Antonia is interested in both applied and methodological research, predominately in the field of predictive/prognostic modelling. Specifically, she uses data-driven methods on routinely collected health data to improve the development, validation and implementation of such prediction models. Her research project is mainly focused on the problem of missing data in electronic health records (EHR).
Electronic health records (EHRs) are an increasingly common source of data for clinical risk prediction. They offer multiple advantages such as allowing one to observe more individuals at more time points. Challenges in using EHRs for CPMs development and validation include those related to missing data.
There are several methods for handling missing data, most common of which is complete case analysis or list-wise deletion. More advanced and reliable method is multiple imputation (MI), since it restores the natural variability of the data. However, if one would like to use this approach when applying a CPM to an individual patient, access to the derivation set data is needed. As a consequence, researchers often use multiple approaches to handle missing values during development and validation of a model.
Currently, there are no established standards for handling missing data in terms of prediction modelling. Therefore, Antonia's PhD project aims to explore the effect of using different approaches to handle missingness on the CPMs predictive performance, developed and validated using EHR. Initial plans include searching the literature for clinically implemented models and evaluating how the missing data are handled during each model’s development, validation and implementation.