The current opioid crisis is significantly impacting millions of lives, healthcare, social welfare, and the
economy. Patient interactions with the treatment system coupled with results of completed studies create a
wealth of data stored in disparate electronic sources. Therapists and health care providers with limited time and
resources face challenges to access, integrate and monitor this vast data set for novel opportunities to improve
care. Significant advances would include predicting when a patient will relapse out of a program, and also
suggesting optimal personalized care strategies to reengage patients before this negative event occurs.
Survive-OUD will solve these challenges by providing a web-based therapist interface integrating survivor
model artificial intelligence (AI) strategies. Based on multiple input data domains leveraged from existing
electronic medical record sources, survivor recurrent neural networks will be trained to recognize when patients
are likely to relapse or drop out of an OUD program. Examples of data domains that can be input to the network
include patient demographics, medical and prescription data, engagement with therapy paradigms, and
compliance with logistical program tasks. Furthermore, once a patient is noted as high risk, a second layer of
algorithms will be developed to recommend a specific and personalized care strategy for retention based on
existing best practices in the literature and clinical trials. Therefore, the Survive-OUD platform will also integrate
with common literature database and clinical trial repositories. Utilizing an existing AI platform for searching,
tagging, and extracting data from database sources, the innovative platform will close the loop on actionable
results by recommending updated care options based on potential outcomes learned from best practices in
existing literature. The AI architecture developed will greatly improve success rates in opioid addition programs
and expand high quality healthcare.
While the commercialized Survive-OUD platform will integrate all features above, Phase I will target feasibility
of data aggregation and AI algorithms to detect relapse and recommend intervention strategies. The innovative
technical challenge in Phase I is to develop and validate targeted AI tools using data already being captured in
patient workflow to allow early prediction of patient retention issues. More specifically, a prototype therapist
interface and data network infrastructure will be developed to source personalized patient data as well as
literature and clinical trial sources. Once the platform architecture has passed verification testing, it will be
deployed in a field data collection study to determine usability and also provide a rich set of de-identified data for
algorithm development. Collected data will then be used to train and test AI algorithms for early detection of
patient dropout/relapse and appropriate treatment recommendation.
