Depression and anxiety are common mental health problems affecting around one in six adults. Treatments include therapy, telephone or group-based activities delivered by therapists. Treatment uses a "stepped care" model where most patients start with very brief treatments. If they remain unwell after this, they access longer and more intensive therapy. This model does not always work, as only one out of two patients fully recover.

The researchers have developed an artificial intelligence tool called "StratCare". It is designed to help health services to offer the best available treatment for each patient. StratCare is a computerised system that guides therapists on how to assess new patients. It asks a few questions about their symptoms, personality, and background. The system makes a recommendation about which treatment might be most effective for that person. This is either starting with brief therapy or starting with intensive therapy. The treatments are already used with depression in the National health Service (NHS). The patient can discuss with their therapist and decide whether to accept the recommendation. Otherwise, they can discuss trying other options. Previous research found that StratCare can help more people to recover from depression compared to the usual stepped care model.

The researchers will investigate if the StratCare tool works on a large scale in the NHS and if it helps patients in the long-term. The researchers will run a trial involving 1252 participants using NHS Talking Therapies services. Half will use the StratCare tool to make a treatment recommendation. The other half will follow the stepped care approach. The researchers will contact participants in the trial after 6, 12 and 18 months to see if their mood and quality of life has improved. The researchers will also interview therapists and participants to see what they think about treatment being guided by the StratCare tool.

Hypotheses:

• Stratified care will result in lower mean depression scores compared to usual stepped care (USC).
• Stratified care will result in a statistically significant higher proportion of cases with reliable and clinically significant improvement in depression symptoms, compared to USC.

Primary Outcome:

Depressive symptoms, as measured by change on the Patient Health Questionnaire (PHQ-9), at 12 months post-enrolment.

Trial Design: A pragmatic, single-blind, multi-site, parallel group cluster Randomised Controlled Trial (RCT), with an internal pilot and stop/go criteria for progression to the full RCT. The researchers will recruit 1252 adult participants seeking treatment for common mental health problems who present with case-level depression symptoms on the Patient Health Questionnaire (PHQ-9) measure. They will be recruited from 16 services in the National Health Service (NHS) Talking Therapies programme across England.

Enrolment: Cluster randomisation will be used to allocate Talking Therapies Teams and their patients to either StratCare or Usual Stepped Care (USC) arms. Sites may have one or more Talking Therapies Teams participating in the trial.

After General Practitioner (GP) or self-referral for treatment, participants will undergo a standard clinical assessment by Talking Therapies clinicians, during which a treatment option will be selected. In the experimental arm, treatment selection will be guided by an Artificial Intelligence (AI) tool (StratCare app). In the USC arm, treatment will be selected following usual clinical practice and guidelines. Participants will then follow their selected treatment option.

NHS Trusts will be recruited through a Practice Research Network, NHS data service provider (PCMIS), and other means, attending to considerations of diversity and generalisability outlined in section seven. Within trusts, individual Talking Therapies teams will be recruited to participate in the trial. Randomisation will occur at team level rather than individual level, to reduce the risk of potential contamination from clinicians becoming aware of how the StratCare app makes treatment decisions.

Participants are adult patients seeking and eligible for treatment for common mental health problems in NHS Talking Therapies services. Patients who are eligible for treatment in the Talking Therapies service will be offered the opportunity to take part in the trial at the point of initial suitability assessment by the service, if seen by a team and clinician participating in StratCare-2.

Informed consent will be received verbally from participants during their initial assessment appointment with the treating Talking Therapies service. The StratCare App will guide clinicians through a brief and standardised script to provide information and seek verbal consent from patients who they assess in routine care. The researchers have chosen to obtain only verbal consent to minimize additional burden to make this viable within the constraints of routine care and due to the minimal risks posed by a new treatment selection method between two routinely delivered treatments.

A post consent 'opt-out' system is in place, so that once they have seen the full PIS, participants can withdraw from the trial shortly after consent and remove all their data. After the consent and assessment appointment, participants are sent a link to the participant information sheet (PIS), which provides further information on the study and how to opt out. If participants decide to withdraw their consent within a week of being sent the PIS, they will be classed as an opt-out, and all their information will be removed from trial records.

Treatment: Talking Therapies teams randomised to the experimental group will implement an AI-driven stratified care treatment pathway, where participants are matched to specific treatments based on their clinical and demographic features. Participants will complete a suitability assessment with a qualified assessing clinician. The assessing clinician will enter the required data from the clinical assessment, participant's previous treatment history and StratCare demographic items into the StratCare app which will give a treatment recommendation of low- or high-intensity treatment. The participant and clinician discuss the assessment outcome and make a joint decision about treatment based on the StratCare App recommendation. This joint decision does not have to follow the recommendation of the StratCare App. The decision will be recorded in the StratCare App and clinical records, and the participant will then proceed to the waiting list for the agreed treatment. If the decision does not follow the StratCare App recommendation, the reason for this will be recorded.

