ENBIS-13 in Ankara

15 – 19 September 2013 Abstract submission: 5 February – 5 June 2013

My abstracts

 

The following abstracts have been accepted for this event:

  • Decision Support with Knowledge and Data: A Bayesian Network for Trauma Surgery

    Authors: Barbaros Yet (Queen Mary University of London), Zane Perkins (Queen Mary University of London), Nigel Tai (Queen Mary University of London), William Marsh (Queen Mary University of London)
    Primary area of focus / application: Modelling
    Keywords: Bayesian networks, Decision support, Clinical decision making, Knowledge engineering
    Submitted at 13-Apr-2013 12:42 by Barbaros Yet
    Accepted
    Models that imitate previous decisions in data often provide limited support to an experienced decision maker. Rather than directly recommending a decision, an effective decision support should assist the expert intuition by calculating the probabilities of risks and outcomes of available decision alternatives. This is possible when causal relations between the decisions and outcomes are modelled using both expert knowledge and statistical data.

    We present a method for developing causal Bayesian networks from knowledge and data with a case study about mangled extremities in trauma care. A critical decision in this domain is whether to amputate or salvage a mangled extremity. This decision has multiple and interrelated adverse outcomes: a salvage decision may not always be better since a salvaged limb may be less functional than a prosthetic limb and aggressive salvage attempts may put a patient’s life in danger especially when the patient is physiologically unwell. Previous models are scoring systems that calculate a score for a patient and recommend an amputation if the score is above a certain threshold value. However, their utility is limited for an experienced clinician’s judgement especially when the score is close to the threshold and decision support is most needed. A more useful decision support can be provided by a model that calculates the probabilities of limb viability and death outcomes following amputation and salvage decisions. Our method combines data with knowledge to develop a causal Bayesian network for predicting these outcomes.
  • EWMA and CUSUM Charts on Transformed Variables

    Authors: Paulo H. C. Maranhão (Military Institute of Engineering), Eugenio K. Epprecht (Catholic University of Rio de Janeiro)
    Primary area of focus / application: Process
    Keywords: Known directions, Transformed variables, Multivariate processes, Control charts
    Submitted at 13-Apr-2013 21:15 by Paulo Maranhão
    Accepted
    Most of the works that propose schemes of Multivariate Statistical Process Control (MSPC) and that analyze the performance of these schemes consider changes in the observed variables. In an earlier paper, we propose a method for the monitoring of multivariate processes, consisting of control charts on the transformed variables, in which changes in the process parameters are due to special causes that affect non-observable variables and occur in known directions. Besides, we compare its performance with that of Shewhart's charts on the observed variables, on the principal components, and with that of T2 charts on the vector of observed variables. Results obtained by simulation show that the proposed scheme, has better performance in most of the cases analyzed. The comparisons were based on the in-control and out-of-control probabilities of signal. However, it is known that of Shewhart´s Chart loses efficiency for small shifts in the mean or variance. This work extends the study of control charts on the transformed variables for versions EWMA and CUSUM. The analysis of performance is done assuming shifts in the mean of the known directions, since these are associated to special causes, and/or increases of the variance in these same directions.
  • Statistical Considerations in the Design and Analysis of Drug Product Stability

    Authors: Hans Coppenolle (Johnson & Johnson)
    Primary area of focus / application: Quality
    Keywords: Stability, Shelf life, Release, Risk, Design, Accelerated
    Submitted at 14-Apr-2013 22:55 by Hans Coppenolle
    Accepted
    18-Sep-2013 09:00 Statistical Considerations in the Design and Analysis of Drug Product Stability
    Quality of a pharmaceutical drug product is defined based on critical quality attributes that should be within specifications established to ensure its identity, strength, quality, and purity. Stability is the capacity of a drug product to remain within the specifications throughout shelf life. A stability study typically requires evaluation of changes of a critical quality attribute over stability time, calculation of the drug product shelf life and calculation of a release limit. These concepts, statistical risk assessment and experimental design issues will be discussed based on real life applications of stability studies in pharmaceutical industry.
    Quality by design is a new approach to pharmaceutical product quality encouraged by regulatory agencies. The focus of this concept is that quality should be built into a product with a thorough understanding of the product and process by which it is developed. An application of quality by design to drug product stability based on accelerated stability testing will also be discussed.
  • Smoothed Rank Method for Two-sample Location Problem

