Behavioral Study to Predict the Efficacy of a Self-Help Tool

Background:

Cognitive-Behavioral Therapy (CBT) is a learning-based psychotherapy treatment that has been established as effective treatment for depression. It consists of two core interventions: cognitive restructuring (CR) and behavioral activation (BA). In recent years, internet-delivered CBT (iCBT) has been developed, which allows for dissemination of standardized, evidence-based, CBT treatments at scale.

Virtually all psychotherapy methods aim to teach clients something new: new behavioral or thought patterns, new responses to triggers and situations, and/or new emotional reactions. As such, learning-based psychotherapies modify targeted brain circuits to the extent that these circuits show flexibility and are amenable to change through learning. The human brain has several learning circuits/mechanisms that work in parallel. Different people will have more flexibility and learn more effectively through some learning mechanisms and not others. Because different psychotherapy methods (e.g., BA and CR) rely on different types of learning, the investigators hypothesize that by characterizing what learning mechanisms are most available and efficient for each person, will allow us to predict what intervention method will be most effective for that person.

The goal of this study is to test if people's individual learning propensities can predict what type of intervention will benefit them more.

Detailed study design:

In a fully online study, the investigators deploy several behavioral tasks to assess individual differences in learning and decision-making processes. Participants then get access to an internet-based self-help tool for depression that is based on CBT principles, and are asked to undergo either the BA modules or the CR modules (random assignment; 5 weeks of 1-hour session per week). The investigators follow up on symptoms at the middle of the period of use of the self-help tool (at this point, some behavioral tasks are repeated as well), at the end of this period, and at different times up to a year after finishing use of the self-help tool.

Participants are recruited online via advertisement, for instance on social media. After checking eligibility and obtaining consent online, participants are randomized to be part of the discovery or validation dataset (the investigators will not analyze the validation dataset until all analyses are pre-registered). Participants are then asked to fill out a range of symptom self-report questionnaires and complete a series of behavioral tasks. They are then randomized to either a cognitive restructuring group or behavioral activation group. All participants are then given access to e-couch, a validated self-help tool. In the first week, all participants complete the depression information submodule from the depression program. Over the following four weeks, participants in the cognitive restructuring group complete submodules on cognitive restructuring and participants in the behavioral activation group complete submodules on behavioral activation and physical activity. After that period, participants are free to engage with any of the other modules e-couch offers. Participants are additionally asked to fill out symptom self-report questionnaires 1,3,5,12,24 and 48 weeks after the start of e-couch engagement, and to repeat a subset of the behavioral tasks 3 weeks after the start of e-couch engagement.

All interaction with participants is conducted online via email (and possibly via zoom to verify identity or technical advice) and online-administered tasks, questionnaires and the self-help tool. The investigators do not offer any medical advice and forward participants to appropriate sources of support (e.g., hotlines) if needed.

Quality assurance plan:

All data are collected online through our in-house custom-built software. The code for the assessments has been reviewed and data quality has been checked prior to study start. The code is backed up on a secure server and the investigators can make the code available for review to relevant authorities.

Data checks:

The investigators use a range of attention checks throughout the study to exclude data from inattentive participants or participants who respond randomly.

Statistical analysis plan:

Note, the goal of this study is not to assess how much the self-help tools affect symptoms on average. This has been examined previously (and will only be verified in this dataset). The primary goal of this study is to develop a tool that predicts for a new individual (whose data was not part of the tool development, i.e. out-of-sample) what interventions they may benefit more from.

When consenting, participants are assigned to a discovery or a validation dataset. After completing analyses on the discovery dataset, the investigators will pre-register a detailed statistical analysis plan and apply that to the validation dataset to confirm and verify any findings. Below is the general approach to analyzing the discovery data:

Step 1) Compute scores for the self-report scales in line with the literature and fit computational models to data from the behavioral task to retrieve individual model parameters that best explain behavior the behavior from each participant.

Step 2) Use prediction models, e.g. elastic nets, to predict from the above parameters and scores, symptom scores (in particular, for depression and anhedonia) from each participant at the end of the engagement with the self-help tool and during follow ups, and/or improvement of symptom scores from before to after the engagement with the self-help tool.

Step 3) Examine whether change in task behavior due to the first half of engagement with the self-help tool mediates a change in symptoms from before to after that engagement.