基于非随机缺失机制的模式混合模型医学应用分析-analysis of medical application of model hybrid model based on non-random absence mechanism.docxVIP

every layer to obtain outputs more correspond to professional explanation.PA-PMM and SA-PMM are constructed respectively with both systolic pressure and diastolic pressure as response variables. PA-PMM assumes that only intercepts are different with four missing patterns, while SA-PMM assumes that all parameters are different. As a result, the linear estimates of intercept in PA-PMM are close to each other, which are both close to that in the mixture effect model. However, SA-PMM has different parameters with four patterns. We regard the male patients with one year hypertension course who are 60 years old as objects, and the results depict that the complete cases has most lowest systolic pressure while pattern with three or more times missing has the highest, with diastolic pressure inverse. This illustrates that patients whose blood pressure not being effective controlled are apt to missing.4. Sensitivity analysis is a important content in MNAR data analysis.For MNAR is affected by unobserved response, and it can’t be certified by observed data, a sensitivity analysis should be conducted. Let l to be the sensitivity parameter and then take values of -10, -5, -3, 0, 3, 5, 10. As a result, , the basic predicted value of systolic pressure is close to that of PMM when l equals to 10, while diastolic pressure is the same when l equals to -3. This illustrates that patients with higher systolic pressure and lower diastolic pressure are apt to missing, which confirms that the missingness is MNAR.Key words: Missing Not at Random (MNAR), Non-ignorable missing, Pattern mixture model (PMM), Generalized score-type test, Sensitivity analysis学位论文独创性声明本人声明，所呈交的学位论文系在导师指导下本文独立完成的研究成果。文中任何引 用他人的成果，均已做出明确标注或得到许可。论文内容未包含法律意义上已属于他人的 任何形式的研究成果，也不包含本人已用于其他学位申请的论文或成果。与我一同工作的 同志对本研究所做的任何贡献均已在论文中作了明确的说明并表示谢意。本文如违反上述声明，愿意承担以下责任和后果：1、交回学校授予的学位证书；2、学校可在相关媒体上对作者本人的行为进行通报；3、本文按照学校规定的方式，对因不当取得学位给学校造成的名誉损害，进行公开道歉。4、本人负责因论文成果不实产生的法律纠纷。论文作者签名：日期：年月日学位论文版权使用授权书本人完全了解山西医科大学有关保留、使用学位论文的规定，同意学校保留或向国家