工業衛生實驗室認證之分析能力比試統計方法

An Evaluation on the Statistical Protocol of the Proficiency Analytical Testing Program as Part of the National Industrial Hygiene Laboratory Accreditation Program

本文之目的在討論並比較數種常用的統計程序，並評估何種方式最適於為國內工業衛生實驗室認證之分析能力比試在統計容許範圍時運用，其中包括極端值的剔除方式、數據組統計平均值與統計變異係數的合理性評估等皆有詳盡的討論。結果顯示。當分析能力比試參加實驗室的家數在十五至五十家左右時，先以MND方法剔除極端值，再配合溫氏平均法所計算得到的統計容許範圍，可有效的做為評估參加實驗室分析能力的參考依據。由於實驗室在持續參加分析能力比試之後，其分析能力往往會有明顯的進步，因此可能使得回報數據組經過統計後所得到的容許範圍過於嚴苛，因此仍必須持續依照政府相關法規要求及實驗室分析能力，來定期檢討統計方法的適用性與統計容許範圍的合理性，以維持分析能力比試具體反應參加實驗室分析能力的功能，避免不合理的容許範圍使得新實驗室望而卻步。

A statistical protocol has been designed to evaluate the performance of member laboratories in the national industrial hygiene proficiency analytical testing (PAT) program. Several approaches were compared to obtain a confident acceptable range of testing results. Removing outliers, estimating the mean and searching acceptable coefficient of variation (cv) are addressed. Maximum normalized deviation associated with Winsorization method was found very suitable for calculating the performance range of the interlaboratory testing when the number of member laboratories was 15~50. Since the quality of member laboratories would improve after joining the PAT, the cv of mean usually will become more restricted. There is a need to continuously monitor the statistical method, and adjusts to meet a reasonably acceptable range according to the governmental regulations and the capability of laboratories. Therefore, the interlaboratory PAT program could really reflect an acceptable and average performance of laboratories, but not too restricted to exclude some new joining good laboratories.