GUANBO WANG
  • Home
  • Research
  • Positions and collaboration
  • Miscellaneous
I STUDY CAUSAL INFERENCE

THEORY

Semi-parametric statistics

Survival analysis

​Machine learning

Optimization/High-dimensional statistics

TOPICS

Causal inference

Data Integration

Personalized medicine

Interpretable variable selection

DATA

(Multi-center) clinical trials

Electronic health records (hospitals/claims/admin)

​The combination of above

​Longitudinal


DISEASE

Heart

Pediatric

Cancer

​Mental, etc.

Papers and softwares
Google scholar
Awards/Grants
Collaborators
Topics
GitHub

Glossary:
  • ATE: average treatment effect
  • HTE/CATE: heterogeneous (conditional average) treatment effect
  • Effect modifier: the factor can modify the treatment effect in different population
  • Common transportability condition: the outcome is independent of population source conditional on covariates. I.e., after adjusting for covariate, the outcomes in different populations have the same distribution.
  • Non-parametric models: machine learning models that put no restriction on the data generating mechanism
  • Green text: links to papers
Picture
1. DATA INTEGRATION
1.1 Robust trial augmentation: trials have limited efficiency due to their small sample sizes. We solve the challenge by integrating external data. 
  • Study design and identification for various scenarios under common transportability conditions
  • Hybrid control in early-phase trials
  • Methods that using parametric, non-parametric, and foundation models that can ensure (asymptotically) unbiased estimation of the ATE of the trial, even if the common transportability condition fails and/or the models are misspecified

​1.2 Transportability: conclusions drawn from a trial are applicable only to its participants, while decision-makers often seek evidence relevant to a broader or specific population.
  • ​A scoping review
  • A method for transportability that assumes a more realistic condition---relative effect measures (w.g., risk ratio) are transportable​
​
1.3 Meta-analysis
  • Causally-interpretable (network) meta-analysis to estimate the ATE, add-on effects, HTE, and subgroup effects (with R package and paper) when patients should take a regime from multiple treatments and patients in different populations have different access​ to the treatments

2. PERSONALIZED MEDICINE
  • Effect score analyses: a method to estimate HTE of a trial, identify exceptional responders and optimal treatment strategies. It has valid statistical inference and is highly interpretable and clinically applicable even if with the use of any generic machine learning methods.
  • HTE in longitudinal high-dimensional data
  • Estimating subgroup and HTE in meta analysis
Picture
Picture
Figures by Wang et al. (JAMA) and Clemson University Open Textbooks
3. VARIABLE SELECTION
  • Prior knowledge integration: incorporating clinician's knowledge to construct coherent and interpretable prediction models and identify clinically meaningful predictors. General framework; guide of using the overlapping group Lasso to achieve the goal for non-survival outcomes; for survival outcomes; R package.
  • Confounder and effect modifier selection
  • Applications
Picture
Figure by ​spot intelligence
Picture
Arriving at an interpretable predictive mechanism and a coherent prediction model with significant predictors.
Powered by Create your own unique website with customizable templates.
  • Home
  • Research
  • Positions and collaboration
  • Miscellaneous