A A A

UCL Partners Biostatistics Network Seminars  - Term 2, 2012

All seminars will be held in Room 102, Department of Statistical Science, 1-19 Torrington Place (1st floor)


Tuesday 14th February (4-5pm)

John Wood (University of East Anglia)
Overdispersion in large datasets

Increasingly, information about healthcare is becoming available in the form of large datasets – for instance on inpatient hospital episodes available from the NHS Information Centre. Such datasets will often comprise millions of observations, and it is fortunate that the computing power needed to handle them is now also fairly routinely available. Clearly, these developments can only be a good thing, but they do bring with them new challenges in statistical analysis. In particular, whilst our ability to model such data is often surprisingly good, the likelihood is that there will always be significant unexplained excess variation which – particularly in view of the size of the datasets – will not be ignorable. A natural approach to this problem is to model it as overdispersion. This can be done in a number of ways, some of which will be discussed in the context of recent examples of analyses of mortality data.


Tuesday 13th March (4-5pm)

Gianluca Baio (UCL)
Bayesian hierarchical models - recent computational advances and applications to pre-implantation genetic screening in IVF

Bayesian hierarchical models are an effective way of accounting for complex structures in the data, including clustering and nested levels of information. Typically, within a Bayesian framework, hierarchical models are estimated using Markov Chain Monte Carlo methods. These are standard in Bayesian analysis but, while generally very efficient, they can be extremely computationally intensive, especially for hierarchical models. Recently, alternative methods have been investigated to increase the computational efficiency and the precision of the estimations. In this talk, we review the theory behind Integrated Nested Laplace Approximation; in particular, we show an example based on data obtained at the UCL Centre for Pre-implantation Genetic Diagnosis, investigating the effect of the length of telomeres (a 6 base repeat found at the end of chromosomes to protect them from degradation) on chromosomal abnormalities in IVF.


Tuesday 24th April (4-5pm)

Jonathan Bartlett (LHSTM)
Accommodating the model of interest within the fully conditional specification multiple imputation framework

Missing covariate data is a common issue in epidemiological and clinical research, and is often dealt with using multiple imputation (MI). When the model of interest is non-linear, or contains non-linear (e.g. squared) or interaction terms, this complicates the imputation of covariates. Standard software implementation of MI may impute covariates from models that are uncongenial with such models of interest. We show how imputation by full conditional specification, a popular approach for performing MI, can be modified so that covariates are imputed from a model compatible with the model of interest. We investigate through simulations the performance of this proposal, and compare it to existing approaches. Simulation results suggest our proposal gives consistent estimates for a range of common models of interest, including models which contains non-linear covariate effects or interactions, provided data are missing at random and the assumed imputation models are correctly specified. In contrast, passive imputation of non-linear or interaction terms generally results in inconsistent estimates of the parameters of the model of interest, while a recently proposed alternative approach, based on treating such terms as 'just another variable' gives consistent results only for linear models and only if data are missing completely at random. 


Tuesday 15th May (4-5pm)

Deborah Stocken (University of Birmingham)
Title to be confirmed - topic: fractional polynomials and restricted cubic splines 



Page last modified on 24 may 12 14:35