請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/59281
標題: | 調整網絡統合分析中相依性治療權重之統計方法 Statistical Approaches to Adjusting Weights for Dependent Arms in Network Meta-analysis |
作者: | Yu-Xuan Su 蘇育萱 |
指導教授: | 杜裕康(Yu-Kang Tu) |
關鍵字: | 統合分析,網絡統合分析,對比基礎模型(contrast-based model),分口研究,特殊研究設計, meta-analysis,network meta-analysis,contrast-based model,split-mouth studies,special study design, |
出版年 : | 2017 |
學位: | 碩士 |
摘要: | 背景:網絡統合分析(network meta-analysis)能夠藉由收集隨機對照試驗(randomized controlled trial)的證據來比較多種治療的治療效果。而在大部份的臨床試驗中使用的是平行設計(parallel design),也就是研究對象會被隨機分派到不同的治療組別,並且只接受一種治療,但是有些臨床試驗採用的是特殊的研究設計(special design),最常見的就是分口設計(split-mouth design)與交叉設計(cross-over design),而這些特殊研究設計裡的研究對象接受不只一種治療,因此治療與治療之間的結果也就不再是獨立的。在分析這類型的資料時,必須將其相關性(correlation)納入考慮,若忽略治療與治療之間的相關性將會導致錯誤的結果。
目的:為解決特殊研究設計所導致治的相依性問題,本篇論文提出新的統計方法解決在網絡統合分析中相依性治療所造成的偏差(bias)問題,並提供更便捷的計算方法。 方法:在此篇論文中,我們提出了三種調整網絡統合分析中相依性治療權重的統計方法,分別命名為標準方法(the standard approach)、調整變異數方法(the adjusting variance approach)、降權重方法(the reducing weight approach),並且使用牙周病患者資料作為分析案例,針對三個統計方法的結果進行比較。 結果:這三種統計方法,主要的差別是在於調整特殊研究設計所導致的相依性。標準方法主要是分別計算平行設計試驗與特殊研究設計試驗的變異數(variance)與共變異數(covariance),並且再去設置研究內的共變異數矩陣(within-study variance-covariance matrix),此矩陣為一個分塊對角矩陣(block-diagonal matrix)。調整變異數方法則是先調整在特殊研究設計試驗裡各個治療的標準差,再去設置研究內的共變異數矩陣。最後,降權重方法主要是分別計算平行設計試驗與特殊研究設計試驗的變異數,並且使用圖論(graph theory)與電路學(electric network)上所推導出的公式與概念來解決相依性的問題。我們發現不論採用何種方法來分析特殊的研究設計的資料,基本上結果都非常的相近。 結論:對於特殊研究設計所導致的相依性問題,這三種統計方法都提供了更可信的結果。相較於標準方法,調整變異數方法的優點為使用的便利性,在分析過程中我們可以利用現有的套件(package)完成。而降權重方法的優點為研究內的共變異數矩陣的設置是相對簡單且直觀的。這些方法都能更有效率的分析由特殊研究設計所導致治的相依性資料。 Background: Network meta-analysis compares multiple treatments in terms of their efficacy and harm by using evidence from randomized controlled trials. Most clinical trials use parallel design, where patients are randomly allocated to different treatments and receive only one treatment. However, some trials use special designs such as split-mouth and cross-over designs, where each patient may receive more than one treatment. Data from treatment arms in trials with special study designs are no longer independent, so the correlations between dependent arms need to be taken into consideration in statistical analysis. Ignoring correlations would lead to incorrect results. Objectives: The main objective was to develop statistical approaches to adjusting weights for dependent treatment arms within trials with a special design. Methods: We proposed and demonstrated three approaches: the standard approach, the adjusting variance approach, and the reducing weight approach. An example of periodontal regeneration was used to show how these approaches could be undertaken and implemented within statistical software packages, and to compare the results from different approaches. Results: The main difference between the standard approach, the adjusting variance approach, and the reducing weight approach is how they deal with the correlations within special design trials. The standard approach calculates the variance of treatment contrast and the covariance between treatment contrasts and then set up the within-study variance-covariance matrix. The adjusting variance approach adjusts the standard deviation of each treatment arm within special design trials. The reducing weight approach modifies the variance of treatment contrast to adjust for the correlations between dependent treatments and applies electrical network method to adjust for correlations within multi-arm trials. We found that these three approaches yielded identical results of consistency and design-by-treatment inconsistency models. Conclusions: These three approaches provide more accurate results for network meta-analysis with dependent arms in a trial. The advantage of the adjusting variance approach is that it can be implemented within the network package in STATA. In contrast, the reducing weight approach provides an easy way to set the within-study variance-covariance matrix structure. As a result, these approaches are more effective to analyze the dependent data yielded by special designs. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/59281 |
DOI: | 10.6342/NTU201701373 |
全文授權: | 有償授權 |
顯示於系所單位: | 流行病學與預防醫學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-106-1.pdf 目前未授權公開取用 | 2.76 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。