On diagnostic codes (2021)

Abstract

Improved survival rates among patients with ischemic heart disease (IHD) have stagnated, potentially because these patients are becoming increasingly multi-morbid. Thus, development of new methods for deep phenotyping is necessary to increase patient outcomes (PMID: 30274592).

Accordingly, the hypothesis is that data-driven models based on comprehensive analyses of Danish healthcare data are of value in patient phenotyping and that they can complement classical epidemiological studies.

The three main objectives of the thesis are:

to present an overview of IHD along with the medical classification systems that comprise the foundation for patient phenotyping using electronic patient record data,
to describe how disease trajectories, mathematical graph theory, and artificial neural networks can be applied in the context of IHD multi-morbidity and,
to discuss strengths and limitations of the models, focusing on multi-morbidity in IHD, the clinical utility and perspectives on future directions.

In the studies that the thesis is based on, it is showcased how the temporal order of diagnoses in this population may be indicative of different subpopulations (manuscript accepted). Also, a strategy for unsupervised clustering of IHD patients is presented (manuscript being revised). And finally, the key elements towards development of an artificial neural network risk prediction tool based on hundreds of input features are described (manuscript in preparation).