Study protocol for a triple-blind randomised controlled trial evaluating a machine learning-based predictive clinical decision support tool for internet-delivered cognitive behaviour therapy (ICBT) for depression and anxiety.

Introduction: Therapist-supported internet-based Cognitive Behavioural Therapy (ICBT) has strong scientific support, but all patients are not helped, and further improvements are needed. Personalized medicine could enhance ICBT. One promising approach uses a Machine learning (ML) based predictive decision support tool (DST) to help therapists identify patients at risk of treatment failure and adjust their treatments accordingly. ICBT is a suitable clinical context for developing and testing such predictive DST's, since its delivery is quite flexible and can quickly be adapted for probable non-responders, for example by increasing the level and nature of therapist support, to avoid treatment failures and improve overall outcomes. This type of strategy has never been tested in a triple-blind randomised controlled trial (RCT) and has rarely been studied in ICBT.The aim of this protocol is to expand on previous registered protocols with more detailed descriptions of methods and analyses before analyses is being conducted. Methods and analysis: A triple blind RCT comparing ICBT with a DST (DST condition), to ICBT as usual (TAU condition). The primary objective is to evaluate if the DST condition is superior to the TAU condition in decreasing diagnose-specific symptoms among patients identified to be at risk of failure. Secondary objectives are to evaluate if the DST improves functioning, interaction, adherence, patient satisfaction, and therapist time efficiency and decreases the number of failed treatments. Additionally, we will investigate the therapists' experience of using the DST.Patients and therapists have been recruited nationally. They were randomised and given a sham rationale for the trial to ensure allocation blindness. The total number of patients included was 401, and assessments were administered pre-treatment, weekly during treatment, at post-treatment and at 12-month follow-up. Primary outcome is one of the three diagnosis-specific symptom rating scales for respective treatment and primary analysis is difference in change from pre- to post-treatment for at-risk patients on these scales. Human ethics and consent to participate: Informed consent to participate in the study was obtained from all participants. Both therapists and patients are participants in this trial. For patients, informed consent to participate in the study was obtained when they registered interest for the study via the study's secure web platform and carried out initial screening before the diagnostic and fit for treatment assessment, they first received the research subject information and were asked for consent by digitally signing that they had read and understood the information. For therapists who were part of the study, consent was requested after they had registered their interest. Therapists then received an email with a link to the study's secure web platform with the research person's information and were asked for consent by digitally signing that they had read and understood the information. All documents are stored in secure, locked filing cabinets on the clinic's premises or on a secure digital consent database. Approval committee: Approved by the Swedish Ethical Review Authority (SERA), record number 2020-05772.