Research
My primary research focuses on developing methodology for statistical inference with complex longitudinal data in comparative effectiveness research. My areas of methodological interest include causal inference, Bayesian statistics, longitudinal data analysis, measurement errors and bias analysis, and semi-parametric/parametric joint modelling.
A complete list of my publications is available from my Google scholar page.
R packages
- causens: an R package that will allow to perform various sensitivity analysis methods to adjust for unmeasured confounding within the context of causal inference.
- bayesmsm: an R package that implements the Bayesian marginal structrual models to estimate average treatment effect for drawing causal inference with time-varying treatment assignment and confoudning with extension to handle informative right-censoring.
Research themes
1. Methdological research in causal inference with longitudinal data
Bayesian estimation methods that permit causal inference in longitudinal observational studies using administrative databases with the following features, repeated measures, high-dimensional confounding, latent variables, and multiple outcomes.
Ongoing projects under this theme (accepting students):
- Bayesian causal joint and mixture models
- Bayesian transfer learning
- Bayesian confounding learning
2. Design and analysis of observational study
I am interested in studying and applying statistical methods on the design and analysis of clinical and public health studies of rare diseases and chronic conditions. Under this theme, Bayesian inference is an appealing framework: it i) provides a flexible framework for data augmentation and adaptive designs, ii) propagates estimation uncertainty and enables the modelling of latent variables, iii) allow direct probability summaries, and iv) can incorporate prior clinical/expert beliefs.
Ongoing collaborative projects:
- Longitudinal phenotyping and trajectory clustering of multiple repeatedly measured biomarkers in dementia and critical care medicine
- Causal analysis quantifying the impact of school closure during and post pandemic
- Bayesian applications in pediatric and critical care medicine
3. Causal inference methods for randomized controlled trials
Causal inference methods have been applied to traditional RCT data to adjust for non-compliance. Newer trial designs such as pragmatic trials, with a focus on providing timely efficacy evidence, often do not feature complete treatment randomization and thus require causal design and methods to estimate treatment effect. Under this topic, my research interests focus on methods for subgroup analysis including identification of patient subgroups and clinical phenotypes that have differential response to treatment.