Problematic alcohol use is the third leading contributor to the global burden of disease and is considered to be the main cause of 3-8% of global mortality. At the same time, alcohol and substance use disorders have the highest treatment gap of all mental health disorders, with the majority not receiving treatment. The consequences of problematic alcohol and drug use impose large costs on societies, and reducing alcohol and tobacco use are two of the three main aims of the national prevention agreement (Preventieakkoord). E-health interventions have been developed since a few decades, yet little is known about their efficacy and what elements in an e-health intervention are related to its effects. By its characteristics (high data availability e.g. on registrations of alcohol/substance use , time spent, assignments made, data logs of type of activities engaged in within an e-health intervention), e-health interventions lend itself excellently for application of innovative data science methods like data mining, machine learning and growth curve modelling. With machine learning, patterns can be identified that are linked to successful outcomes of the e-health intervention. Specifically, machine learning enables single subject inference, and thereby selection and adaptation of E-health interventions on the single-person level. With growth curve modelling, individual trajectories and transitions in substance use behaviour through time can be estimated. Thus, with the use of machine learning and growth curve modeling, personalizing interventions at the single-subject level is enabled.
Research aims and methods:
In this project therefore, machine learning and growth modelling is applied to (1) investigate what types of use of the ehealth interventions (e.g. modules activated; exercises completed) is linked to (1) adherence and (2) meeting health goals (i.e.: outcome in alcohol and substance use). In a second phase of the project, a pilot implementation of adaptions to the programme based on the insights gained from machine learning and growth modeling in the Jellinek Selfhelp Ehealth interventions will be tested. Machine learning algorithms will be applied to the e-health intervention, in order to see whether implementation of knowledge gained from these algorithms leads to more effective e-health intervention outcomes. Growth curve modelling will be used to evaluate individual differences in and correlates of adherence, in a prospective cohort study in which the selfhelp ehealth interventions will be offered with for instance feedback and/or sms/email coaching (dependent on the focus group outcomes). The Ehealth intervention that is studied is the evidence-based Jellinek online selfhelp for alcohol and substance use disorders. Jellinek is at the forefront of developing e-health interventions, developing and testing early e-health interventions in mental healthcare since 2008, and innovating web- and app-based selfhelp since then. The Jellinek current selfhelp programs are used by over 12000 users each year, rendering high feasibility for this project in terms of available data for the proposed data science methods. By investigating both alcohol, tobacco, cannabis, and cocaine selfhelp interventions, we will investigate whether similar machine learning patterns are present for the e-health interventions for these substances, or whether unique patterns appear.
Relevance and impact:
The proposed project will delineate how machine learning can be applied to e-health interventions in order to improve usage and successful completion of e-health interventions, to ultimately prevent substance use disorders. Our results will therefore have high relevance for the evaluation of other (mental health) e-health interventions as well, as the evaluated substance use disorder e-health programmes of Jellinek are based on cognitive behavioral therapy (CBT), which is one of the main evidence based treatment methods for depressive and anxiety disorders as well.
As the primary aim is to generate new knowledge on how innovative methods like data mining and machine learning can be applied to improve the use and effects of ehealth interventions, this is of interest to researchers and clinicians in Ehealth (e.g. addiction, mental health and Ehealth interventions more broadly), to prevention organisations; to ehealth software developers, to translational researchers in psychiatry and psychology, and to machine learning researchers. Knowledge dissemination and implementation by use of focus groups, project meetings and an expert meeting geared at these groups are an important part of the proposed project.