Data analysis and methodologies
We aim to improve analysis methods in epidemiology.
We do this by:
- Developing new methods when necessary, including:
- developing rules for stopping a trial early when a treatment has been shown to be futile;
- developing methods to reduce confounding by unmeasured variables by using incorporating data from multiple sources;
- exploring modern methods of identifying patterns in temporally rich, novel and complex data using machine learning and text-mining;
- developing methods to provide evidence that a given propensity score will balance all measured covariates to an acceptable level;
- examining reweighting methods to estimate expected treatment effects in target populations different from the one in which the treatment effect has been measured.
- Implementing existing methods in a more user friendly manner to encourage wider use, including:
- weighted cumulative exposure models;
- missing cause approach to controlling for unmeasured confounding;
- balance checking following the application of propensity score methods.