REPOSITORY > RESULTS

Abstract

Parkinson’s disease is the second most common neurodegenerative disorder that affects
people worldwide. Its symptoms affect different aspects of patients’ lives that influence
the quality of life and that of their families. Given the wide variability of symptoms
among different patients and also for the same patient at different times, it is difficult
for the clinicians to manage the Parkinson’s disease. Using data mining for analyzing of
Parkinson’s disease data can lead to the identification of similar patients, with the aim to
assist the clinicians to respond more promptly and in a more personalized fashion to the
changes of the patients’ status.
The management of Parkinson’s disease to improve the patients’ quality of life is not
widely researched. In this thesis, we address the question of data mining support for the
analysis of Parkinson’s disease progression and management with antiparkinson medications.
The thesis is divided into three parts addressing improved descriptive modeling,
grouping of patients based on similar disease progression, and analysis of the most important
symptoms and medication changes.
High-quality descriptive models are highly appreciated by the clinicians. Classification
rules and rules describing interesting subgroups are important components of descriptive
data mining. Rule learning algorithms typically proceed in two phases: rule refinement selects
the conditions for specializing the rule and rule selection selects the final rule among
several rule candidates. While most conventional algorithms use the same heuristic for
guiding both phases, recent research indicates that using two separate heuristics is conceptually
better justified, improves the coverage of positive examples, and may result in better
predictive accuracy. The thesis presents and evaluates two new beam search rule learning
algorithms: DoubleBeam-SD for subgroup discovery and DoubleBeam-RL for classification
rule learning. The algorithms use two separate beams and can combine various heuristics
for rule refinement and rule selection, which widens the search space and allows for finding
rules with improved quality. In the classification rule learning setting, the experimental results
confirm previously shown benefits of using two separate heuristics for rule refinement
and rule selection. In subgroup discovery, DoubleBeam-SD algorithm variants outperform
several related state-of-the-art algorithms. We use the newly introduced algorithms in
Parkinson’s disease data analysis.
Careful management of the patient’s status is crucial to ensure the patient’s independence
and quality of life. This is achieved by personalized treatment based on individual patient’s
symptoms and medical history. We address the issue of determining patient groups
with similar disease progression patterns coupled with patterns of medication changes that
lead to the improvement or decline of the patients’ quality of life symptoms. To this end,
we propose a new methodology for clustering short time series of patients’ symptoms and
prescribed medications. The methodology also employs time sequence data analysis using
skip-grams to monitor disease progression. The results demonstrate that the motor
and autonomic symptoms are the most informative for evaluating the quality of life of
Parkinson’s disease patients. We show that Parkinson’s disease patients can be divided into clusters ordered in accordance to the severity of their symptoms. By following the evolution
of symptoms for each patient separately, we were able to determine the patterns of
medication changes that lead to the improvements or degradations of the patients’ quality
of life.
Based on discovered groups of similar patients, we present a step towards personalized
management of Parkinson’s disease patients. We present two novel approaches. The first
algorithm analyzes the symptoms’ impact on the Parkinson’s disease progression. Experiments
on the Parkinson Progression Markers Initiative (PPMI) data reveal a subset of
symptoms influencing the disease progression which are already established in Parkinson’s
disease literature, as well as symptoms that have been considered only recently as possible
indicators of disease progression. The second novelty is a methodology for detecting
patterns of medications dosage changes based on the patient’s status. The methodology
combines multitask learning using predictive clustering trees (PCTs) and short time series
analysis to better understand when a change in medications is required. The experiments
on PPMI data demonstrate that using the proposed methodology, we can identify some
clinically confirmed patients’ symptoms suggesting medications change. In terms of predictive
performance, the PCT approach is comparable to the random forest model but is
advantageous due to the model interpretability.