As a statistician for the Rochester Institute of Technology, I provided consulting services with 10+ professors and directors and professional development with researchers. I conducted advanced statistical analyses using Python, R, and SPSS. At that time, I worked with a professor and researcher to analyze Parkinson's disease and depression.
Objective Determine if the presence of Parkinson's disease (PD) with depression is associated with increased depression severity, structure causal-effect relationships surrounding PD, and identify magnitudes of motor and nonmotor symptoms (NMS).
Challenge
Depression is not directly observed, complicating accurate measurement and analysis.
Unlike Alzheimer's disease, no gold-standard directed acyclic graph (DAG) is established for Parkinson's disease.
Each depressive symptom varies in severity and often overlaps with others, possibly impeding the analysis.
My Approach
Collaboration with Subject-Matter Expert: Shared need-to-know concepts of confounding and directed acyclic graph and worked together to develop DAG based on priori knowledge and literature review.
Item Response Theory: Applied 2-Parameter Logistic (2-PL) model with a Structural Equation Model approach (a statistical technique that tests and verifies relationships between variables in a complex model), including differential item functioning (DIF).
Tobit Model: Implemented a Tobit model based on the DAG, set up according to how the 2-PL model's factor scores were formed, and interpreted independent variables of interest using Average Treatment Effects (ATE).
Results
Support for NMS Research: With 30-40% of people with PD affected by NMS and most treatments focused on motor impairments, depression has a more severe impact on individuals with PD compared to those without.
Key Findings: Based on the DIF analysis, the question regarding Dropped interests and activities indicated the highest severity of depression. Furthermore, the Tobit model's (ATE) analysis showed that an increase in the average number of hours spent walking decreases depression severity.
Further Directions: The DAG needs verification using a new dataset with two causal discovery algorithms (fast greedy equivalence search and fast causal inference).