陳秀熙臺灣大學:流行病學研究所李芳儀Li, Fang-YiFang-YiLi2010-05-052018-06-292010-05-052018-06-292009U0001-1808200914022200http://ntur.lib.ntu.edu.tw//handle/246246/180827闡明難治型癲癇病患接受手術治療的預後情形十分重要，然而，病患每段時間癲癇發作的追蹤資料伴隨著異質性和相關性，兩者都妨礙我們使用傳統的統計方法進行分析，因此，在有無考慮共變項的前提下，我們提出高階和移動-停留馬可夫模型以適用於擁有這些棘手特性的資料。74名曾經接受手術治療的病患在手術後五年內接受追蹤，以癲癇的月發作次數分類成以下三種狀態：良好：月發作0次；輕微：月發作1-2次；嚴重：月發作3次以上。收集的共變項包含年齡、性別、初次發病年齡、術前用藥總數、出生前後損傷、顳葉內側硬化以及初次發病到手術間隔時間等。將一階馬可夫模型、二階馬可夫模型以及移動-停留模型應用於模式化這些資料，並使用最大概似法進行參數估計，一階馬可夫模型的結果顯示進展和復原的速率相依於前一個狀態，給定前一個狀態為良好或輕微，進展到嚴重狀態的力量分別是1%和15%；二階馬可夫模型以及移動-停留模型移動者的狀態轉移行為也有相似的結果，值得注意的是，移動-停留模型的結果顯示停留者的比例為62%，而且根據AIC，移動-停留模型有最佳的配適程度。共變項特定回歸模型方面，我們發現對於狀態惡化的總效應而言，術前用藥總數和出生前後損傷是兩個加速因子，相反的，顳葉內側硬化則是一個減速因子。總結來說，高階馬可夫鏈和移動-停留模型可以有效地模式化病患狀態的進展和回復，此外，對於難治型癲癇病患手術治療預後因子的鑑別也十分有用。Elucidating the prognosis of intractable epileptic patients treated with surgery is important. However, as follow-up data on the episodes of seizure of these patients are fraught with heterogeneity and correlated property, both preclude one from analyzing the data with conventional statistical method. The higher-order and mover-stayer Markov model with or without considering covariates were therefore proposed to accommodate these intractable properties. A total of 274 patients who had undergone surgery were followed over five years on the monthly episodes of seizure with classification of three states: Normal – 0 count of episode; Mild – 1 to 2 counts of episode; Severe – above 3 counts of episode. Baseline covariates were collected including age, gender, age of onset, medication, perinatal insult, MST, and duration of epilepsy.irst-order Markov model, second-order Markov model, and mover-stayer model were applied to modeling this data by using maximum likelihood estimate (MLE) method. The result of first-order Markov model showed the rates of progression and regression depended on initial status. The forces of progression to severe state were, 1% and 15%, respectively, for normal and mild in nascent state. The similar findings were noted in the second-order Markov model and the transitions for mover in the mover-stayer model. Note thate the results of mover-stayer model found 62% stayer. In the light of AIC criteria, we found the mover-stayer model had the best fit.n the covariate-specific regression model, we found medication and perinatal insult were two accelerated factors for the net force of progression and MTS was a decelerated factor.n summary, higher order and mover-stayer model were very useful for modeling the progression and regression and identification of prognostic factors in intractable epileptic patients treated with surgery.第一章 前言 1二章 文獻回顧 2.1 高階馬可夫鏈 2.2 共變項特定馬可夫鏈模型 3.3 移動-停留模型 4.4 癲癇症 5三章 資料 6.1 研究對象 6.2 資料探索 6.3 資料結構 7四章 模型建立 8.1 符號定義 8.2 模型建立 8.2.1 一階馬可夫鏈模型 8.2.2 二階馬可夫鏈模型 9.2.3 一階馬可夫鏈移動-停留模型 9.2.4 二階馬可夫鏈移動-停留模型 10.3 參數估計 11.4 模式選擇 12.5 預測t-階段轉移機率 12.6 共變項特定模型 13.6.1 一階馬可夫鏈模型 14.6.2 二階馬可夫鏈模型 15.6.3 移動-停留模型 15.6.4 模式選擇 15.6.5 共變項對於轉移行為的影響 15.7 軟體 16五章 結果 17.1 描述性統計 17.2 參數估計結果 17.1.1 一階馬可夫鏈模型 17.1.2 二階馬可夫鏈模型 17.1.3 一階馬可夫鏈移動-停留模型 17.1.4 二階馬可夫鏈移動-停留模型 18.3 模式選擇 18.4 預測t-階段轉移機率及平穩分布 18.5 共變項-特定模型 18.4.1 一階馬可夫鏈模型 19.4.2 二階馬可夫鏈模型 19.4.3 移動-停留模型 20六章 討論 21考文獻 23application/pdf467533 bytesapplication/pdfen-US高階馬可夫模型移動-停留馬可夫模型難治型癲癇病患High-order Markov modelMover-stayer Markov modelIntractable epileptic patientsthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/180827/1/ntu-98-R96842007-1.pdf