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

ISCB2014_abstract_book

34 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsMonday, 25th August 2014 • 16:00-17:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust delayed entry through a simulation study. To maintain flexibility in model- ling the breast cancer data, we use restricted cubic splines to model the baseline hazard function, and model the longitudinal trajectory using fractional polynomials. User friendly Stata software is provided.   C15.3 Dynamic time process model for the association among two longitudinal markers in the presence of survival: application to healthABC cohort D Geva1 , DR Shahar1 , TB Harris2 , S Tepper1 , G Molenberghs3 , M Friger1 1 Ben Gurion University, Beer Sheeba, Israel, 2 Laboratory of Epidemiology, Demography, and Biometry, Bethesda, United States, 3 Katholieke Universiteit Leuven, Leuven, Belgium Background: Further expansion of joint models to include several mark- ers led to the formulation of the joint model with multiple longitudinal- markers. Due to the integral over the random effects, estimation of such models is complex. Thus, in practice a model to study the impact of an ad- ditional marker on the joint longitudinal-survival outcome is lacking. The aim of our work is to offer an appropriate model to study the impact of a single marker measured repeatedly over time on longitudinal outcome in the present of survival. Methods and Results: Founded on the shared parameter joint model (SREM), we propose the Dynamic Time Process Model(DTPM) as follows: For each time cut-points t=1,2,3…T obtain the trajectory of the marker by het- erogynous latent class mixed model for the period up to t. Then, use the trajectory class in the SREM for time period beyond t.The model result in a T frames dynamically capturing the marker trajectory up to t and the joint longitudinal-survival projection beyond t. We wrote an R function, dtpm() and illustrate the model on the Health- ABC cohort data and again explore its boundaries using simulation. Remark: The proposed model provides an elegant analytical framework to study the impact of marker- trajectories on longitudinal outcome in the presence of survival.The unique feature is the dynamic progression before and after time cut-point. Although, the proposed model does not have a natural expansion to multiple outcomes, it is valuable for testing new hypotheses for the joint survival-longitudinal setting. C15.4 Student Conference Award Combined dynamic predictions using joint models of multiple longitudinal outcomes and competing risk data E-R Andrinopoulou1 , D Rizopoulos1 , JJM Takkenberg1 , E Lesaffre1 1 Erasmus Medical Center, Rotterdam, The Netherlands Nowadays there is increased medical interest in personalized medicine thereby tailoring decision making to the needs of the individual patient. In this work we focus on the statistical methodology for providing sub- ject-specific predictions of survival probabilities used in this context. Our developments find their motivation in a Dutch study at the cardiotho- racic department of the Erasmus Medical Center. It is our aim to utilize the available follow-up measurements of the current patients to predict both survival and freedom from re-intervention for future patients. Since the human tissue has limited durability due to calcification and degen- eration resulting in valvular dysfunction, it is of interest to physicians to predict the need for future re-interventions using all available repeated echo measurements. To analyze the data and obtain subject-specific risk predictions we use the joint modeling framework. In this work we extend the concept of predic- tion to multiple longitudinal combined with competing risk survival out- comes and we derive the dynamically updated cumulative incidence func- tions. Moreover, we investigate whether different features of the longitu- dinal processes would change significantly the prediction for the events of interest. Our final contribution focuses on optimizing the quality of the derived predictions. In particular, instead of choosing one final model over a list of candidate models, we propose to suitably combine predictions from all considered models using Bayesian model averaging (BMA). The advantage of using BMA in this setting is that predictions are tailored to each individual patient because the model weights are both subject- and time-dependent.   C15.5 Development and validation of individualized dynamic predictions based on repeated biomarker data according to scenarios of new treatments M Sène1 , JM Taylor2 , H Jacqmin-Gadda1 , C Proust-Lima1 1 INSERM U897, Bordeaux, France, 2 University of Michigan, Ann Arbor, United States With the emergence of rich information on biomarkers after treatments, new types of prognostic tools are being developed: dynamic prognostic tools that can be updated at each new biomarker measurement. Such pre- dictions are of particular interest in oncology where after an initial treat- ment, patients are monitored with repeated biomarker data. However, in such setting, patients may receive second treatments to slow down the progression of the disease, which is not currently handled by dynamic predictive tools. This paper aims to develop and validate dynamic individual predictions that allow the possibility of a new treatment in order to help understand the benefit of initiating new treatments during the monitoring period.The prediction of the event in the next x years is done under two scenarios: (1) the patient initiates immediately a second treatment, (2) the patient does not initiate any treatment in the next x years. The dynamic predictions are derived from joint (shared random-effect) models. The predictive accuracy of the dynamic predictions is evaluated with two measures (the Brier score and the prognostic information) for which approximated cross-validated estimators that correct the usual over-optimistic predictive performances are proposed. Applied to the monitoring data of more than 2300 men initially treated by radiation therapy for a localized prostate cancer, different specifications for the dependence between the PSA repeated measures, the initiation of a second treatment (hormonal therapy) and the risk of clinical recurrence are investigated and compared.   C16 High-dimensional data analysis II C16.1 A comparison of prediction models for gene expression data by resampling techniques S Yilmaz Isikhan1 , E Karabulut1 , CR Alpar1 1 Hacettepe University, Ankara, Turkey Gene set methods aim to assess the overall evidence of association of a set of genes with a phenotype, such as dose values or a quantitative trait. The purpose of this study is to compare the effects of such frequently used methods of generalization as Bootstrap, Cross Validation on some machine learning regression methods. To assess the Root Mean Squared Error and Coefficient of Determination performance of these generalization methods for different regression methods, an extensive simulation study was completed in which the scenarios varied according to: sample size, number of genes associated with the phenotype, regression coefficients, correlation between expres- sion of genes within a gene set and the original correlation structure of

Pages Overview