|
Abstract:
|
Synthetic data is increasingly being generated to meet the rising demand for clinical data, especially in the context of the COVID-19 pandemic. In this presentation we will discuss various generative methods that can scale down to small datasets, as well as scale up to large databases, and present a framework for evaluating the utility of synthetic data. The methods will be demonstrated using real world datasets covering different clinical and public health domains. Generative methods that will be discussed include statistical machine learning and deep learning models suitable for cross-sectional and longitudinal datasets, with comparisons of their various strengths and weaknesses. In practice, multiple models need to be used to accommodate the heterogeneity of real world data. In terms of applications, generative models enable the sharing of data in a privacy protective manner, and also can be used as data simulators to augment small datasets and for complex predictions.
|
