腹膜透析相关腹膜炎危险因素研究的统计模型筛选

目的筛选适用于腹膜透析相关腹膜炎危险因素研究的统计方法,为腹膜炎的临床研究提供新的思路。 方法选择九种统计模型进行对比分析,包括：不含时间依存性协变量(时依协变量)的Logistic回归、泊松回归、负二项回归、COX回归、AG模型、PWP-CP模型6个模型,含时间依存性协变量的COX回归、AG模型、PWP-CP模型3个模型。统计数据来源于解放军总医院第一医学中心腹膜透析中心2013年1月至2016年12月开始腹膜透析治疗的终末期肾脏病患者,数据收集至2018年12月。分析腹膜炎的临床特点,对比各统计模型拟合情况和统计结果,筛选统计模型。 结果从腹膜炎临床数据构成来看,作为因变量,腹膜炎事件由"发生时间"、"发生次数"、"发生顺序"三个要素构成,作为自变量,大部分协变量是随时间变化的,三维的因变量和变化的自变量共同构成了复杂的临床数据。从"腹膜炎三要素"纳入情况看,只有PWP-CP模型能够全部纳入,最大限度保证了数据的完整性。从统计模型拟合程度看,泊松模型存在过离散,不适用于本研究数据;使用时依协变量的模型较使用基线数据的模型拟合好。从统计结果看,含时间依存性协变量的模型避免了单纯使用基线数据造成的偏差,使统计结果更接近于真实情况。 结论在腹膜炎危险因素的研究中,只有尽可能地保证数据信息的完整性和准确性,才能得到更真实的研究结果。本研究通过剖析腹膜炎的数据构成,对比各模型的统计结果,发现含时依协变量的PWP-CP模型更适用于腹膜炎危险因素的研究。

Screening of statistical models for risk factors of peritoneal dialysis-related peritonitis

ObjectiveTo screen statistical methods that are suitable for studying risk factors of peritoneal dialysis-related peritonitis so as to provide new ideas for clinical research of peritonitis. MethodsNine statistical models were selected for comparative analysis, including six models without time-dependent covariates: logistic regression, Poisson regression, negative binomial regression, COX regression, AG model, and PWP-CP model; and three models with time-dependent covariates: COX regression, AG model, and PWP-CP model. The statistical data were obtained from patients with end-stage renal disease who started peritoneal dialysis treatment from January 2013 to December 2016 in the Peritoneal Dialysis Center of The First Medical Center, Chinese PLA General Hospital. The data were collected until December 2018. Through the analysis of the clinical characteristics of peritonitis, and the comparison between the statistical models and results, the suitable statistical model was selected. ResultsFrom the clinical data of peritonitis, as a dependent variable, the peritonitis event consists of three elements: "time of occurrence" , "number of occurrences" , and "order of occurrence" . As independent variables, most of the covariates were time-dependent. The three-dimensional dependent variables and the varying independent variables together constituted complex clinical data. In terms of data integrity, only the PWP-CP model could include the "three elements of peritonitis" . In terms of statistical models fitting, the Poisson model was too discrete to enter the data of this study. The model based on the time-dependent covariates was better to fit than the model using the baseline data. In terms of statistical results, the model with time-dependent covariates avoided the bias caused by the simple use of baseline data, making the statistical results closer to the real situation. ConclusionIn the study of risk factors of peritonitis, more realistic research results could be obtained only by ensuring the integrity and accuracy of the data as much as possible. In this study, by analyzing the data of peritonitis and comparing the statistical results of each model, it was found that the PWP-CP model with time-dependent covariates was more suitable for studying the risk factors of peritonitis.