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

ISCB2014_abstract_book

ISCB 2014 Vienna, Austria • Abstracts - Poster Presentations 109Tuesday, 26th August 2014 • 10:30-11:00 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust P2.2.185 Time series analysis of Campylobacter incidence in Switzerland W Wei1 , G Schüpbach2 , L Held1 1 University of Zurich, Zurich, Switzerland, 2 Veterinary Public Health Institute, University of Bern, Bern, Switzerland   Campylobacteriosis is the most common food-associated infectious dis- ease in Switzerland since 1995. Contact with and ingestion of raw or un- dercooked broiler are considered the dominant risk factor for infection. In this study, we investigate the temporal relationship between disease incidence among humans and prevalence of Campylobacter in broiler in Switzerland from 2008 to 2012. We use a time series approach to describe the pattern of the disease by incorporating seasonal effects and autocor- relation. Our analysis shows that prevalence of Campylobacter in broiler, lagged by two weeks, has a significant impact on disease incidence in humans. Therefore Campylobacter cases in human can be partly explained by con- tagion through broiler meat. We also found a strong autoregressive effect among human infections, and a significant increase of infections during Christmas and new year holiday. In a final sensitivity analysis, we corrected for the sampling error of broiler prevalence estimates which gave similar conclusions.   P2.3 Methods for handling missing data P2.3.2 Multiple imputation is not necessary for performing analyses in pre-post studies U Aguirre1,2 , I Arostegui2,3 , JM Quintana1,2 1 Hospital Galdakao-Usansolo, Galdakao-Usansolo, Spain, 2 REDISSEC Health Services Research on Chronic Patients Network, Bilbao, Spain, 3 Applied Mathematics, Statistics Operational Research, UPV/EHU, Bilbao, Spain   Aims: Pre-post studies based on health related quality of life (HRQoL) variables are motivated to determine the potential predictors of the mean change of the outcome of interest. It is very common in such stud- ies for data to be missing, which can bias the results. The use of Multiple Imputation (via Markov Chain Monte Carlo, MCMC) has been increased when handling missing data. However, it has been discussed whether only Complete Case (CC) with mixed models are also effective for this perfor- mance. Methods: We compared CC analysis and MCMC methods to assess their performance for handling missing data under different situations (rate: 10% and 30%; mechanisms: missing completely at random (MCAR), miss- ing at random (MAR), and missing not at random (MNAR)). Moreover, in both cases mixed-models techniques were used. These strategies were applied to a pre-post study of 400 patients with chronic obstructive pul- monary disease (COPD). We analyzed the relationship of the changes in subjects’HRQoL over one year with clinical and sociodemographic charac- teristics. A simulation study was performed (500 and 1000 runs), where the standardized bias of the regression coefficient of the interaction between the Time effect and the covariate was computed. Results: In both 500 and 1000 simulation-runs, CC with mixed models showed the lowest standardized bias coefficients for MCAR and MAR sce- narios. However, in MNAR setting, both approaches provided biased coef- ficients. Conclusions: MCMC has not additional benefit over CC when handling missing data for MCAR and MAR settings. There is no consensus in MNAR scenario.   P2.3.44 A new method for significance testing of categorical covariates after multiple imputation I Eekhout1,2 , MA van de Wiel1,3 , MW Heymans1,2,3 1 VU University Medical Center, Amsterdam, The Netherlands, 2 EMGO Institute for Health and Care Research, Amsterdam, The Netherlands, 3 VU University, Amsterdam, The Netherlands   In medical prognostic research, logistic regression analysis is frequently used. Unfortunately, missing data is common in these studies. As a solu- tion multiple imputation is recommended, which generates multiple im- puted datasets. Subsequently, logistic regression models are applied in each imputed dataset and finally parameter estimates are pooled using Rubin’s Rules (RR). For significance testing of dichotomous and continuous covariates in these models, RR can easily be applied. However, to consider whether a categorical covariate as a whole signifi- cantly contributes to the model, RR cannot be used. Instead, to obtain an overall p-value, Meng and Rubin (MR) proposed to pool the log likelihood ratio test statistics for each parameter and obtain the significance level from that pooled statistic. This procedure is complicated and not available in standard statistical software. We propose a new method which is much easier to use with power at least equal to that of the MR method. Our method uses the median of the p- values of all separate likelihood ratio tests in each imputed dataset: the Median P-Rule (MPR). In a large simulation study, it was shown that for non-significant categorical covariates the type I error is controlled and the statistical power of the MPR was at least equal to that of the MR method for significant ones. An illustrative empirical data example showed similar results. We recommend using the median of the p-values from the imputed data analyses (MPR). This method performs at least equally well as the MR method, but is much easier to apply. P2.3.137 Missing categorical data: the influence of imputation technique on regression analysis in an opioid maintenance treatment setting M Riksheim1 , J Røislien1,2 1 Norwegian Centre for Addiction Research, University of Oslo, Oslo, Norway, 2 Department of Biostatistics, University of Oslo, Oslo, Norway   Missing data is a recurring topic in observational studies and can be deci- sive in some settings. In Opioid Maintenance Treatment (OMT) mainly one of two substitution medications is used to treat opioid dependence and best practice is discussed. Studies on the OMT population are important to optimize treatment, but with OMT patients being hard to reach, the issue of missing data is non-negligible. For missing continuous data, several imputation methods have been pro- posed and extensively researched. For categorical data, however, no clear recommendations exist. In this methodological study we applied four dif- ferent imputation techniques on missing categorical data to explore the influence of method choice on results in a subsequent regression analysis on data from the Norwegian OMT programme. In the Norwegian OMT programme questionnaire data regarding patients’ treatment status are collected annually. For the present study, we used data from the eastern region of Norway collected 2005-2010. The data comprised of 9039 questionnaires with 12 questions from 2886 patients. Missing ranged from 0% to 10% per question. Four missing data techniques were tested: Expectation Maximization with Bootstrapping; Multiple Imputations by Chained Equations; Hot Deck Imputation, and Multiple Imputation using Latent Class. The imputed data sets were tested in logistic regression analyses with type of medication as outcome and 11 covariates, including measures of social situation, age and sex. The imputation methods gave differences in both

Pages Overview