Biostatistics In-Person Seminar:

Event Description

Presenter:

Joseph Hogan, Ph.D.
Professor and Chair of Biostatistics
Brown University, School of Public Health

Abstract:

In the analysis of large scale observational data, such as from electronic health records of surveillance systems, the difference between predictive and causal inference is not always appreciated. For example, if seriously ill patients are selectively given a new treatment, it's possible that receiving the treatment can be predictive of a poor outcome even though the causal effect of receiving that treatment prevents poor outcomes. It's also the case that many methods for prediction are based more on algorithms than on proper statistical models.
In this talk we illustrate the use of Bayesian likelihood-based generative models that can be used for both prediction and causal inference. The models are 'generative' because they reflect the distribution of random variables that represent processes of interest. New advances in Bayesian machine learning enable the models to be competitive with algorithm-based techniques for predictive accuracy, with the added benefit of properly reflecting uncertainty.
The methods are illustrated using electronic health records data from a large-scale HIV care program in Kenya, and using COVID surveillance data from the Rhode Island Department of Health.

WebEx Password: euA2GJ2Swq9

Event Site Link

https://uthealth.webex.com/uthealth/j.php?MTID=m454c191067563b3f4dc6a39a6dcc3aad

Additional Information