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

ISCB2014_abstract_book

14 ISCB 2014 Vienna, Austria • Abstracts - Oral PresentationsMonday, 25th August 2014 • 9:00-10:30 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust treatment arm than in the sequential treatment arm. This is an encourag- ing finding, since it means that we can recommend integrated treatment, which reduces mortality, without the risk of worsening HIV outcomes.   C02.4 Proposition of a multiple imputation approach for MNAR mechanism using Heckman’s model J-E Galimard1 , M Resche-Rigon1 , S Chevret1 1 INSERM UMR 1153, Université Paris 7, Paris, France Standard implementations of multiple imputation (MI) approaches provide unbiased inferences assuming underlying MAR mechanisms. However, in presence of missing data generated though MNAR mecha- nisms, although the MAR assumptions can be approached by collecting more explanatory variables, MI is not satisfactory and difficult to validate. Coming from econometric statistics, the Heckman’s method, also called the sample selection method, deals with selected sample using two joined linear equations, namely the selection equation and the outcome equation, respectively. The Heckman´s method has been successfully applied to missing outcomes in presence of MNAR mechanism. Nevertheless, such a method deals with missing outcomes only, and this is a strong limitation in clinical epidemiology settings where covariates are also often missing. We propose to extend the validity of MI to MNAR mechanisms by using the Heckman´s model as the imputation model, using a two-step estimation process. This will provide a solution that can be used in a MI by chained equation framework to impute missing variables (outcomes or covariates) resulting either from a MNAR or a MAR mechanism. This approach will be validated by a simulation study.We will first consider MNAR missing outcomes, and evaluate the robustness of the approach in case of model miss-specification. Then it will be evaluated in presence of missing outcome and missing covariates either under MNAR or MAR mechanisms. C02.5 Allowing for nonignorable missingness in HIV status using multiple imputation with delta- adjustment: applications to causal mediation analysis and prevalence estimation FP Leacy1,2 , IR White1 , TA Yates3 , S Floyd4 , HM Ayles4,5 , P Godfrey-Faussett4 1 MRC Biostatistics Unit, Cambridge, United Kingdom, 2 University of Cambridge, Cambridge, United Kingdom, 3 University College London, London, United Kingdom, 4 London School of Hygiene & Tropical Medicine, London, United Kingdom, 5 ZAMBART Project, University of Zambia, Lusaka, Zambia Aims: The primary goal of this analysis is to assess the mediating influence of HIV status on the relationship between socioeconomic status and prev- alent tuberculosis in Zambia, allowing for nonignorable missingness in the HIV test result variable. Anticipated sensitivity of HIV prevalence esti- mates to subgroup-specific departures from MAR will also be investigated. Methods: Utilising data from the Zambia South Africa Tuberculosis and AIDS Reduction (ZAMSTAR) trial, we first present results from a complete case causal mediation analysis. We then analyse the data using multiple imputation under MAR, comparing results from two imputation models that differ only with regard to inclusion of information on self-reported HIV status.We proceed to perform a sensitivity analysis to departures from MAR using multiple imputation with delta-adjustment. We extend the standard procedure by allowing the value of the adjustment, δ, to vary across subgroups defined by participants‘ self-reported HIV status. Results: While estimates of conditional natural direct effects, conditional natural indirect effects and conditional total effects on the odds ratio scale exhibited significant sensitivity across the investigated δ range, estimates of the corresponding marginal quantities were relatively insensitive to departures from MAR. Subtle variations in HIV prevalence estimates by gender were observed across the range of missingness mechanisms con- sidered. Conclusion: Multiple imputation with delta-adjustment offers a transpar- ent and flexible means to allow for nonignorable missingness in HIV sta- tus.This method may represent a particularly important tool for sensitivity analysis in contexts such as mediation analysis where multiple subcompo- nent models must be fitted to the data.   C03 Regression modelling in epidemiology C03.1 Interpretation of linear regression coefficients under mean model misspecifications W Brannath1 , M Scharpenberg1 1 University of Bremen, Faculty 3, Bremen, Germany Linear regression is an important and frequently used tool in medical and epidemiological research. However, its validity and interpretability relies on strong model assumptions. While robust estimates of the coefficients’ covariance matrix extends the validity of hypothesis tests and confidence intervals, a clear and simple interpretation of the regression coefficients is lacking when the mean structure of the model is misspecified. To over- come this deficiency, we suggest a new mathematical rigorous interpreta- tion that is independent from specific model assumptions. The idea is to quantify how much the (unconditional) mean of the dependent variable Y can be changed by changing the distribution of an independent vari- able X in the population. We show that with a suitable standardization of the distributional changes, the maximum change in the mean of Y is well defined and equals zero if and only if the conditional mean of Y given X is independent of X. Restriction to linear functions for the distributional changes in X provides the link to linear regression. It leads to a conserva- tive approximation of the newly defined and generally non-linear mea- sure of association. The conservative linear approximation can then be estimated by linear regression. We show how the new interpretation can be extended to multiple regression analysis with the goal of adjusting for confounding covariates. We illustrate the utility (and limitations) of the new interpretation and point to perspectives for new analysis strategies. C03.2 Simulation study to assess and compare strategies for modelling two continuous covariates with a spike at zero C Jenkner1 , E Lorenz2 , H Becher2 , W Sauerbrei1 1 University Medical Center Freiburg, Freiburg, Germany, 2 University of Heidelberg, Institute of Public Health, Heidelberg, Germany In epidemiology and clinical research, a common goal is to estimate the effect of predictors on an outcome using appropriate regression models. Such predictors often consist of an amount of individuals with a value of zero while the distribution of the remaining ones is continuous (variables with a spike at zero). Examples in epidemiology are smoking or alcohol consumption. Since the risk for a certain disease may be substantially dif- ferent between unexposed and exposed individuals, it is important to al- low a separate estimate for the unexposed and a continuous function for those exposed. A strategy for the univariate case was proposed in Royston et al, Becher et al. For a logistic regression model, theoretical odds ratio functions for selected bivariate distributions were calculated (Lorenz et al., submitted). Four possible methods how to include information of the zero values in the bivariate case using fractional polynomials were proposed.

Pages Overview