定量體外檢驗試劑線性確校統計評估方法之研究

Abstract

在檢定確效性的評估中，線性是最重要的特性之ㄧ。目前，評估線性的統計方法是由 Clinical Laboratory Standard Institute (CLSI) EP6-A準則 所提出。這個方法直接比較點估計值和允許區間並且完全忽略點估計值的抽樣誤差。另一個評估線性的方法是由 Kroll, et al. (2000) 所提出，他使用了線性平均離散程度 (ADL) 當作統計檢定方法，但是卻使用了不正確的統計假設與對應之統計檢定方法。因此，現有兩個方法的型一誤差可能會因而變大而無法做出正確評估。我們提出了雙尾檢定方法與 corrected Kroll’s 方法來改善現有方法之缺點。另一方面，我們亦建議了一個以廣義樞紐量(Generalized Pivotal Quantity, GPQ) 為基礎的 ADL 方法來克服由於ADL的機率分布存在未知之參數 (nuisance parameter)，而使得型一誤差受到未知之參數干擾的問題。 此外，我們亦建議了兩個新的用來評估線性程度的聚合型測度 (aggregate measure)。其中 SSDL 代表線性離散程度平方和。另一方面，CVDL則同時考量了變異程度的影響，而定義為相對於變異之線性平均離散程度平方和。經由模擬研究結果顯示，我們所提出各個方法皆比現有由 CLSI EP6-A 準則 與 Kroll et al. 所提出之方法不僅能有效控制型一誤差並且達到一定水準的檢定力。最後，針對我們提出的方法，也利用了數個例子進行資料分析與方法間之比較。

Linearity is one of the most important characteristics for evaluation of the accuracy in assay validation. The current estimation method for evaluation of the linearity recommended by the Clinical Laboratory Standard Institute (CLSI) guideline EP6-A (Tholen et al., 2003) directly compares the point estimates with the pre-specified allowable limit and completely ignores the sampling error of the point estimates. An alternative method for evaluation of linearity proposed by Kroll, et al. (Kroll, 2000) considers the statistical testing procedure based on the average deviation from linearity (ADL). However this procedure is based on the inappropriate formulation of hypothesis for evaluation of the linearity. Consequently, the type I error rates of both current methods may be inflated for inference of linearity. Therefore, we propose a two one-sided test (TOST) procedure and a corrected Kroll’s procedure as the more appropriate procedure for assessment of linearity. On the other hand, for the purpose to overcome the issue raised by the unknown nuisance parameters of the distribution of ADL, the GPQ-based ADL procedure is also proposed. In addition, we introduced two new alternative measures SSDL and CVDL which are defined as the sum of square of deviations from linearity and the deviations scaled by the variability, respectively, as the aggregate criteria for assessment of linearity. Unlike ADL and SSDL, CVDL can consider linearity and repeatability of an assay method simultaneously. The relationship among the dofferent aggregate criteria is discussed. The simulation studies are conducted to empirically investigate the size and power among the current and proposed methods. The simulation results show that all proposed methods can adequately control size better than the current methods. Numerical examples are also used to illustrate the application of the proposed methods.