Back to search

Boston University

Predicting psychological treatment dropout using graph neural networks

Abstract

dc:description.abstract

Psychological treatment dropout remains a persistent challenge in mental healthcare, with significant effects for patient outcomes and healthcare resource utilization. This study proposes a framework for predicting treatment dropout using probabilistic Graph Neural Networks (GNNs) applied to data collected from the Wellness and Recovery After Psychosis (WRAP) program at Boston Medical Center. The approach extends the BernGraph [10] model by generalizing probabilistic node encoding to support both binary and continuous clinical variables, the latter transformed into probabilistic representations via z-score standardization and cumulative distribution functions. Edge weights between nodes are computed using conditional probabilities for binary pairs and Gaussian similarity for continuous variables, in order to capture nuanced interdependencies among clinical features. The study’s results show a substantial performance improvement across all models when using GNN-generated features compared to raw clinical inputs. This work presents a promising step towards more GNN applications in computational healthcare research.

Degree

thesis:*
Name thesis:degree_name
Master of Science
Level thesis:degree_level
masters
Discipline thesis:degree_discipline
Systems Engineering
Grantor
Boston University
Year dc:date.issued
2025

Author and committee

dc:creator, dc:contributor.*
Author dc:creator
  • Le, Khang
Advisor dc:contributor.advisor
  • Venkataraman, Archana
ORCID dc:identifier.orcid

Subjects

dc:subject × 1

Rights

Language dc:language.iso
en_US

Identifiers

dc:identifier.*
Handle dc:identifier.uri
https://hdl.handle.net/2144/51175

Chain of custody

source
Harvested from
Boston University
Base URL
open.bu.edu/oai/request
Last updated
2026-11-18
Source record
OAI-PMH GetRecord
related terms
citation

Le, Khang. Predicting psychological treatment dropout using graph neural networks. masters thesis, Boston University, 2025. https://hdl.handle.net/2144/51175