04/04/2024
By David Musoke

The Biomedical Engineering and Biotechnology program invites you to attend a doctoral proposal defense by David Musoke on "A Multi-Headed Dense Neural Network for Healthcare Surveys: Forecasting the behavior of health workers in longitudinal surveys."

Candidate Name: David Musoke
Defense Date: April 18, 2024
Time: 3 to 5:30 p.m.
Location: Perry Hall Room 415, North Campus

Committee:

  • Advisor: Laura Punnett, Biomedical Engineering, UML
  • Karyn Heavner, Public Health, UML
  • Milo DiPaola, Mechanical Engineering, UML
  • Mazen Elghaziri, Susan and Alan Solomont School of Nursing, UML

Abstract:

Throughout the COVID-19 pandemic, healthcare workers faced many stressful challenges that affected their physical and emotional health and personal safety. Many risked daily exposure to the virus despite the extensive health measures to lessen such exposure. The heightened exposure, long extended working hours, and the increased load of distressed patients further worsened the challenges faced by these healthcare workers during the pandemic. These unique events resulted in increased stress levels, anxiety, depression, loneliness, and other mental health concerns for these workers.

It would be beneficial if there were a way to prepare healthcare institutions and workers for the adverse impacts of such stressful events before a pandemic occurs. The proposed research seeks to identify predictive factors that could be addressed in the future to prevent repeat impacts on healthcare workers.

The research involves a longitudinal employee health and safety survey, originally part of the Safety and Health through Integrated, Facilitated Teams (SHIFT) study at five public sector healthcare facilities in the New England region of the United States. However, the pandemic disrupted that study, and the team investigators decided to continue but at a reduced capacity and with a different mission: to measure the impact of the pandemic on worker health, safety, and well-being of mental health workers while noting relevant working conditions and organizational factors that could impact worker performance at these same five unionized facilities in the New England area. We shall focus on two related work performance outcomes, 'Job Satisfaction' and worker 'Intention to Leave within 2 years', and determine how they were affected by the COVID-19 pandemic.

The initial baseline survey was taken in 2018 (pre-COVID-19), while the second was conducted after facility lockdowns were lifted in 2021 (post-COVID-19). The disruption of the pandemic caused only a subset of 428 healthcare workers to participate in both sets of surveys. The healthcare workers consisted of nursing assistants, mental health workers, and others who provided inpatient, outpatient, and long-term care to civilian and veteran patient populations. The average healthcare worker was 47 years old and had 10 years of on-the-job experience; 64% were female. The racial distribution was 69% Caucasian, 19% African American, 8% Hispanic, 2% American Indian and 2% Asian.

We are designing a novel artificial neural network model, a Multi-Headed Dense Neural Network,
that accepts pre- and post-COVID-19 survey data about healthcare worker health, safety, and well-being and synthesizes a new near-identical post-pandemic COVID-19 health, safety, and well-being dataset. A transfer function of the model is developed from the original pre- and synthetic post-COVID-19 datasets that should be able to accurately 'predict' future health, safety, and well-being metrics to new but similar healthcare worker populations.

We shall establish all critical, independent factors that significantly affect Job Satisfaction and Intention to Leave within 2 years through rigorous bivariate and multivariate regression analyses, plus identify any covariates capable of modulating these outcomes. The computed regression coefficients will determine the effect sizes of the COVID-19 pandemic on the Job Satisfaction and Intention to Leave within 2 years outcomes for healthcare workers.

We shall also determine the statistical significance (SS) and effect size (ES) metrics between the pre and synthesized post-COVID-19 paired datasets. All statistical computations will be made via SAS 9.4 analytical software.

The models' regression coefficients, SS, and ES data will be compared to that computed from the original pre- and post-COVID-19 paired datasets. A Relative Similarity Index (RSI) shall be computed to denote the similarity between these computed data sets. If all indices are high (0.9 ≤ RSI ≤ 1.1), implying the model can synthesize post-pandemic behavior to within 10%, we would be justified in using the model to predict future post-event outcomes with similar accuracy, given new pre-event surveys from similar populations working under similar conditions as the ones used to develop this model.

All interested students and faculty members are invited to attend the defense.