Genetic disorder classification based on existing patient data
Diagnosing genetic disorders can benefit patients in various ways. Most importantly, a diagnosis makes it more possible to understand the symptoms of a patient and find an appropriate treatment. However, a large portion of the process of diagnosing is done manually by physicians, which makes misclassification a considerable risk. Accurately automating part of the process can reduce this risk significantly. A tool that attempts to do so is the Phenomizer. This widely used tool ranks diseases that describe the symptom set of the patient best, along with a corresponding p-value. For this, comparisons are made between the patient and disease profiles, which consist of all phenotypic abnormalities associated with the corresponding disease. The similarity between these patients is calculated using the Resnik score. While the Phenomizer is widely used, its performance can still greatly be improved. In order to do so, a new approach is introduced in this research. In this new approach, new patients are compared to already diagnosed patients, using the same similarity score as is used for the Phenomizer. This approach thus needs patient data in order to function. This data is not yet publicly available. Results show that the new approach performs significantly better than the Phenomizer. Where the Phenomizer has an accuracy of .400, the new approach has an accuracy of .834. However, the data needed to apply this approach to all syndromes is not yet available. This research shows the importance of collecting the data needed in order to diagnose and treat patients as good and early as possible.
Faculteit der Sociale Wetenschappen