Predicting relapse in substance use disorders using machine learning approach using mobile sensing of neurocognitive functioning
Although there has been made substantial progress in the last decade concerning the neurobiology of addiction and the understanding of how these chronically relapsing brain disorders develop and are maintained, they still remain a significant public health challenge. The number of patients that relapse is very high. Studies show that between 40 and 60 % of patients remained abstinent one year after treatment initiation. A recent local study at the University Hospital in Trondheim showed that 37 % of patients with substance use disorders (SUD) relapsed during the first 3 months after in-patient treatment.
Most research on predicting relapse to SUD is based upon retrospective and crossectional data from pscyhometric scales, observational indexes, and even intuition. Psychosocial factors addressing increased likelihood of relapse are plentiful, but there is still scarce research based on more objective assessments, such as neurocognitive impairments, physiological and behavioural variables, as well as longitudinal and prospective research designs with high data granularity.
This study aims to broaden our understanding of the objective factors leading to relapse of substance use and enhance our ability to predict and prevent relapse in real time. We will focus on developing more objective real-time assessments of variables predicting relapse in general based on data from smart phone sensors and using machine learning methods. And specifically study how neurocognitive impairments might mediate and determine the probability of relapse, so called precision health care.
A special issue of the journal 'Addictive Behavior' (2018)concerns exactly the usage of these ambulatory assessment methods to better understand the '…etiology, maintenance, treatment, and remission of addictive disorders'. Assessments based on mobile sensor data from mobile phones and wearables show promise as a method to understand the dynamics and predict clinical changes in this population.
Precision Health Care for SUD will depend upon the early and sensitive detection of precursors of substance use and relapse. Moreover, to achieve this, there is a need to narrow the gap between theoretical models based on retrospective data, crossectional research designs and longitudinal, real-world behavior assessment in as real-time as possible.
First objective: We aim to study the cross-sectional and longitudinal relationships between polysubstance abuse and executive neurocognitive functioning. Using a lagre population study and a prospective cohort study of patients. Executive neurocognitive factors are thought to mediate the effects of other probability influencing factors on relapse (e.g.mental health).
Second objective: We aim to develop a statistical model accounting for both trait and state executive neurocognitive factors influencing the probability of relapse to substance abuse.
Third objective: We aim to investigate the accuracy of mobile sensing and machine learning methods in predicting different aspects of executive neurocognitive functioning.
Fourth objective: We aim to validate the ability of the statistical model to predict craving and relapse in real time with adequate specificity and sensitivity using a mobile sensing application with changes in neurocognitive status as a presumed mediator.