预测原发性青光眼发生风险的分类回归树及列线图模型的初步建立及评估

发布时间：2018-06-29 18:31

本文选题：原发性闭角型青光眼 + 原发性开角型青光眼　；参考：《复旦大学》2012年博士论文

【摘要】：目的：通过对PACG、POAG、正常对照组的临床资料进行单因素和Logistic多因素回归分析,以明确PACG、POAG的相关独立危险因素。方法：在2009.12-2011.11期间,共筛出在复旦大学附属眼耳鼻喉科医院住院的PACG患者200例、POAG患者100例、正常对照组200例。通过病史回顾性收集这些患者的相关临床资料变量,个人一般情况包括：年龄、性别、是否有青光眼家族史、是否有高血压史、是否有糖尿病史、是否有全身其他病史。临床相关资料包括：矫正视力、屈光情况、眼内压、前房深度、杯盘比、中央角膜厚度、角膜曲率、眼轴、晶状体厚度、晶状体位置、相对晶状体位置。同时,比较这些因素在PACG组、POAG组和正常对照组三组上的分布有无差异性,并进行相关的Logistic多因素回归分析。结果：在单因素分析结果的基础上,对相关变量进行Logistic多因素回归分析后发现,最终入组预测PACG模型的变量有糖尿病、杯/盘比、眼轴、角膜曲率、中央角膜厚度,其中除角膜曲率外,其他4个变量是预测PACG的独立危险因素(p0.05)。整个Logistic多因素回归模型的C-index为0.956。最终入组POAG模型的变量有性别、高度近视、杯/盘比、眼轴,其中除性别外,其他3个变量是预测POAG的独立危险因素(p0.05)。整个Logistic多因素回归模型的C-index为0.975。结论：糖尿病、杯／盘比、眼轴、中央角膜厚度是PACG的独立危险因素,而高度近视、杯／盘比、眼轴是POAG的独立危险因素。目的：建立并验证预测PACG发生风险的CART及列线图模型,并通过与其他模型或标准比较,以明确最佳的模型或标准,从而根据该最佳模型或标准以减少不必要的干预措施。方法：CART模型的建立及评估：对相关变量进行CART统计分析以建立用于预测PACG发生风险的CART模型,并采用10倍交叉验证方法对此CART模型进行内部验证以减少过度拟合偏倚。列线图模型的建立及评估：根据PACG的Logistic多因素回归分析确定模型入组变量,并依据相关变量的回归系数画出相应的列线图模型,并采用Bootstrap自抽样方法对列线图模型进行内部验证以减少过度拟合偏倚,同时评价列线图模型预测PACG发生风险的符合度。最后,采用AUC. C-index、DCA统计方法比较列线图模型、CART模型和前房深度指标在预测PACG的准确性及临床应用价值上的优劣性。结果：CART模型4个节点上的PACG发生率分别为99.3%、92.9%、87.5%及8.8%,并且在经过内部验证后得到的C-index为0.965,表现出较好的预测准确性。PACG的列线图模型输入变量包含糖尿病、杯/盘比、眼轴、角膜曲率、中央角膜厚度,在经过内部验证后C-index为0.953。根据AUC、C-index、DCA统计方法,CART及列线图模型均优于前房深度指标,而CART及列线图模型按阈值概率范围的不同,在临床应用价值上各有其优势。结论：在预测PACG发生风险的准确性及临床应用价值上,CART及列线图模型均优于前房深度指标。在临床应用价值上,可以在青光眼筛查中联合应用CART模型、列线图模型和前房深度指标,不以单一模型筛选。目的：建立并验证预测POAG发生风险的CART及列线图模型,并通过与其他模型或标准比较,以明确最佳的模型或标准,从而根据该最佳模型或标准以减少不必要的干预措施。方法：CART模型的建立及评估：对相关变量进行CART统计分析以建立用于预测POAG发生风险的CART模型,并采用10倍交叉验证方法对此回归树模型进行内部验证以减少过度拟合偏倚。列线图模型的建立及评估：根据POAG的Logistic多因素回归分析确定模型入组变量,并依据相关变量的回归系数画出相应的列线图模型,并采用Bootstrap自抽样方法对列线图模型进行内部验证以减少过度拟合偏倚,同时评价列线图模型预测POAG发生风险的合度。最后,采用C-index、DCA统计方法比较两模型在预测POAG的准确性及临床应用价值上的优劣性。结果：CART模型2个节点上的PACG发生率分别为98.9%和5.2%,并且在经过内部验证后得到的C-index为0.973,表现出较好的预测准确性。POAG列线图模型输入变量包含性别、高度近视、杯/盘比、眼轴,在经过内部验证后C-index为0.970。DCA统计结果显示,列线图模型在临床应用价值上优于CART模型。结论：CART模型及列线图模型在预测POAG发生风险的准确性上相似,但在临床应用价值上,列线图模型优于CART模型。
[Abstract]:Objective: to analyze the clinical data of PACG, POAG and normal controls by single factor and Logistic multivariate regression analysis in order to determine the independent risk factors of PACG and POAG.
Methods: during the period of 2009.12-2011.11, 200 cases of PACG patients hospitalized in Otolaryngological Hospital Affiliated to Fudan University, 100 cases of POAG and 200 cases of normal control group were collected, and the related clinical data of these patients were collected through the history of disease, and the general situation included: age, sex, family history of glaucoma, whether there were high History of blood pressure, whether there is a history of diabetes, or whether there are other systemic history. Clinical data include: corrected visual acuity, refraction, intraocular pressure, anterior chamber depth, cup disc ratio, central corneal thickness, corneal curvature, eye axis, lens thickness, lens position, relative crystalline body position. At the same time, these factors are compared in group PACG, POAG and normal control. There was no difference in the distribution of the three groups, and Logistic regression analysis was performed.
Results: on the basis of the results of the single factor analysis, after Logistic multiple regression analysis of the related variables, the final group to predict the variables of the PACG model were diabetes, cup / disc ratio, eye axis, corneal curvature, central corneal thickness, and the other 4 variables were independent risk factors (P0.05) for predicting PACG except the corneal curvature. The whole Logi was the whole Logi. The C-index of stic multi factor regression model was 0.956. final entry group POAG model variables: sex, high myopia, cup / disc ratio, eye axis, except for sex, the other 3 variables were independent risk factors (P0.05) for predicting POAG. The C-index of the whole Logistic multifactor regression model was 0.975.
Conclusion: diabetes, cup / disc ratio, axial length and central corneal thickness are independent risk factors for PACG, while high myopia, cup / disc ratio and axial length are independent risk factors for POAG.
