I am a statistician working in the pharmaceutical industry. Interests include Bayesian statistics, graphics, modelling, prediction particularly in the clinical realm.
MSc in Medical Statistics, 2004
London School of Hygiene and Tropical Medicine
R Shiny application examples and reproducible analyses using R Markdown. Click on the blue menu ‘R Shiny’ button to see the Shiny apps. The ‘Other’ option displays all non-R shiny analyses. Explore the Git repository for more…
Random variation is very likely to lead to observing responders and non responders in a parallel RCT
Power simulation for one way analysis of variance
More informative presentation of diagnostic test data
Boxplots can hide a lot of information
Nothing more, nothing less
Simulating allows us to play God. We know the true data generating mechanism.
Simulating allows us to play God. We know the true data generating mechanism.
Why analyse variances (ANOVA) when interest is in differences in population means?
Getting to grips with power
‘Crossover-Trial-Simulation’ and ‘Crossover-Trial-Simulation-Power’
The analysis of an RCT based on a continuous endpoint that is log-normal distributed
Using R we explore the ‘FDA PBE Statistical Analysis Procedure Used in Bioequivalence Determination of Budesonide’
A game where gamblers simultaneously flip a single coin…
Estimating treatment effect with patients followed over time.
A simulation of a RCT demonstrating why it is wrong to analyse arms of a trial separately to identify responders and non responders.
The biggest coincidence of all would be if there were no coincidences.
Plotting longitudinal data and estimating means with the help of a natural log transformation
Investigation of adjusted and unadjusted power in the setting of RCT when there is a predictive biomarker
Simulating balanced or unbalanced nested study design, plotting and a variance components analysis
Simulating a 3 level nested study design, plotting and variance components analysis
Investigating a crossover trial, a longitudinal study in which subjects receive a sequence of different treatments
Two data points may convey more information than you think
Sample size/power for an RCT based on a continuous biomarker endpoint that is log-normal distributed