@MASTERSTHESIS\{IMM2012-06260,
    author       = "J. E. Blindum",
    title        = "Quality Control Techniques to Validate Analytical Methods in Biopharmaceutical Applications",
    year         = "2012",
    school       = "Technical University of Denmark, {DTU} Informatics, {E-}mail: reception@imm.dtu.dk",
    address      = "Asmussens Alle, Building 305, {DK-}2800 Kgs. Lyngby, Denmark",
    type         = "",
    note         = "Supervised by Associate Professor Murat Kulahci, mk@imm.dtu.dk, {DTU} Informatics, and Co-Supervisor Kim Westi Joergensen, Novo Nordisk. Thesis not publicly availble.",
    url          = "http://www.imm.dtu.dk/English.aspx",
    abstract     = "The purpose of this project is to analyse a control charting approach for handling a more complexed correlation structure than the usual structure in which all observations are assumed to be independent. The correlation structure of interest is the one that arise from a situation where observations obtained within a subgroup are correlated, whereas independence is assumed between subgroups. The most frequently used control chart approach, denoted the Shewhart approach, cannot be applied as a result of the proposed correlation structure. As a consequence, it is necessary to consider alternatives which are more unknown from a research point of view. The special correlation structure described above is often found in data obtained from an analytical method. A case study is conducted in cooperation with Novo Nordisk by analysing the control chart set-up for one of their analytical method. The idea is to give a detailed analysis of the control chart method that is currently applied by Novo Nordisk. The detailed analysis mainly consists of analysing the two types of risks,  and , related to this current method. The two types of risks combined explain a control chart system's ability to distinguish between 'good' and 'bad' samples, which is the main purpose of any control chart method. An additional object is to analyse data obtained from the specific process at {NN} in order to determine the significance of the underlying covariates and analyse specific tendencies from the parameters in the process. The main outcome of the detailed analysis is that the method is very capable of dealing with this type of correlation structure. However, a couple of recommendations are given to small changes that can improve the general ability of the current method. The most important recommendation is to use a larger number of subgroups (m) when establishing the control limits compared to current practise at Novo Nordisk. The specific  findings obtained for the {NN} process are more limited mainly due to a poor experimental design. Although a complimentary method is suggested for providing faster warnings on shifts in the process parameters, because such shifts are noted previously in the process. Finally, thoughts are given on all  findings and potential future work."
}