Objective: to establish and verify the CART and line map models for predicting the risk of PACG occurrence, and to identify the best models or standards by comparing with other models or standards to reduce unnecessary interventions based on the best model or standard.
Method: the establishment and evaluation of the CART model: CART statistical analysis of related variables to establish the CART model for predicting the risk of PACG occurrence, and the 10 times cross validation method is used to verify the CART model to reduce the overfitting bias. The establishment and evaluation of the column graph model: according to the Logistic multiple factor regression of PACG Analyze the model into the group variables, and draw the corresponding line graph model according to the regression coefficient of the related variables, and use the Bootstrap self sampling method to verify the model of the line graph in order to reduce the overfitting bias, and evaluate the line graph model to predict the coincidence of the risk of the occurrence of PACG. Finally, the AUC. C-index, DCA statistical method is used. Compared with the nomogram model, the CART model and the anterior chamber depth index, the accuracy and clinical application value of PACG were better.
Results: the incidence of PACG on the 4 nodes of the CART model was 99.3%, 92.9%, 87.5% and 8.8% respectively, and the C-index was 0.965 after the internal verification. The good predictive accuracy of the line graph model input variables included diabetes, cup / disc ratio, eye axis, corneal curvature, central corneal thickness, and C-in after internal verification. DEX for 0.953. according to AUC, C-index, DCA statistical methods, CART and line graph model are better than the front room depth index, and CART and line graph model are different in the range of threshold probability, each has its advantages in clinical application.
Conclusion: in predicting the accuracy and clinical value of the risk of PACG, both the CART and the line map model are superior to the anterior chamber depth index. In the clinical application, the CART model, the line map model and the anterior chamber depth index can be applied in the glaucoma screening.
Objective: to establish and verify the CART and line map models for predicting the risk of POAG occurrence, and to identify the best models or standards by comparing with other models or standards to reduce unnecessary interventions based on the best model or standard.
Methods: the establishment and evaluation of the CART model: CART statistical analysis of related variables to establish the CART model for predicting the risk of POAG occurrence, and the 10 times cross validation method is used to verify the regression tree model to reduce the overfitting bias. The establishment and evaluation of the line map model: Based on the Logistic multiple factor regression of POAG Analysis determines the model into the group variables, and draws the corresponding line graph model according to the regression coefficient of the related variables, and uses the Bootstrap self sampling method to verify the model of the line graph in order to reduce the overfitting bias. At the same time, we evaluate the alignment of the line graph model to predict the risk of POAG occurrence. Finally, the C-index and DCA statistical methods are used to compare two The model is used to predict the accuracy and clinical value of POAG.
Results: the incidence of PACG on the 2 nodes of the CART model was 98.9% and 5.2% respectively, and the C-index obtained after the internal verification was 0.973, which showed good prediction accuracy. The input variables of the.POAG line graph model included sex, high myopia, cup / disk ratio, eye axis, and C-index for 0.970.DCA statistics after the internal verification. The line model is superior to the CART model in clinical application.
Conclusion: the CART model and nomogram model are similar in predicting the accuracy of POAG risk, but the nomogram model is better than the CART model in clinical application.
【学位授予单位】：复旦大学
【学位级别】：博士
【学位授予年份】：2012
【分类号】：R775

【引证文献】