    Authors: Feridun Tasdan (Western Illinois University)
    Primary area of focus / application: Design and analysis of experiments
    Keywords: Rank estimation, Shift parameter, Smoothing, Two-sample problem, Mann-Whitney-Wilcoxon
    Submitted at 15-Apr-2013 07:38 by Feridun Tasdan
    Accepted
    17-Sep-2013 10:30 Smoothed Rank Method For Two-Sample Location Problem
    Smoothed rank procedure is proposed to estimate the shift parameter in the Two-Sample Location Problem. The regular ranks in Wilcoxon’s two sample rank sum method are replaced with smoothed ranks which are computed by using a continuous distribution function, Φ, with an appropriate smoothing (called bandwidth in kernel density estimation) parameter h. There are several advantages of using smoothed ranks. One of them is to
    avoid using discrete step function in the process since the smoothed ranks gives continuous dispersion or gradient functions for estimating the shift parameter. As a result, smoothed ranks gives faster rate of converge in the iterations compare to the methods which use regular ranks. In application, the smoothing parameter h chosen to reach the fastest rate of converge that satisfy the bandwidth constrain. A similar approach can be found in Heller (2006) who studied smoothed
    rank regression with censored data. Asymptotic distribution of the proposed shift estimator is developed and Monte Carlo simulations are performed to examine its finite sample properties against known shift estimators.
  • Using DoE in Industry: Understanding and Optimizing a Grease Production Process

    Authors: Winfried Theis (Shell Global Solutions International B.V.)
    Primary area of focus / application: Design and analysis of experiments
    Keywords: Design of experiments, Case study, Oil industry, Process understanding
    Submitted at 15-Apr-2013 08:55 by Winfried Theis
    Accepted
    16-Sep-2013 15:25 Using DoE in Industry: Understanding and Optimizing a Grease Production Process
    In this case study we show, how we used a series of experimental designs to improve the understanding of a special grease production process by devising a series of experimental designs. The first challenge was the duration of single runs on the pilot plant installation limited the feasible number of experiments substantially, such that a screening on the combined set of factors regarding the recipe and the process parameters was not feasible on the pilot plant. We devised a way to do parts of the screening on laboratory scale first and then after eliminating most recipe factors we were able to set up pilot plant tests. The large amount of product produced per batch in the pilot plant allowed us to test the effects of three post-processing methods per recipe. This experimental process settled a number of discussions going on amongst the experts on the influences of some of the factors. It revealed some unexpected additional influences, which had not been well-understood before. In total this work convinced other colleagues of the power of Design of Experiments and led an increased used of DoE in the development of Grease products.
  • A Comparison Study of On-line Data-driven and Model-based Denoising Methodologies

    Authors: Marco P. Seabra dos Reis (Department of Chemical Engineering, University of Coimbra), Ricardo S. Rendall (Department of Chemical Engineering, University of Coimbra)
    Primary area of focus / application: Process
    Keywords: On-line denoising, Model-independent denoising, Model-based denoising, Multiscale analysis
    Submitted at 15-Apr-2013 10:29 by Marco P. Seabra dos Reis
    Accepted (view paper)
    17-Sep-2013 10:10 A Comparison Study of On-line Data-driven and Model-based Denoising Methodologies
    Signal denoising is a pervasive operation in most on-line applications, such as engineering process control and on-line optimization, strongly affecting the outcome of these higher-level tasks and therefore the final variability of processes and products. In this work we compare the performance of a set of currently available on-line denoising filters, for several types of test signals, affected with noise with different signal to noise ratios (SNR). Both model-independent (or data-driven) and model-based approaches are contemplated. A new class of multiscale denoising algorithms is also considered in this study, based on the on-line wavelet multiresolution decompositon. After proper tuning, the methods are tested and their performances compared. This work will provide clear guidelines on the use of denoising methodologies, allowing for a better management of the impact of the propagation of unstructured components of variability in the final outcome of the process. These components might be magnified by the action of automatic control loops or other supervisory control operations. This is an often overlooked source of variability whose impact might be significant, but that can be addressed in systematic and effective way, as shown in this work.