘤logistic判别分析及预后分析，诊断病理学杂志，1994，1(3)：152
鲁宗相 编著。Delphi5入门与提高，第一版，北京：清华大学出版社，2002，1，382
吴小前 马亮 等编著。Delphi5编程基础，第一版，北京：清华大学出版社，2000，1
陈豫龙 何旭洪 编著。Delphi6数据库系统开发实例导航，第一版，北京：人民邮电出版社，2002，1,66～67
飞思科技产品研发中心 编著。Delphi6数据库开发，第一版，北京：电子工业出版社，2002，1,95,6～9,204～205,215～216
伍俊良 编写。Delphi6实例编程50讲，第一版，北京：北京电子出版社，2002，217～221，275
胡良平 主编。现代统计学与SAS应用，第一版，北京：军事医学科学出版社，2000，260
刘荷一，陈文雪，刘一晨。多普勒超声诊断卵泡囊肿的演变过程，天津医药，2000，28(3):138～139
23沈其君 主编。SAS统计分析，第一版，南京：东南大学出版社，2001，167