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


ISCB 2014 Vienna, Austria • Abstracts - Poster Presentations 115Wednesday, 27th August 2014 • 11:00-11:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust variate linear mixed effect model. By including the marker values as latent terms in the event sub-model, both sub-models shared the random ef- fects of the markers. Because we had random effects for all four biomarkers and for all nine event types, our joint model involved a high dimensional integral. In this case, quadrature approximations are too computationally expensive. As a solution, a Monte-Carlo approach was used to evaluate the integral. To improve the accuracy of the approximation, we used a Quasi-Monte Carlo (QMC) approach with a deterministic point set based on scrambled Sobol sequences. With simulations, we showed that the joint model with QMC integration gave accurate estimates. Moreover, the joint model successfully captured informative drop-out. We fitted the joint model on our post-kidney trans- plantation data. P3.1.75 Analyzing clinical pathways in observational studies: pitfalls and approaches M Huebner1 , DW Larson2 1 Michigan State University, East Lansing, United States, 2 Mayo Clinic, Rochester, United States   An enhanced recovery pathway in colorectal surgery consists of a se- quence of treatments in preoperative, intraoperative, and postoperative phases before discharge from the hospital. It is of interest to determine key pathway elements that are associated with short-term outcomes such as length of hospital stay. Studying the individual association of elemental compliance with the outcome is not sufficient to establish their importance, since other fac- tors impact the compliance or have an effect on the outcome. For exam- ple, occurrence of complications may lead to a longer length of stay and may modify the postoperative pathway. Preoperative and intraoperative elements may be associated with a decreased need for opiates and thus lead to a faster recovery after surgery. Comparisons of research reports between institutions are difficult due to heterogeneity of patient popula- tions and ignoring confounding factors in the analyses. We will present some approaches including multistate models that have been helpful in shedding some light on these issues. P3.1.111 Remedy for‘IntCox’in partly interval-censored survival data M Nishikawa1 , H Mizukami2 , T Morikawa3 , T Yokoyama1 1 National Institute of Public Health, Wako,Saitama, Japan, 2 Sanofi K.K., Shinnjyuku, Japan, 3 Kurume University (formerly), Kurume, Japan   In survival analysis, common data type are right-censored. However, we often encounter “partly interval-censored data” (PICD) in oncology stud- ies where observed data include both exact times of event and interval- censored, for example progression free survival (PFS). PFS is defined as the time from randomization to the date of disease progression or death, whichever is earlier. The most common regression model in survival analysis is proportional hazard model. It is also called Cox model for right-censored data. Several interval-censored data regression analyses have been proposed (e.g. Finkelstein(1986), Pan(1999)). Among them, Pan(1999)’s method is imple- mented in the R package‘IntCox’as an extension of Cox model. Chen et al. (2012) report performance of‘IntCox’for interval-censored data. However, its performance for PICD is unknown. ‘IntCox’ cannot work for PICD. We found that some data manipulation (jittering) is necessary be- fore calling ‘IntCox’, and incomplete result of ‘IntCox’(no improvement of likelihood possible) is frequently observed. We consider some remedies for ‘IntCox’ in the analysis of PICD; these are deterministic imputations and bootstrap methods. In this talk, we compare the performances and properties of the three de- terministic imputation methods (the right-point, the mid-point and the left-point of the censoring interval), and ‘IntCox’ with/without bootstrap for PICD by simulation study of which design was based on actual clinical trials. Our simulation suggests that the left-point imputation may be generally better in MSE than the right-point or the mid-point imputations if there are right-censored data before the planned end-of-the-study. The details will be shown in the presentation. P3.1.125 Comparison of survival between allogeneic haematopoietic stem cell transplantation and continued drug treatment when differentiating between risk groups at diagnosis M Pfirrmann1 , R Hehlmann2 , M Lauseker1 , J Hasford1 1 IBE, Ludwig-Maximilians-Universität München, München, Germany, 2 III. Medizinische Klinik Mannheim, Universität Heidelberg, Mannheim, Germany Aims: In the German study IIIA, patients with chronic myeloid leukaemia providing consent and eligibility for allogeneic haematopoietic stem cell transplantation (HSCT) were randomized in accordance with the avail- ability of a (matched) related donor to receive either HSCT or continued drug treatment. Primary endpoint was overall survival (OS) from diagno- sis. Subgroups with different survival risks after transplantation were of particular interest. Methods: At transplantation, survival probabilities can be differenti- ated through subgroups defined by the prognostic factors age, phase of disease, time to transplantation, donor matching, and recipient-donor sex combination. At diagnosis, patients were in chronic phase of dis- ease and age known. It was assumed that for all patients, transplanta- tion was planned in the first year. For non-transplanted patients, values for the factors “donor matching” and “recipient-donor sex combination” were randomly drawn from their actual distribution among the patients transplanted. This was repeated to gain 999 results of a log-rank statistic for each subgroup. Treatment comparisons were in accordance with ran- domisation. Results: Of 166 patients randomized to HSCT, 151 were transplanted. Here, donor matching and recipient-donor sex combination were as- sumed to have been known at diagnosis but were randomly drawn for the 15 remaining patients and all 261 patients randomized for drug treatment. Median p-values for the 999 OS comparisons of HSCT versus drug treat- ment were not significant for any of the subgroups. Conclusions: This innovative approach prevents time-to-transplantation and selection bias. To decide for the option “HSCT” at diagnosis depicts reality. The issue of statistical power needs to be discussed.   P3.1.148 Modeling cause-specific survival in cancer patients compared to the general population, a large population based study MC Småstuen1 , DA Barzenje2 , A Kolstad3 , H Holte3 1 University of Oslo, Department of Biostatistics, Oslo, Norway, 2 Department of Oncology, Ostfold Hospital Trust, Oslo, Norway, 3 Department of Oncology, Oslo University Hospital, Oslo, Norway Aims: Competing risk is a recognised methodology however not often used and correctly interpreted by clinicians. Kaplan Meier approach is still the method of choice even though it might result in wrong and mis- leading results. Cause-specific survival is often cited and compared in the medical literature however in many studies calculated using a wrong methodology.

Pages Overview