WPI - Computer Science Department, PhD Proposal Defense , Trusting Inekwe "Predictive and Causal Machine Learning Models of the Impact of Pandemics on Patients with Cardiovascular Diseases"

Trusting Inekwe

PhD Proposal Defense

WPI - Computer Science Department 

Thursday, October 10, 2024,

Time:  11:30 a.m. – 1:30 p.m.

Location: FL 141

Committee members:

Advisor: Prof. Emmanuel Agu, WPI – Computer Science Dept.

Prof. Dmitry Korkin, WPI – Computer Science Dept.

Prof. Chun-Kit Ngan, WPI – Computer Science Dept. and Data Science Program 

Prof. Andres Colubri, Genomics and Computational Biology UMass Medical School – External Committee member

 Abstract:

Pandemic-induced disruptions to routine healthcare and lifestyle changes in cardiovascular diseases (CVDs) patients triggered changes in critical CVD biomarkers (measurable parameters of the body that can indicate health or illness). Prior work has overlooked models for predicting these biomarker trajectories or modeling causality during pandemics. Utilizing a first-of-a-kind Electronic Health Record (EHR)  dataset of over 400,000 patients treated at the UMass Memorial hospital before and during the Covid pandemic, this doctoral dissertation proposes ML predictive and causal models of the COVID-19 pandemic’s impact on the following CVD patient biomarkers. 

This research proposes three thrusts.  The first, published in the flagship IEEE CHASE conference, explored traditional ML models on EHR data attributes, to predict the impact of the COVID-19 pandemic on CVD patient biomarker trajectories (BP,  LDLchol, HbA1c and BMI) and ML causal analysis exploring the Debiased ML for Difference-in-Differences approach.  Two innovative research thrusts are proposed to address limitations of the first work including low predictive capacity of traditional ML models (low r2 values, high MSE values), the lack of uncertainty estimation, neglecting to exploit temporal data relationships and not performing multi-target predictions 

1. Genetic Algorithm (GA) Neural Architecture Search (NAS) for automated, large CVD DL model design, and enhance the low predictive capacity of traditional ML models.  NAS offers an automated approach to DL model design and determines optimal hyperparameters, obviating the need for expertise. For the first time, we apply GA, an optimization technique inspired by the principles of natural selection and evolution, to optimize NAS to generate a CVD biomarker DL model.

1. Bayesian Transformers for multi-CVD-biomarker trajectory prediction, leverage temporal relationships and mitigate data uncertainty. Bayesian Transformer (BT) is a variant of the Transformer model that incorporates Bayesian inference techniques for uncertainty estimation in the model’s predictions. We innovatively adapt BT for multi-target prediction of the pandemic's impact on CVD patient biomarkers, capturing temporal and biomarker relationships and modeling uncertainty. 

This research will advance predictive and causal modeling of pandemic impacts on CVD patients and facilitate tools for early detection of health deterioration, uncovering latent trends, and forecasting pandemic-induced health risks. Such tools could be utilized by healthcare professionals, epidemiologists, and public health policymakers for more informed decision-making and targeted interventions during pandemics.