Copyright © Lippincott Williams & Wilkins. Unauthorized reproduction of this article is prohibited. Determining the adequate number of internal quality control levels: the example of coagulation factor VIII assay Fre´de´ ric Sobas a , Laurent Mazliak b , Audrey Bellisario a , Mathieu Lefranc a , Anne Lienhart a , Christophe Nougier a and Claude Ne´ grier a Taking the specific case of coagulation factor VIII assay, we determined the characteristics of an internal quality control panel assuring control of the assay method for all of the critical factor VIII concentrations. The precision of the assay method was determined on six control materials C1– C6, with expected factor VIII levels of 1, 5, 30, 50, 80 and 150 U/ dl, respectively. Given that, when two control levels correlate statistically, the information provided by one of them is redundant, we used correlation and principal components analysis to define a priori adequate and inadequate control panels. For each of these panels, we calculated the number of runs required, using Hotelling’s method, to detect a shift expressed on C1 and impacted on C2, C3, C4, C5 and C6 in relation to the correlation phenomena among the six levels. The C1/C6 panel proved to be as informative in this regard as the complete panel for a 1 U/dl shift simulated on C1 and impacted on other levels too. These correlation phenomena allow the biologist to implement fewer control levels than there are critical concentrations needing to be explored in the internal quality control plan. Blood Coagul Fibrinolysis 19:433–437 ß 2008 Wolters Kluwer Health | Lippincott Williams & Wilkins. Blood Coagulation and Fibrinolysis 2008, 19:433–437 Keywords: average run length, correlations between internal quality controls, factor VIII, Hotelling, number of control levels a Haemostasis laboratory, Edouard Herriot hospital, Lyon and b Laboratory of Probabilities and Random Models, Paris VI University, France Correspondence to Fre´de´ ric Sobas, Pavilion E sous sol Hoˆ pital Edouard Herriot, Place d’Arsonval, 69 437 Lyon, France E-mail: frederic.sobas@chu-lyon.fr Received 13 February 2008 Revised 7 April 2008 Accepted 13 April 2008 Introduction There is no consensus as to the optimal number of internal quality control (IQC) levels. Some experts advise using as many control levels as there are critical concentrations [1]. Others argue for using at least two control levels spanning the range of critical concentrations involved in patient management [2,3]. Medical biologists thus have no precise recommendations as to how to construct the optimal IQC panel with respect to the multiple specifications, which the measurement methods are intended to meet. From the literature data, clinical decision intervals around the coagu- lation factor VIII concentrations representing clinical decision thresholds for hemorrhage and thrombosis risk can be set at 1, 5, 30, 50, 80 and 150 U/dl [4– 8]. The IQC plan should be able to detect increase of either random error (loss of precision) or systematic error (increase of bias) or both by the method [9]. Methods tend nowadays to be automated (as in the case of the method here under study), so that analytic error is mainly a matter of increased systematic error [10]. We therefore approached the deter- mination of the optimal control panel in terms of shift detection. We used a multivariate approach. The multi- variate approach – in which Hotelling’s T 2 is the reference method [11,12] – consists of calculation of a single index which will allow an overall interpretation of results of all control levels. If the measurement apparatus sustains a shift, this will obviously show up, to a greater or lesser degree, on all of the various controls of the series, that is, the concentrations of these different controls are not independent [13]. Hotelling’s T 2 index enables variations, which may be simultaneous, to be taken into account. According to Marquis and Masseyeff [14], errors are thereby detected between two and five times as fast as when unidimensional quality control is employed. Shift detection capability of each panel may be compared with the concept of average run length (ARL) – the number of control runs required to detect an analytic problem [15]. Methods Reagents and automated coagulation analyzer Our lab uses lyophilized control plasma from Dade Behring (Marburg, Germany) VEQA and VEQB, with expected factor VIII concentrations of around 80 and 30U/dl, respectively. Precision BioLogic (Dartmouth, Canada) produced six frozen control concentrations for the present study C1, C2, C3, C4, C5 and C6, with expected factor VIII concentrations of 1, 5, 30, 50, 80 and 150U/dl, respectively. In step 1, citrated, pooled normal plasma was depleted in von Willebrand factor and factor VIII by immunoaffinity chromatography. The resulting deficient plasma was found to be less than 1U/dl for factor VIII Original article 433 0957-5235 ß 2008 Wolters Kluwer Health | Lippincott Williams & Wilkins