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

ISCB2014_abstract_book

132 ISCB 2014 Vienna, Austria • Abstracts - Poster PresentationsWednesday, 27th August 2014 • 15:30-16:00 Monday25thAugustTuesday26thAugustThursday28thAugustAuthorIndexPostersWednesday27thAugustSunday24thAugust P4.2.102 Determining the risk of Neisseria gonorrhoeae infection by meeting location among men who have sex with men in Amsterdam A Matser1,2 , R Geskus1,3 , M Schim van der Loeff1,3 1 Public Health Service of Amsterdam, Amsterdam, The Netherlands, 2 University Medical Center Utrecht, Utrecht, The Netherlands, 3 Academic Medical Center of Amsterdam, Amsterdam, The Netherlands   Introduction: In individuals with concurrent sexual partnerships, the source of Neisseria gonorrhoeae (NG) infection is usually unknown. For di- rected prevention measures, knowing the infection risk per meeting loca- tion is useful. Methods: In 2008-2009, we collected information from 3034 men who have sex with men at the Public Health Service of Amsterdam. For up to four partners per participant (n=8032), meeting location, partner and partnership characteristics were asked. For NG-infected participants, we assumed that one of the reported partners was the source. We used lo- gistic regression to relate infection risk per partnership to participant and partner characteristics and meeting location. In the likelihood we accounted for partially unobserved transmission status per partnership. We predicted the partners’probability of having NG based on this model, replacing index characteristics by partner characteristics. Each partner- ship was assumed to be the source based on the relative size of the prob- abilities. Each source was linked to a meeting location. We used a Bayesian approach. Results: We considered 16 meeting locations; two (A and B) were streets in Amster-dam with gay venues. When the infection risk was equally al- located over the partnerships, the NG probabilities were 0.085 at A and 0.028 at B.The multivariate model predicted 0.019 (95% CI 0.017-0.021) for A and 0.056 (95% CI 0.052-0.060) for B. Discussion: Modelled NG probabilities differed from equal allocation results. NG typing data can be included, which may improve accuracy of estimates. The model can be applied to other data with partially unknown source of infection.   P4.3 Novel designs and methods for simulations P4.3.13 Bi-cross-validation for the choice of optimal number of non-zero loadings in sparse PCA methods N Assi1,2 , P Ferrari1 , V Viallon2,3 1 International Agency for Research on Cancer, Lyon, France, 2 Université Claude Bernard Lyon I, Lyon, France, 3 UMRESTTE, Bron, France   Recent developments in data analysis and dimension reduction intro- duced a variety of sparse methods to overcome problems related to high dimensionality. While PCA is a tool of choice for such analyses, factor load- ings related to its components may be difficult to interpret. Alternate tech- niques allow yielding eigenvectors/components with sparse loadings i.e. with a few non-zero elements. These methods require the user to give the number of non-zero elements to appear on each eigenvector. However the cardinality is not known a priori, and while some methods have a simple cross-validation routine included, the usual approach to ob- tain this parameter is prone to a level of subjectivity often found by trial and error. Inspired by the work of Owen and Perry on unsupervised cross-validation for singular value decomposition, non-negative matrix factorization and k-means, we devised a bi-cross-validation algorithm to help in deciding the optimal cardinality to use in these methods. A bi-cross-validation for 3 different sparse eigenvalue problems was developed, including: Sparse PCA using the elastic net penalty (Zou, Hastie, Tibshirani, 2006), SPC, that uses a penalised matrix decomposition (Witten, Tibshirani, Hastie, 2009) and Truncated Power Method (Yuan and Zhang, 2011). The performance of this algorithm in terms of successfully recovering the underlying sparse covariance structure of the data, together with the reproducibility of re- sults, was evaluated with a simulation study. The bi-cross-validation was applied to nutritional data and application to metabolomics is planned. Developments to make the procedure help identify the optimal number of components to retain are currently on-going.   P4.3.70 Calculating sample size for cluster-randomised trials with mid-point sample size re-assessment R Hooper1 1 Queen Mary University of London, London, United Kingdom   Sample size re-assessment in a clinical trial involves estimating nuisance parameters at an interim analysis, and using them to re-calculate sample size. This can rescue an initially underpowered trial. Though interim esti- mation biases the significance level of a naive statistical analysis, an unbi- ased test can be achieved by combining p-values from the two stages of the trial using methods for adaptive trials. In individually randomised trials it is known that little bias arises from a naive analysis if the interim analysis includes 20 individuals or more, but adaptive methods could be particularly useful for cluster-randomised tri- als, where two nuisance parameters must be estimated at interim from (perhaps) a small number of clusters. In the latter case it may make sense to optimise the information available at the interim analysis by planning it to fall half-way through recruitment. I describe how to choose a sample size re-assessment design with given power under initial assumptions about nuisance parameters, where the interim analysis is planned to occur after recruitment reaches its expected mid-point. Sample size in this case cannot be calculated analytically, but may be determined by simulation (I will demonstrate using Stata). Important considerations when deciding whether to use a sample size re- assessment design are the variability in actual sample size, and the extent to which power is successfully controlled at the desired level under depar- tures from initial assumptions. These aspects of performance should also be simulated before deciding against the use of a simpler, fixed sample size trial design.   P4.4 Observational studies and causal inference methods P4.4.14 Prevalence of internet use and Internet addiction disorder among medical students: a case from low income country S Awan1 1 Aga Khan University, Karachi, Pakistan   Background: Prolonged use of internet caused a series of problems such as internet dependence, problematic internet use, compulsive internet use, internet abuse and Internet addiction disorder (IAD), This has aroused attentions of researchers all over the world, with IAD being recognized as a mental disorder. The aim of study is to investigate the prevalence and risk of Internet addic- tion disorder (IAD) and the associated factors amongst medical students

Pages Overview