UNFOLDING THE VARIABILITY OF CLINICAL DATA IN PARKINSON TREATMENT USING MULTI-OBJECTIVE ANALYSIS
OBJECTIVE QUANTIFICATION OF OBSERVATIONAL DATA
SELECTION OF IMMUNE PARAMETERS FOR TREATMENT OF PARKINSON
Parkinson’s disease (PD) is the second most common neurodegenerative disease with many motor and non-motor symptoms that range from impairments of motor capabilities to decreased sense of smell and cognitive abilities. Thus, unidimensional quantification of PD progression is inaccurate and can potentially introduce biases into the analysis of pathology. At the same time, the long-term nature of PD means that randomized controlled experiments are difficult to set up and researchers often need to draw conclusions from observational data collected during the treatment of patients. In this project, we utilize a multi-objective-based approach to cluster the clinical data according to the test subjects’ clinical behavioral performances to objectively represent their relative levels of PD progression. We then perform a correlation-based feature selection to rank the available immune parameters in terms of significance, to further analyse the underlying biological mechanisms of PD. The aim of this analysis is to identify parameters that best explain treatment effectiveness over treatment timeline.
What we want to achieve
Our Project Goals
Establish a framework for quantification of relative PD progression within a observational study subject group
The observational study at hand quantifies PD progression within the test subjects using three clinical behavioral metrics, namely sniffing score, UPDRS-III score and PANDA score, which measures the olfactory, motor and cognitive capabilities of the subjects respectively. Since all three metrics are ordinal with clear direction of desirability, we consider all three metrics using the principle of Pareto optimality and focus on the relative pairwise Pareto dominance among the subjects.
Divide test subjects into clusters with different levels of PD progression
We sort the test subjects using non-dominated sorting with respect to the available metrics and assign each subject a front number which represents their relative optimality in clinical behavior and thus level of PD progression. The test subjects can be clustered in accordance to the direction of non-dominated fronts, and the clusters will be differentiated by relative levels of PD progression.
Divide test subjects into clusters with different trade-offs with respect to different PD behavioural metrics
When the test subjects are clustered perpendicular to the direction of non-dominated fronts, the clusters will have mixed levels of PD progression but be differentiated by different trade-offs when it comes to different clinical behavioral metrics.
Select features among available immune parameters based on correlation with cluster membership
For every parameter, we compare the statistical distribution of the patients within and outside a considered cluster, via the mean and standard deviation, which are normalized for the entire population. We then perform non-dominated ranking on the immune parameters to identify the ones with significant differences within a cluster and conduct further investigations on them.
Validate the results with expert knowledge
Our results are checked by medical researchers for agreement with established knowledge in the area and potential new discoveries based on the available data.
Project Team
Dr. Christiane Desel
Alexander Duscha
Univ.-Prof. Dr. Aiden Haghikia
Dr. Tobias Hegelmeier
Prof. Dr. Sanaz Mostaghim
Qihao Shan
Publications
Support our research
Become a study participant!
How do our brains, our bodies and our environment interact? How do physical illnesses affect our mental performance? And why are we more efficient on some days than others?
We would like to get to the bottom of these questions together with you. Register now and take part in exciting studies.
