A data-driven approach to evaluating the effectiveness and adverse outcomes of antidepressant exposure using longitudinal real-world EHR data

This dissertation explores the complex dynamics of treatment outcomes and body weight changes in patients undergoing antidepressant therapy, employing advanced machine learning techniques to model and predict clinically significant events. Depression remains one of the most pervasive and challenging mental health disorders worldwide, significantly impacting the quality of life and well-being of millions. The variability in individual responses to antidepressant treatments necessitates a data-driven approach to improve predictive models and optimize therapeutic strategies. Leveraging a robust dataset from the PCORnet Common Data Model, which encompasses comprehensive electronic health records (EHR) of over 2.25 million patients, this dissertation employs advanced machine learning techniques, specifically Bidirectional Long Short-Term Memory (Bi-LSTM) with an attention mechanism, to capture the temporal dynamics and complexities inherent in longitudinal EHR data. These models predict diverse treatment failure outcomes, categorized by clinical severity from dosage adjustments to critical events like hospital admissions and mortality, and assess the impact of antidepressants on body weight, a critical aspect of treatment affecting patient health outcomes. Integral to this research is the commitment to model transparency and explainability, achieved through the integration of SHAP (SHapley Additive exPlanations), which enhances the interpretability of the models for clinical practitioners by detailing how patient characteristics and treatment variables contribute to model decisions. The findings suggest that personalized treatment plans based on predictive analytics can substantially improve patient outcomes, advancing the field of health informatics by demonstrating the application of machine learning in understanding and predicting the outcomes of antidepressant treatments and setting a precedent for future research in personalized medicine approaches in psychiatry.

Name thesis:degree_name

Ph. D.

Level thesis:degree_level

Doctoral

Discipline thesis:degree_discipline

Informatics (MU)

Grantor dc:publisher

University of Missouri--Columbia

Year dc:date.issued

2024

Author dc:creator

• Alaboud, Khuder Ali

Advisor dc:contributor.advisor

• Song, Xing

Language dc:language.iso

eng, English

DOI

https://doi.org/10.32469/10355/105974

OAI identifier oai:identifier

oai:mospace.umsystem.edu:10355/105974