Big Data Lunchtime Seminar: Are predictions from test accuracy meta-analyses valid in practice? Experiences from simulating data in R when developing a validation statistic

Location
B01, School of Mechanical Engineering
Dates
Wednesday 17 February 2016 (13:00-14:00)
Contact

If you would like to attend and join the Institute of Advanced Studies Big Data network which brings together researchers and professional services from across the campus please email Sarah Jeffery.

BrianWillis-Cropped-110x146

Dr Brian H. Willis (Institute of Applied Health Research) presents this lunchtime seminar as part of the Institute of Advanced Studies' Big Data Network.

Evidence based medicine is at the heart of good clinical decision-making and relies upon high quality research. A powerful method used by researchers to combine multiple studies in clinical research is that of meta-analysis and here I will consider the example of meta-analyses of diagnostic test studies which provide summary estimates of a test’s accuracy.

The problem faced by clinicians when applying such meta-analyses to their own practice is that it is difficult to ascertain whether the summary estimates would be valid (accurately represent) their own clinical practice. This problem provides the motivation for developing a validation statistic which is able to measure the likely validity of test research in practice.

Dr Willis will provide the background to this problem and propose a validation statistic with its exact distribution. He will discuss the evaluation of the validation statistic using statistical simulation trials using R. Predominantly the problems faced when coding such simulations involve extensive use of Ioops and in this case, processing matrices. He will discuss the problems faced when using loops and manipulating matrices in R, and consider the options for improving the time efficiency when executing large scale simulations.