Predictive Model of Early SUD Treatment Dropout
At the Florida Recovery Center Research Lab (FRC), we have started a new project that aims to examine early dropout from substance use disorder (SUD) treatment with the goal of developing more effective methods of dropout prevention. By utilizing a machine learning approach, we aim to design a predictive algorithm that identifies a patient's risk of early dropout, and indicates cases of increased risk of dropout. We hope to apply this tool to improve clinical practice regarding addiction treatment.
FALL 2024 - SPRING 2025
Principal Investigator
Dr. Ben Lewis
Institution
University of Florida
Department of Psychiatry
UF Florida Recovery Center Research Lab
Project Details
Research Focus
Addiction recovery and maintaining abstinence from substance use relies on successful completion of SUD treatment. Early dropout from SUD treatments has been associated with negative outcomes, including a greater risk of relapse. Therefore, identifying factors that contribute to early dropout, as well as designing a model that can preemptively identify cases of high dropout risk, can inform clinical practice and improve treatment approaches. We aim to design a predictive algorithm that serves this purpose, and that can be applied in practice for clinical use.
Research Methodology
The predictive model of early SUD treatment dropout was designed using a random forest machine learning approach. The model was applied using 300+ baseline features and longitudinal data collected from patients at FRC. Data gathered at treatment entry will be used by the model to predict risk of subsequent dropout from treatment.
Project Responsibilities
With this project I was involved in the administration of patient health assessments to individuals seeking SUD treatment. The data collected from these assessments compile a comprehensive biopsychosocial databank, from which our data was extracted from. Thus, I held data collection and management responsibilities. I was also involved in analysis of this raw data and the development of a predictive model of treatment dropout.