多状态Markov模型在糖尿病纵向数据转归研究中的应用

发布时间：2018-06-23 04:18

本文选题：多状态Markov模型 + 状态转移　；参考：《重庆医科大学》2017年硕士论文

【摘要】：目的本研究将多状态“疾病-预后”Markov模型应用到2型糖尿病患者慢性并发症罹患个数的动态转归研究中,探讨状态间传递规律,估量状态间的转移概率、转移危险及滞留时间,并将其与Logistic回归结合用以寻找影响状态改变的可能因素,从而为2型糖尿病相关慢性并发症的防治提供科学依据。方法回顾性地收集重庆市3所三甲医院2011年1月至2015年5月期间出院,且出院诊断包括2型糖尿病及其慢性并发症的患者数据。根据慢性并发症的合并数量,本研究将合并不同数量慢性并发症的2型糖尿病患者分为5个不同的状态,并拟合一个连续时间、离散状态的齐次Markov过程。然后引入R3.3.2软件中的msm软件包进行多状态Markov模型的分析,并估计5状态之间的传递规律、转移风险、转移概率及滞留时间。与此同时,根据是否发生状态转移来筛选Logistic回归的结局变量,并在SAS9.2软件中进行简单统计分析及Logistic回归分析。结果从msm软件包的plot图和Pearson型拟合优度检验的结果可知,多状态Markov模型拟合效果良好。多状态Markov模型结果提示患者在状态1,状态2,状态3和状态4的总滞留时间分别为3.19月,14.51月23.87月和15.21月。转移强度矩阵显示状态2转化为状态3的转移强度是转化为状态1的1.5倍,而状态3转化为状态4转移强度是转化为状态2的接近10倍,而状态4转移到状态3的转移强度约为转移到状态5的242倍。转移概率矩阵显示经过足够长的时间,处于状态1的个体将向状态2转移,状态2的个体将向状态3转移,状态3的个体将向状态4转移,状态4的个体也将向状态3转移。将单因素Logistic回归中有影响的变量纳入多因素逐步Logistic回归中分析,结果显示:高血压、空腹血糖、尿素、尿微量白蛋白和高密度脂蛋白对状态1→状态2的转移有影响;年龄、空腹血糖、空腹胰岛素、低密度脂蛋白、总胆固醇、载脂蛋白A1和尿素对状态2→状态1的转移有影响;空腹血糖、空腹胰岛素、高密度脂蛋白、甘油三酯和游离脂肪酸对状态2→状态3的转移有影响;空腹血糖、空腹胰岛素、高密度脂蛋白、甘油三酯和游离脂肪酸对状态3→状态2的转移有影响;年龄、高密度脂蛋白、甘油三酯、肌酐和游离脂肪酸对状态3→状态4的转移有影响;甘油三酯、总胆固醇、肌酐和空腹胰岛素对状态4→状态3对的转移有影响;空腹血糖、空腹胰岛素和甘油三酯对状态4→状态5的转移有影响。多因素多状态Markov模型分析结果显示高血压、糖化血红蛋白、游离脂肪酸、脂蛋白a、尿白蛋白/尿肌酐比值和尿微量白蛋白对状态1→状态2的转移有影响;高血压、尿白蛋白/尿肌酐比值、年龄、空腹胰岛素、低密度脂蛋白、载脂蛋白A1对状态2→状态1的转移有影响;高血压、尿白蛋白/尿肌酐比值、糖化血红蛋白、游离脂肪酸、脂蛋白a、尿微量白蛋白对状态2→状态3的转移有影响;高血压、游离脂肪酸、载脂蛋白A1、高密度脂蛋白、肌酐、尿素对状态3→状态2的转移有影响;高血压、游离脂肪酸、载脂蛋白A1、入院时情况对状态3→状态4的转移有影响;游离脂肪酸、肌酐、尿微量白蛋白、空腹胰岛素、甘油三酯对状态4→状态3的转移有影响;游离脂肪酸、肌酐、甘油三酯、低密度脂蛋白、空腹血糖对状态4→状态5的转移有影响。结论多状态Markov模型和传统Logistic回归模型均表明糖尿病慢性并发症合并数量受到年龄、入院时情况、血糖、血脂、血压和肾功能损害等指标的影响,但不同变量对不同状态转移的影响程度有所差异。多状态Markov模型对多结局事件的分析是基于全局的视角,并且考虑了时间对目标结局出现的影响,因此相对于传统Logistic回归而言,其结果更为准确、科学。故而,多状态Markov模型完全可以作为传统Logistic回归模型的有益补充。同时,转移强度矩阵和转移概率矩阵的研究结果表明,各状态患者的病情均有进一步加深的趋势,且患者在状态3停留的时间较长,提示临床上完全可利用该时间窗来逆转疾病的进展。
[Abstract]:Objective the purpose of this study was to apply the multi state "disease and prognosis" Markov model to the study of the number of chronic complications of type 2 diabetic patients, to explore the transfer rules between States, to estimate the transfer probability between States, to transfer the risk and the time of detention, and to combine it with Logistic regression in order to find the possible cause of the change of state. This provides a scientific basis for the prevention and treatment of chronic complications related to type 2 diabetes. Methods 3 Three A hospitals in Chongqing were collected from January 2011 to May 2015, and the discharge diagnosis included data of patients with type 2 diabetes and its chronic complications. According to the number of chronic complications, the study would be combined with different numbers. The patients with chronic complications of type 2 diabetes are divided into 5 different states and fit a continuous time, the homogeneous Markov process in a discrete state. Then the MSM software package in the R3.3.2 software is introduced to analyze the multi state Markov model, and the transfer of the 5 states, the transfer of risk, the transfer probability and the time of detention are estimated at the same time, The outcome variables of Logistic regression were selected according to the occurrence of state transfer, and the simple statistical analysis and Logistic regression analysis were carried out in SAS9.2 software. The results showed that the multi state Markov model had good fitting effect. The results of multi state Markov model showed that the patient was in the MSM software package. State 1, state 2, state 3 and state 4 of total retention time are 3.19 months, 14.51 months 23.87 and 15.21 months respectively. Transfer strength matrix shows that the transfer intensity of state 2 converted to state 3 is converted to 1.5 times of state 1. The degree of transfer is