探討臨床試驗遺失資料機制對ANCOVA分析之影響

The effect of missing data mechanisms on ANCOVA for clinical trials.

翁啟倫(Chi-Lun Weng)

指導教授 ： 歐士田

國立臺北大學/商學院/統計學系/碩士(2011年)

摘要

在臨床試驗中，即使是一個經過謹慎設計的試驗，資料的收集過程也是有可能會發生資料遺失的，若是沒有適當的補救辦法，最後的分析結果不僅有偏誤而且也可能導致錯誤的結論。因此如何使用正確的統計方法來進行遺失資料的分析一直是臨床試驗發展中重要的一部份。最後觀察資料挪動之插補法 (Last Observation Carried Forward，記作 LOCF) 是蠻常見的插補方法，因為此方法在執行過程是較為容易的。使用此方法後，分析結果會有何影響也是學者們一直以來研究的內容。Shao 等人(2005) 認為使用此方法後有一相對應的不偏假設檢定。此假設檢定不同於以往關心病人在試驗終了時的療效平均之假設檢定。在這樣的假設底下，一般 ANCOVA 模型分析在某些情況下是不確效的。 Shao 等人也提出了一個新的統計量，此統 計量在新的假設檢定條件下是穩健的(Robustness)。
本篇文章延續 Shao 等人 (2005) 之研究內容。希望探討在這樣新的假設檢定下，面對不同的遺失情況，一般 ANCOVA 模型下之分析與 Shao 等人 (2005) 所提出的新統計量之表現情況為何，將藉由統計模擬，分別對兩種方法之型一誤差率以及檢定力進行討論。

關鍵字

ANCOVA；LOCF；遺失資料；假設檢定；型一誤差率；檢定力

並列摘要

Missing data often occurs in clinical trials with multiple visits. If we don’t use an appropriate method to handle this problem, the analysis result may be not valid. Therefore, how to apply appropriate statistical methods to handle missing data is an important issue for drug development. Last observation carried forward (LOCF) is a common method because it is easy and simple to implement. There has been concern that implementing LOCF creates biases in statistical tests for treatment effect. Recently, Shoa and Zhong (2003) proposed an unbiased hypothesis by implementing LOCF. Under this hypothesis, the asymptotic size of the LOCF test is not α for some scenario in one-way ANCOVA. Shao et al. (2005) derived asymptotically valid tests when LOCF tests are invalid.
This article follows the work by Shao et al. (2005). Under the unbiased hypothesis and different missing data mechanisms, we compare the performances
between LOCF tests and the tests derived by Shao et al. (2005). The statistical simulation will be conducted with respect to the type I error rates and powers of two tests.

並列關鍵字

ANCOVA；LOCF；Missing date；Hypothesis；Type I error rate；Power

參考文獻

Dawson, J. D. (1994a). Comparing Treatment Groups on the Basis of Slopes, Areas-under-the-curve, and Other Summary Measures. Drug Infor. J. 28, 723-732.

EMEA (1998). Note for Guidance on Statistical Principles for Clinical Trials.

EMEA (2001). Point to Consider on Missing Data.

EMEA (2007). Recommendation for the Revision of the Points to Consider on Missing Data.

Heyting A. and Toolboom, J. T. B. M. and Essers, J. G. A. (1992). Statistical Handling of Drop-outs in Longitudinal Clinical Trials. Statist. Medicine 11, 2043-2061.