Real time relapse risk scoring for Opioid Use Disorder (OUD) from clinical trial datasets Project Summary Any clinician treating a patient with Opioid Use Disorder (OUD) would like to know whether this patient would relapse in the next week or month. Such a score, analogous to a credit score in consumer finance, may be similarly obtained from longitudinal data streams derived from patient behavior during SUD treatment. There are several common data sources that every OUD patient in treatment produces: a binary, longitudinal data survey of use patterns for a set of pre-determined substances of abuse, treatment session attendance records, and medication records. In particular, urine drug screens (UDS) or alcohol and nicotine breathalyzers and standard Timeline Follow Back (TLFB) questionnaires are universal surveys in every treatment delivery context, including large pragmatic clinical trials. While these data streams are incomplete, of different lengths and sampling frequencies, and correlate in complex ways, contemporary machine learning methods allow us to overcome these challenges. We aim to build a toolbox that would allow for the following: 1) standardized methods for risk scoring and visualization from UDS and TLFB datasets in existing large clinical trials; 2) standard methods for inferences of risk scores: procedure for hypothesis testing whether an intervention made a difference in the risk scores and their trajectories. 3) user-friendly software modules aimed toward researchers and administrators for quality improvement projects and customized predictive modeling pipelines, and interpretable web portal for clinicians, analogous to a credit report. This proposal will also incorporate usability survey and evaluation for algorithmic bias. These applications will provide a computational framework for future real time predictive modeling work for many other different substance use disorders.