ORCID
0009-0001-8574-6826
Date of Award
2025
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Complex Systems and Data Science
First Advisor
Nicholas Cheney
Abstract
Internalizing mental health disorders, namely Anxiety and Depression, begin presenting as early as preschool, and incident rates in children have been rising. Without appropriate interventions, childhood internalizing disorders are linked to long-term negative health outcomes, including higher rates of suicide and substance abuse. The current diagnostic procedure is based on caregiver-reported surveys and interviews, which require an external observation of the child’s symptoms. As given by their name, internalizing disorders are difficult to observe, as their effects are largely unseen, making reliable screening challenging for young children who cannot yet verbalize their feelings. Wearable sensors and digital phenotyping have shown promise in enabling objective and rapid point-of-care screening of these disorders and others, such as ADHD. To this end, this study investigates the relative role of survey-based data in a multimodal diagnostic classification model for anxiety, depression, and ADHD for children aged 4-8. We compare model performance across the three diagnostic targets and assess the importance of subsets of the feature space when modeled alone (either just biometric data, which is physiological data such as heart rate variability and skin conductance, or just survey data) versus together. Results found that models targeting ADHD benefited most from a multimodal approach, which incorporates both survey-based and biometric data. Still, models targeting depression achieved the highest overall performance across data groups. Among the survey feature subsets, it was found that age and gender provided significant context for models using multimodal data. When biometric data was included in the models, standardized screening surveys such as CBCL, BIQ, and Spence did not significantly improve model performance compared to other survey-based categories, suggesting that the information they provide may be redundant when used alongside physiological data. Among the biometric feature subsets, Heart Rate Variability (HRV) was the most consistently informative modality across biometric-only and aggregated models. The location of biometric sensors on the children’s bodies is also an important metric. Sensors placed on the wrist and sacrum were consistently ranked higher in importance compared to those placed on other locations, such as the thigh, which were not found to be significantly different from other categories. However, it is important to note that not all sensor locations provide the same type of data, so these findings are based both on the physical placement of the sensors and the specific physiological signals they capture at each location. Overall, these findings support the development of a multimodal, noninvasive diagnostic framework for childhood internalizing mental health disorders.
Language
en
Number of Pages
44 p.
Recommended Citation
Rees, Anna Sofia, "Bridging Subjective And Objective: The Role Of Caregiver-Reported Surveys In Multimodal Diagnostic Models Of Childhood Depression, Anxiety, And Adhd Anxiety, And" (2025). Graduate College Dissertations and Theses. 2074.
https://scholarworks.uvm.edu/graddis/2074
Included in
Biomedical Engineering and Bioengineering Commons, Computer Sciences Commons, Medicine and Health Sciences Commons