Mikkel Runason Simonsen
MSc in Statistics
PhD @ Aalborg University
Improving the Clinical Utility of Prognostic Tools through Calibration
For a wide range of medical conditions, prognostic models are used routinely to inform patients about their outlooks, guide treatment choice, and recruit patients into clinical trials. However, many prognostic models are developed and used only knowing the discriminatory capacity of the model, and not the model calibration and clinical utility. This PhD program aims to develop a method that can improve calibration, and thus clinical utility, of prognostic models, such that they will apply in heterogenous clinical settings across borders and continents. Additionally, new prognostic models for specific hematological cancers that outperforms existing models will be developed.
The project consists of two elements. Firstly, we will develop a new methodology to improve external validation of prognostic models particularly aiming at improving model calibration. This is of particular interest as new prognostic models developed in a Danish setting may not perform as well in other countries with different clinical standards, background mortality, and culture. Secondly, we will develop new prognostic models within hematological cancers using the newly developed methodology in combination with machine learning and artificial intelligence (AI) approaches. Denmark holds numerous comprehensive clinical registers, which the model development will be based on.
Development of a methodology for improving performance, particularly model calibration, of prognostic models will allow for the development of prognostic models that perform well in a variety of economic, cultural, and clinical settings. Improving the precision of prognostic models will provide health care planners, patients, and clinicians with a better foundation for making important clinical decisions. For instance, accurate prognostic models for hematological cancers can identify high-risk patients more accurately at the time of diagnosis, which can be used to guide treatment or recruit patients for clinical trials. Identification of low-risk patients is also important as these will be candidates for de-escalating treatment, which can avoid severe side effects from the treatment.