Causal Inference and Machine Learning Methods in Parkinson's Disease Data Analysis

<p>This dissertation documents an investigation into Parkinson’s Disease utilizing machine learning and causal inference methods. I will cover a descriptive analysis of Parkinson’s Disease (PD) in a vast, high-quality database and present costs associated with Parkinson’s Disease medications. I also researched a causal inference method assessing the Carbidopa-Levodopa effect on two-year survival and a causal survival analysis on a one-to-five-year survival comparing no drug use and Carbidopa-Levodopa in Parkinson’s Disease patients.</p> <p>For my classification with Parkinson’s gait, patients were monitored with a smartphone and an additional 6 Inertial Measurement Unit (IMU) sensors to collect clinical gait measures. I used classical machine learning algorithms on raw smartphone data to distinguish between ON and OFF times. With an average accuracy of 92.5%, this work demonstrates the feasibility of using smartphone data to distinguish between ON versus OFF walking and lays the groundwork for a real-world, corrective feedback system.</p> <p>I also researched the causal effect of the most prevalent PD medication in terms of survival. In particular, I focused on the probability of two-year survival with PD patients taking Carbidopa-Levodopa and no drug use and assessing whether there was an effect on survival utilizing the doubly robust method. My results with the differences of causal effects showed a 0.013 positive increase taking Carbidopa-Levodopa indicating this medication had a significant positive effect on the two-year survival of PD patients.</p> <p>I then furthered this study and conducted a causal survival analysis from one-to-five-year survival with two treatments, no drug use and Carbidopa-Levodopa. The results showed that Carbidopa- vii Levodopa had a significant effect on survival when the drug was prescribed within three years from first diagnosis and no drug use had a significant effect at four and five years of survival.</p> <p>In the process of better using the current data, a descriptive statistical analysis was conducted. As such, I studied a vast and high-quality database Cerner Real-World Data and focused on people who were diagnosed with Parkinson’s Disease from 2016 to 2022. I researched the demographics, comorbidities, and medications of PD patients. After cleaning the database, my final cohort size was 110,037 subjects.</p>

Name thesis:degree_name

Doctor of Philosophy (PhD)

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Computational and Data Sciences

Year dc:date.available

2023

Author dc:creator

• Pierce, Albert

Contributors dc:contributor

• Dr. Cyril Rakovski
• Dr. Adrian Vajiac
• Dr. Sidy Danioko

• Causal Inference
• Machine Learning
• Parkinson's Disease
• Doubly Robust Method
• Carbidopa-Levodopa
• Data Science

Repository record dc:identifier

https://digitalcommons.chapman.edu/cads_dissertations/36

OAI identifier oai:identifier

oai:digitalcommons.chapman.edu:cads_dissertations-1037