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


ISCB 2014 Vienna, Austria • Abstracts - Oral Presentations 25Monday, 25th August 2014 • 14:00-15:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust for estimated coefficients. We conduct a Monte Carlo simulation study to assess the finite sample performance of our procedure. The proposed method is illustrated by an empirical analysis of smoking cessation data, in which the important question of interest is to investi- gate the association between urge to smoke, continuous response, and the status of alcohol use, the binary response, and how this association varies over time. C09.3 Joint modelling of longitudinal and time-to-event data: a comparison between shared random- effect and latent class model M Bastard1 , J-F Etard1,2 1 Epicentre, Paris, France, 2 UMI 233 TransVIHMI, IRD, Université Montpellier 1, Montpellier, France In clinical research, a growing interest is to investigate the effect of a longi- tudinal marker on the occurrence of an event. Joint models have recently been developed to investigate this association within two approaches: the shared-random effect (SRE) and the latent class approach (LC).We propose to present, apply and compare the two approaches. The SRE consists in including the random effects of the mixed model used for longitudinal data as covariates in the model for the event to model the correlation between the marker profile and the event. The LC makes the assumption that the population is composed of several latent sub- populations in which the profile of the longitudinal marker and the risk of the event are different, using a discrete latent variable to link the marker profile and the event. To illustrate the two models, we apply them to a cohort of HIV-infected pa- tients supported by Médecins sans Frontières to study the effect of longi- tudinal CD4 profile on survival. Both approaches show consistent results, with a higher risk of death for patients with low CD4 profile. SRE allows a straightforward interpretation of the effect of longitudinal marker on the event, but is numerically intensive and assumes homo- geneity in marker evolution. LC is particularly useful when population is heterogeneous and provides an easier interpretation for clinicians but requires multiple fits. SRE and LC approaches are two powerful tools to model jointly longitudinal and time-to-event data and should be used carefully according to the research question of interest. C09.4 Longitudinal patterns of stages of change and lifestyle intervention outcomes - a latent class analysis with distal outcomes L Jiang1 , S Chen1 , J Beals2 , CM Mitchell2 , SM Manson2 , Y Roubideaux3 1 Texas A&M University, College Station, United States, 2 University of Colorado Anschutz Medical Campus, Aurora, United States, 3 Indian Health Service, Rockville, United States Stages of change measure the readiness to change a health behavior. To examine the transition patterns of stages of change for regular exercise over time and to investigate the association between longitudinal pat- terns of SoC and lifestyle intervention outcomes, we analyzed data from a lifestyle intervention program to prevent diabetes among American Indian and Alaska Natives (AI/ANs). Latent class analysis (LCA) was con- ducted to identify the longitudinal patterns of SoC for regular exercise re- ported at the three time points. LCA with distal outcomes was performed to investigate the associations between latent class membership and behavioral changes after the intervention. Traditionally, LCA with distal outcomes were estimated using maximum probability classification fol- lowed by standard regression approach (classical three-step approach). Yet, simulations have shown that this approach could underestimate the associations of interest. Two new methods have been proposed recently: a one-step approach and an improved three-step approach. In the current study, various estimation approaches were used to estimate the param- eters and standard errors for the LCA with distal outcomes models and the results were compared. We identified three latent classes: Pre-action, Transition, and Maintenance classes. The participants in the Transition class moved from pre-action stage at baseline to action or maintenance stage post-intervention. Compared to the other two classes, those in the Transition class had the greatest improvements in physical activity and weight outcomes at both time points post-baseline. C09.5 Joint latent class model for longitudinal data and competing interval-censored events: application to the study of Alzheimer’s disease A Rouanet1 , H Jacqmin-Gadda1 1 INSERM U897 - ISPED, Bordeaux, France Alzheimer´s disease is a chronic illness characterized by a continuous cognitive degradation process and a progressive loss of autonomy. This work aims at developing a descriptive model of the natural history of Alzheimer´s disease, specifically the cognitive decline before diagnosis, considering the competing risk of death. The cognitive decline, measured with repeated psychometric tests, is jointly modeled with the risk of de- mentia in a latent class model.The risk of death is also considered because the population under consideration corresponds to elderly people with high risk of death and most of the risk factors of dementia are associated to death too. Moreover, the cohort data used in these analyses are interval censored be- cause dementia can be diagnosed at visit times only. The exact date of on- set of dementia is thus unknown and a subject who becomes demented and dies between two visits is not diagnosed as demented. To consider both the competing risk of death and interval censoring, it is necessary to use a multi-state Illness-Death model and to calculate the likelihood ac- counting for interval censoring. The transition intensities depend on age but the Dementia-Death transition can possibly depend on the time spent in the demented state, in a semi-Markov model. In this work, we propose an Illness-Death joint model for competing in- terval-censored events and repeated measures of a marker. This model is applied to the French Paquid cohort, which includes 3777 patients, older than 65, followed every 2 or 3 years during 20 years.   C10 Genome-wide association studies C10.1 Entropy-based statistics to detect gene-gene interactions PG Ferrario1,2 1 Universität zu Lübeck, Lübeck, Germany, 2 Universitätsklinikum Schleswig-Holstein, Campus Lübeck, Lübeck, Germany Despite the success of genome-wide association studies (GWAS) in the identification of genetic regions associated with complex diseases, an im- portant proportion of the assumed heritability is as yet unexplained for many traits. To fill this gap, it has been proposed not only to analyse if genes act to a certain phenotype but also if genes interact between each other. In the last ten years different methods have been introduced in order to detect gene-gene interactions. Different approaches from the data min- ing, machine learning, Bayes-statistics and from the multidimensional re- duction (MDR) were applied on genome-wide data and the results have been presented regularly in the literature.

Pages Overview