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

ISCB2014_abstract_book

130 ISCB 2014 Vienna, Austria • Abstracts - Poster PresentationsWednesday, 27th August 2014 • 15:30-16:00 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust P4.1.134 Bayesian meta-analysis without MCMC KM Rhodes1 , RM Turner1 , D Jackson1 , JPT Higgins2,3 1 MRC Biostatistics Unit, Cambridge, United Kingdom, 2 University of Bristol, Bristol, United Kingdom, 3 University of York, York, United Kingdom Background: Many meta-analyses combine results from only a small number of studies, a situation in which between-study variance is impre- cisely estimated when standard methods are applied. Bayesian meta-anal- ysis allows incorporation of external evidence on heterogeneity, providing the potential for more robust inference on the effect size of interest. Methods: We propose two methods for performing Bayesian meta-analy- sis, using data augmentation and importance sampling techniques. Both methods are implemented in standard statistical software and provide much less complex alternatives to Markov chain Monte Carlo (MCMC) approaches. In a simulation study, we compare the performance of the proposed methods. Results: An importance sampling approach produces almost identical re- sults to standard MCMC approaches, and results obtained through data augmentation are very similar.We compare the methods formally and also apply them to real datasets. For example, a meta-analysis combining four studies evaluating the effec- tiveness of fluoride for lower limb pain is considered. In a conventional random-effects meta-analysis, the between-study variance τ2 is high at 1.78, but very imprecisely estimated (95% CI 0.39 to 52.2). The estimated summary odds ratio is 4.14 (95% CI 0.92 to 18.4). When in- corporating an informative inverse-gamma prior for τ2 using importance sampling, the heterogeneity estimate reduces to 0.54, with 95% credible interval 0.04 to 5.33. The summary odds ratio changes to 3.46 (95% CI 1.17 to 14.3). Conclusion: The proposed methods facilitate Bayesian meta-analysis in a way that is accessible to the applied researchers who are commonly car- rying out meta-analyses. P4.1.135 Individual patient data network meta-analysis with a time-to-event endpoint in head and neck cancer L Ribassin-Majed1 , G Le Teuff1 , J-P Pignon1 , S Michiels1 1 Institut Gustave Roussy, Service de Biostatistique et d’Épidémiologie, Villejuif, France Background: While the use of individual patient data (IPD) is the gold standard for meta-analysis (MA), network meta-analysis typically only uses summary data. We investigated how to extend network meta-analysis to IPD survival data in head and neck cancer. Materials and methods: From a previous IPD-MA comparing radiothera- py (RT) + neoadjuvant chemotherapy (A) or RT + concomitant chemother- apy (B) to RT (reference group) or directly A to B, a network meta-analysis combining direct (6 trials, 861 patients) and indirect (84 trials, 14 317 pa- tients) evidence was constructed, which included two-arm and three-arm trials. The aim was to compare B versus A. We constructed stratified Cox model using both fixed and random effects with and without adjustment for pa- tient covariates. The two random effects for the treatment contrasts are both assumed to follow a ~N(0, σ²) distribution. We propose to study consistency using an interaction test. Results: Adjusted hazard ratios (HR) comparing B to A from direct, indi- rect and mixed treatment comparisons are respectively HR = 0.84 95%CI = [0.67-1.06]; 0.85 [0.79-0.92] and 0.84 [0.78-0.91] with a significant effect for the last two.The corresponding HRs using the approach with 2 random effects gave similar results with small variance components (σ² between 6.82×10-5 and 0.0414). The consistency was respected with an interaction p-value =0.65. Discussion: Our framework allows combining direct and indirect evi- dence in a single Cox model, either with fixed effects or with two random effects. Mixed treatment comparison using IPD survival data allows a bet- ter control of potential confounding bias than summary data.   P4.1.177 Bayesian evidence synthesis and combining randomized and nonrandomized results: a case study in diabetes PE Verde1 , C Ohmann1 1 University of Duesseldorf, Düsseldorf, Germany   There is an increasing interest in combining results from randomized con- trol trials (RCTs) and non-randomized studies in evidence synthesis. One motivation is the generalization of results of RCTs into clinical practice, in particular to a group of patients which may not be included in RCTs for ethical reasons. In this work we present a Bayesian hierarchical meta-regression model for combining results from different study types (randomized and non- randomized) and different data types (aggregated and individual data). Under this model, experimental and observational data are viewed as complementary sources of evidence. The model explicitly includes two types of parameters: those which are the focus of inference (e.g. treatment effect) and those which are used to describe the data collection processes. The data collection processes parameters are used to directly model po- tential sources of bias or inconsistencies between sources of evidence. We illustrate this approach by combining results from recent RCTs, which investigated treatment efficacy of diabetic foot problems and we extrapo- la these results to patients enrolled in a prospective cohort study. P4.1.186 Meta-analysis of continuous outcomes: systematic review of methods available for dealing with missing mean and standard deviation values I Butcher1 , M Brady2 , SC Lewis1 , GD Murray1 , P Langhorne3 , CJ Weir1,4 1 Centre for Population Health Sci., University of Edinburgh, Edinburgh, United Kingdom, 2 NMAHP Research Unit, Glasgow Caledonian University, Glasgow, United Kingdom, 3 Inst. of Cardiovascular & Medical Sci. University of Glasgow, Glasgow, United Kingdom, 4 Edinburgh Health Services Research Unit, Edinburgh, United Kingdom   Background: Omission of mean or standard deviation (SD) data from clin- ical trial reports, perhaps due to the skewed distribution of the outcome, prevents the inclusion of the trial in a meta-analysis, potentially causing bias. Aims: To identify and develop improved methods for handling continu- ous outcomes within meta-analysis, enabling best use to be made of clini- cal trial evidence available. Methods: We investigate how best to impute the mean or SD where either of these has not been reported. In certain circumstances these are suitable summary statistics to analyse in a meta-analysis, regardless of the underly- ing continuous distribution. We report on a comprehensive review of the literature to identify all methods of deriving missing means and standard deviations using electronic resources (including MEDLINE, EMBASE, Web of Science, BioMed Central and The Cochrane Library), relevant journals and grey literature from inception up to March 2014. This updates a previous review (Wiebe 2006) of methods used to deter- mine the SD and extends it to include methods for imputing the mean. Our search focuses on trial-level imputation from non-parametric summa- ries, but also considers algebraic recalculation from parametric summa- ries, trial-level imputation from external data sources or another treatment

Pages Overview