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

ISCB2014_abstract_book

18 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsMonday, 25th August 2014 • 9:00-10:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust C05.5 Comparison of analysis approaches for multi-level vascular imaging data JR Forman1 , SJ Bond1 , KM Mäki-Petäjä2 , IB Wilkinson1,2 1 Cambridge Clinical Trials Unit, Cambridge, United Kingdom, 2 Clinical Pharmacology Unit, University of Cambridge, Cambridge, United Kingdom Vascular PET-CT imaging studies produce multi-level data structures, but at present there is no consensus on how best to analyse such data. We present our analyses applied to data observed from a cohort of patients with rheumatoid arthritis and a matched control cohort with stable car- diovascular disease on whom vascular PET-CT scans were taken (UK ethics ID 08/H0305/19). The primary endpoint is change in vascular inflamma- tion, as measured by PET-CT. Each participant receives a PET-CT scan at baseline and 8 weeks later. Each scan covers up to five vessels per partici- pant, and each vessel is analysed in “slices”, generating a three-level data structure: participants, vessels, slices. Here, we present and compare three analysis approaches that have been used in previous studies. 1: Data are pooled by vessel and by patient, by considering the change in the mean (or max) inflammation. 2: An index vessel is selected at baseline and followed-up in each patient. The index vessel is selected as the vessel with highest inflammation at baseline. 3: A multi-level model incorporates the complete data set and data struc- ture. Through comparing these analysis approaches, we aim to identify a method which provides an accurate measure of the treatment effect, and a straightforward interpretation.   C06 The biostatistician’s toolbox I C06.1 Translational statistics and dynamic nomograms J Newell1 , A Jalali1 , A Alvarez-Iglesias1 , M O’Donnell1 , J Hinde1 1 National University of Ireland, Galway, Ireland Translational Medicine promotes the convergence of basic and clinical research disciplines and the transfer of knowledge on the benefits and risks of therapies. In an analogous fashion we propose the concept of Translational Statistics to facilitate the integration of Biostatistics within clinical research and enhance communication of research findings in an accurate manner to diverse audiences (e.g. policy makers, patients and the media). One example of this knowledge transfer is in Classification problems, com- monly presented to statisticians, typically involving a binary outcome. The usual summary is the Odds Ratio. It has been argued that, when possible, a summary quoting the underlying probabilities is more informative than one based on ratios of odds or indeed of probabilities. As statistical in- ferential methods become more computational the models arising are increasingly complex and difficult to interpret. Nomograms can be used as calculators of a predicted response for differ- ent values of the explanatory variables but can become cumbersome as the model becomes more complex. Tree based models allow prediction on a probabilistic scale but as the classifier becomes more complex (e.g. random forests) a simple classification rules is unavailable. In this presentation dynamic nomograms are introduced that can be cre- ated automatically from any glm model object in R. In theory any model appearing in a scientific publication can be accompanied by a URL direct- ing the ‘user’ to the accompanying dynamic nomogram from which the results of the models are directly translational and the robustness of the model verified through automatically generated model diagnostic plots. C06.2 Dynamic graph generation and data analysis of complex data: a web-application based on R and shiny L Lusa1 , C Ahlin1 1 University of Ljubljana, Ljubljana, Slovenia R statistical environment includes facilities for data display and analysis that are extremely flexible. It recently became rather straightforward to create interactive web applications and interactive graphics based on code written in R, using the shiny package and Scalable Vector Graphics (SVG). We illustrate, through an example, the feasibility of developing a user friendly web application that incorporates a variety of interactive graphical displays and tools for the analysis of complex data. The medplot application was developed to help medical researchers to explore and analyse longitudinal data, where numerous variables are re- corded for each patient over time. Several interactive graphical displays allow an easy exploration of the data. The analyses tools evaluate the as- sociation of the variables with other characteristics of the patients, taking into account the multiple testing problem, the repeated measurements and the possibility of non-linear associations between the covariates and the outcomes. The application can be used by users that are not familiar with R or other statistical programs. It can be used through a web-browser and it does not require the installation of any program. Template spread sheets for data preparation are also provided, together with example data from a clinical study including patients with erythema migrans, where the variables are the presence and intensity of numerous symptoms recorded over time. C06.3 The grammar of parametric boxplots R Vonthein1 1 Universität zu Lübeck, IMBS, ZKS, Lübeck, Germany Boxplots were introduced as distribution-free summary plots of distribu- tions shown side by side. They may be defined for specific distributions, of an assumed model, say. Then, numbers counted and numbers measured need to be treated differently. One may accomodate censoring. Following the grammar of graphics approach, there may be several gra- phems or layers to a plot. First, all data may be plotted neatly side by side; then summaries are added. Summaries may be nonparametric or para- metric, e.g. estimated quantiles, densities or both. Color and symbols add more information or help discern data from summary. You may decide between joint and conditional distributions and you may pick one out of several theoretical distributions. Distributions of counts may be drawn to emphasize discreteness or to hint at an approximate continuous distribu- tion by confluent graphems. There are several ways to illustrate the fit of the model to the data. Applications to real data of an R-function available from the author will illustrate possibilities and limitations. Especially the dotplot layer needs special attention in case of very small or large samples or widely differ- ing sample sizes. Parametric boxplots look focused on central tendency rather than extreme quantiles when compared to distribution or survival functions. Boxplots readily show joint distributions with many categories when density and distribution functions get unreadable.

Pages Overview