about 242 times that of the state 5. The transfer probability matrix shows that the individual in the state 1 will transfer to the state 2 for a long time, the individual in state 2 will transfer to the state 3, the individual of the state 3 will transfer to the state 4, and the individuals of the state 4 will be transferred to the state 3. The variables of the single factor Logistic regression will be included in the multiple factors. In the stepwise Logistic regression analysis, the results showed that hypertension, fasting blood glucose, urea, microalbuminuria and HDL had an influence on the state 1 to state 2; age, fasting blood glucose, fasting insulin, low density lipoprotein, total cholesterol, apolipoprotein A1 and urea had an effect on the state 2 to state 1; fasting blood glucose, empty Abdominal insulin, high density lipoprotein, triglyceride and free fatty acids have an effect on the transfer of state 2 to state 3; fasting blood glucose, fasting insulin, high-density lipoprotein, triglyceride and free fatty acids have an effect on the state 3 to state 2 transfer; age, high density lipoprotein, triglyceride, creatinine and free fatty acids are 3 to state. The transfer of state 4 was affected; triglyceride, total cholesterol, creatinine and fasting insulin have an effect on the transfer of state 4 to state 3; fasting blood glucose, fasting insulin and triglyceride have an effect on the transfer of state 4 to state 5. The results of multifactor and multi state Markov model analysis show hypertension, glycated hemoglobin, free fatty acid, lipoprotein. A, urinary albumin / urine creatinine ratio and urine microalbumin have an effect on the transfer of state 1 to state 2; hypertension, urine albumin / urine creatinine ratio, age, fasting insulin, low density lipoprotein, apolipoprotein A1 affect state 2 to state 1 transfer; hypertension, urinary albumin / urine creatinine ratio, glycosylated hemoglobin, free fatty acid, Lipoprotein a, urinary microalbumin has an effect on the transfer of state 2 to state 3; hypertension, free fatty acids, apolipoprotein A1, high density lipoprotein, creatinine, and urea have an influence on the transfer of state 3 to state 2; hypertension, free fatty acids, apolipoprotein A1, influence on state 3, state 4, free fatty acid, creatinine, Urinary microalbuminuria, fasting insulin and triglyceride have an effect on the state 4 - state 3 transfer. Free fatty acids, creatinine, triglycerides, low density lipoprotein and fasting blood glucose have an influence on the state of the state 4 to state 5. Conclusion the multistate Markov model and the traditional Logistic return model all indicate that the chronic complications of diabetes are affected by the number of diabetic complications. The influence of age, admission, blood sugar, blood lipid, blood pressure, and renal function damage, but different variables have different influence on different state transfer. The analysis of multi state Markov model is based on the global perspective, and the effect of time on the outcome of the target is considered, so relative to the traditional Logistic The result of the regression is more accurate and scientific. Therefore, the multistate Markov model can be used as a useful supplement to the traditional Logistic regression model. At the same time, the results of the transfer intensity matrix and the transfer probability matrix show that the patient's condition in each state has a further deepening trend, and the patient's stay in the state 3 is longer. This time window can be used clinically to reverse the progression of the disease.
【学位授予单位】：重庆医科大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：O212.1;R587.1

【参考文献】