Please activate JavaScript!
Please install Adobe Flash Player, click here for download

ISCB2014_abstract_book

140 ISCB 2014 Vienna, Austria • Abstracts - Poster PresentationsWednesday, 27th August 2014 • 15:30-16:00 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust P4.5.176 An empirical comparison of methods for a conjoint analysis survey of knowledge translation in women’s addiction agencies T Vanniyasingam1 , C Cunningham1 , G Foster1 , A Niccols1 , L Thabane1 1 McMaster University, Hamilton, Canada   Research has identified effective models of addiction services for the prevention and treatment of substance abuse among women, including programs that incorporate pregnancy-, and child-related services. Despite these effective evidence-based methods, there is a lack of implementation of such findings in the delivery of addiction services. We conducted a discrete choice experiment with 1379 service providers and administrators from 333 addiction agencies. The participants were presented with a set of scenarios comprised of combinations of different attributes and corresponding levels, using a fractional factorial design. Our objective is to determine professional development preferences by addiction service providers and administrators for the enhancement of addiction services. We will empirically analyze discrete choice experiment data using various statistical methods to account for within-participant correlation—specifi- cally, multinomial logit and probit models. We will report the relative im- portance (or ranking) of each attribute and level of attribute in determin- ing participants’preferences. P4.5.179 Outlier detection in functional time series: an application to mortality rates J Vilar1 , G Aneiros1 , P Raña1 1 University of A Coruña, A Coruña, Spain   A new procedure to detect outliers in functional data, which takes into account the dependence, is proposed. An application to mortality rates is presented. There are some contexts in which the data should be treated as functions, rather than a string of values. This kind of data, known as functional data, appears in many areas, such as bio-medical. In this work we study a func- tional time series, which is a time series made of functional data. It is im- portant then to consider the dependence found in this kind of data. Our approach to detect outliers follows the idea of Febrero et al. (2008) which assumes the independence between the data, modifying some steps and adapting it to consider the dependence of the functional data. Briefly, the procedure obtains the functional depth of the data and looks for a cutoff using bootstrap techniques. The data with depth lower than the cutoff are classified as outliers. A comparison between our proposal and other different methods to de- tect outliers in functional data, using simulated data, has been carried out showing the good behavior of our approach. In this paper, the method is applied to mortality rates. Specifically, we use data of the French male age-specific log mortality rates, between 1899 and 2005. The effect of historical events, as the World War I or the Spanish flu pandemic, is reflected in the mortality rates and some of these years will be classified as outliers.   P4.5.184 Split and merge techniques for Gaussian mixture learning Y Wang1 , M Titterington2 1 King’s College London, London, United Kingdom, 2 University of Glasgow, Glasgow, United Kingdom   EM algorithm has been the first choice for mixture density estimation. However, the EM, essentially a local search algorithm, has a number of lim- itations such as slow to converge, sensitive to initialization, and may get stuck in one of many local maxima of the likelihood function. Inspired by the idea of splitting and/or merging components sequentially or simulta- neously based on certain criteria, a number of algorithms were introduced to overcome these limitations. We tested 3 algorithms of this kind, namely, the IPRA (Iterative Pairwise Replacement algorithm) (merge) by Scott and Szewczyk (2001), the Greedy EM (split) by Vlassis and Likas (2002), and the SMEM (split and merge) by Ueda et al (2000), on both simulated and real-world genomics data. We commented on their performances with comparison to the stan- dard EM+BIC approach (MCLUST package in R). Further, we extended the IPRA into multidimensional problems by pro- posing a multivariate IPRA (mIPRA), which uses the minimal spanning tree (MST) to limit searches, thereby reduces the number of comparisons from O(n2 ) to O(nlogn). We found the mIPRA was efficient and effective when fitting mixtures with a large number of components. P4.5.187 Preanalytical variation: computing variance estimates from systematic differences and models for clinical practice MS Sylte1 , T Wentzel-Larsen2,3,4 , BJ Bolann1,5 1 Haukeland University Hospital/Lab. of Clinical Biochemistry, Bergen, Norway, 2 Centre for Child and Adolescent Mental Health, Oslo, Norway, 3 Norwegian Centre for Violence and Traumatic Stress Studies, Oslo, Norway, 4 Haukeland University Hospital/ Centre for Clinical Research, Bergen, Norway, 5 University of Bergen, Department of Clinical Science, Bergen, Norway   Aims: A framework for estimating preanalytical uncertainty using linear mixed-effects models is previously developed, with fixed factors for sys- tematic differences between preanalytical settings and random effects for structural levels. Settings for the systematic factors vary within clinical practice, and our aim is to combine explicit modelling of this variation for each systematic difference to be added to the appropriate random effect. Methods: For a discrete fixed effects covariate, the additional variance is based on fixed effects coefficients and corresponding probabilities from clinical practice. For a continuous covariate, it is based on the fixed effect and an assumed rectangular distribution from clinical practice. Results: The systematic and random effects created from different han- dling of blood samples before analysis e.g. using different needles, mix- ing of the blood samples and different transportation is combined with frequencies and calculated into a total preanalytical uncertainty. By this approach the interpretation of patient results can be improved. Conclusions: We have developed a unified procedure for combining fixed effects with estimates from clinical practice to produce variance estimates to combine with random effects. The procedure is quite general and will be investigated further in our laboratory.

Pages Overview