The overall aim of this study is to test the clinical benefits of an AI-based decision support tool (DST) and to evaluate how the DST affects therapists and their patients with depression, social anxiety, or panic disorder during 12 weeks of ICBT.

The project aims to evaluate if therapist-supported ICBT with a DST that guides therapists in addition to a traditional therapist manual (the DST arm) are superior to ICBT using only the therapist manual (TRAD) arm.

The investigators hypothesize that DST, in comparison to TRAD, will:

• decrease the diagnose-specific symptoms during the treatment period, among patients identified to be at risk of failure (primary outcome),
• decrease the number of failed treatments among patients identified to be at risk,
• improve everyday functioning, health related quality of life, patient satisfaction, number of Adverse Events experienced by the patient, need for further treatment and sudden symptom changes among patients at risk,
• Increase the therapists' amount of interaction and time spent on patients at risk,
• increase the adherence to treatment among patients at risk, and
• improve levels of symptoms, functioning, interaction, and adherence when all patients (also those not at risk) are included
• make therapists overall more time efficient, defined as the ratio of 'decrease in symptoms / therapist time spent on patient'
• increase therapist experience of using the DST / the therapist manual, including higher perceived helpfulness and credibility and the overall experience of supervision, clinical routines, and guidance of their clinical decisions

The study is designed as a triple blind randomized controlled trial, where half of the therapists are allocated to DST and the other half to TRAD. Each patient is also randomized between those conditions. The therapists and patients will be blind to what group they are allocated to and the post-treatment assessors will also be blind to what group the patients belong to. The therapists and patients will know that they are randomized, but will be informed that the randomization concerns two different types of clinical routines and supervision, and will thus be blind to the true purpose of randomization. After randomization both groups will be informed that they are in the experimental group to lower the risk for nocebo effects.

The DST provides visual feedback about patient current adherence and symptoms levels, as well as predictions of final outcome and adherence, and also use four color indicators to guide the therapist:

• Green: Patient likely to succeed, consider spending less time on patient.
• Yellow: Prediction too uncertain to give guidance.
• Light Red: Patient likely to fail, consider adjustments.
• Dark Red: Patient very likely to fail, adjustments must be implemented.

A patient is considered to be at risk to fail if he/she gets a Light Red or Dark Red indication.

Primary outcome is the diagnose-specific symptom rating for each of the three ICBT-treatments that are used in the trial for Depression (MADRS-S), Panic Disorder (PDSS-SR), and Social Anxiety Disorder (LSAS-SR) respectively. Failed treatment are defined as being neither a responder (50% symptom reduction) nor a remitter (under a previously specified cut-off for each respective diagnose-specific symptom measure).

Primary analyses are performed according to intent-to-treat and, to handle missing data, with an HLM-model that test the interaction of Time x Arm and includes all available symptom data from pre to post, including weekly measures during treatment. It includes the estimated 50% of all patients that will be indicated as at risk (i.e. also control patients in TRAD, although their therapists never see the prediction because they do not use the DST). In a previous proof-of-concept trial the effect corresponding to our primary analysis was Cohen's d=0.59. To be conservative it was estimated to be 0.50. A power of 80% and an attrition of 20% would require 158 patients identified as at risk of failure, and thus 316 in total. To increase the less easily estimated power for secondary measures, the aim is to include 350 patients.

The trial will follow the guidelines of Good Clinical Practice adapted for psychological treatment.

Inclusion Criteria:

• 18 years or older.
• Social anxiety, panic disorder or depression diagnosis.
• Stable or no antidepressant medication for at least 2 months.
• No diseases, disorders, or substance abuse that required other, immediate attention (e.g., severe depression or suicidality).
• Available time for treatment and acceptance of its format.
• Proficient in Swedish.
• No ongoing CBT.

Exclusion Criteria:

• Not available for assessment and diagnostic telephone interview.
• Will not be in Sweden during the treatment period.
• Not able to receive text messages on a Swedish mobile phone.
• Not proficient in Swedish.
• No access to computer and internet.
• Not able to set aside about one hour a day to work on treatment.
• Bipolar disorder, if seeking depression treatment.

The therapists and patients will know that they are randomized, but will be informed that the randomization concerns two different types of clinical routines and supervision, and will thus be blind to the true purpose of randomization. Outcome assessors will not be informed that any randomization has occurred.