Talking Therapies teams randomised to the USC arm will complete a standard suitability assessment for the Talking Therapies service with a qualified assessing clinician. The StratCare App will be used by the assessing clinician to record the necessary data from the clinical assessment, the participant's previous treatment history and StratCare demographic items but the App will not be used to make a treatment recommendation. Treatment recommendation decisions will be made in the usual way, following stepped care principles - where most patients initially access low-intensity treatments and can subsequently access high-intensity treatments if the first step of care is unsuccessful. The participant and clinician discuss the assessment outcome and make a joint decision about treatment based on the clinician's recommendation. The decision will be recorded in clinical records, and the participant will then proceed to the waiting list for the agreed treatment.

Whichever group the participant is randomised to, they will still access the usual evidence-based interventions available in routine Talking Therapies services. These include low-intensity guided self-help, usually lasting up to eight sessions, and high-intensity psychological therapies which can last up to 20 sessions. These interventions will not be modified in any way, to preserve the integrity of routinely delivered care.

Assessing clinicians will be qualified Talking Therapies Psychological Wellbeing Practitioners (PWPs) or qualified Talking Therapies clinicians. Talking Therapies teams will have access and training (three hours) to use the StratCare technology described above, and study processes. Assessing clinicians will also be provided with bespoke Good Clinical Practice (GCP) training to cover the GCP principles required for their safe and ethical involvement in the trial, and an online portal for refresher training.

Data Collection: Data will be collected in five ways, depending on the stage of the trial, the type of data being collected, whether the participant is still being treated by the Talking Therapies service, and participant preference.

1. via direct entry into the StratCare App by the assessing clinician (baseline only).
2. as part of clinical care within the Talking Therapies service, collected by Talking Therapies staff. Data collected this way will be entered into the study database by study staff, or where available using an automated download from clinical records, transferred securely using an encrypted system to the study database.
3. via an online data collection system, self-reported.
4. via paper forms, self-reported.
5. via telephone or online video call, collected by study staff.

Data collection windows will be +/- one month of the target date. Where more than one data point exists within the data collection window, the data closest to the target date will be used. In exceptional circumstances where data does not exist within the data collection window, data within +/- three months can be used.

Whilst participants are in treatment, there will be routine collection of the PHQ-9 and GAD-7 measures. This data will not be used for the primary or secondary analyses, unless trial-collected data is missing for those time points. In this case, the routinely collected PHQ-9 or GAD-7 scores closest to the target dates, and within the data collection window, may be used.

Beyond the AI-guided treatment suggestion made by the StratCare App in the experimental arm, participant reported outcome data will not inform the clinical care of individual trial participants.

Safety: Further detail of adverse event reporting can be found within the protocol.

Statistical Analysis: The statistical analysis will follow intention-to-treat principles and CONSORT guidelines for Cluster RCTs and it will be pre-registered in an international register for controlled trials. The unit of inference for the StratCare-2 RCT, i.e., the quantity of interest that is to be estimated in a statistical analysis, is the effect of the intervention on a typical individual. Hence, the researchers are interested in the 'participant-average treatment effect', which answers the question 'How effective is the intervention for the average participant?'

Qualitative Sub-study: Semi-structured interviews with a purposive sample of clinicians and participants who took part in the trial will be conducted to carry out a process evaluation investigating implementation barriers, enablers, and factors affecting adherence to the AI-driven stratified care model. This will focus on aspects of "explainable" and "ethical" use of AI: (a) whether participants understand and accept AI-driven recommendations, and (b) whether there are situations where algorithmic recommendations are deemed clinically inappropriate by clinicians. Qualitative interviews will also capture information about participants' and clinicians' experiences of the shared decision-making process. Interviews will be analysed by the qualitative researcher using framework analysis. This will be informed by Sekhon's acceptability of healthcare interventions framework and Normalisation Process Theory, which can help identify whether interventions are likely to become embedded and integrated as part of routine practice or not.

Health Economic Analysis: An economic analysis will be conducted from the NHS and Personal Social Services perspective over the 18-month study time-horizon. A cost-utility analysis will use quality-adjusted life years (QALYs derived from the EuroQol-5D (EQ-5D) questionnaire, and tariff based on the United Kingdom public value set) as the measure of quality of life.

Process Evaluation: The researchers will collect fully pseudonymized clinical pathway and outcomes data for all participants to characterise the full treatment pathway for patients. Both quantitative (e.g., anonymised electronic health records data) and qualitative data will be used to undertake a thorough process evaluation of the logic model. Data on reach, dose and fidelity will be reported alongside qualitative findings on implementation, following guidelines for the process evaluation of complex interventions.

Evaluation of Generalisability: As StratCare-2 is embedded within each participating NHS Talking Therapies service where they all collect the nationally mandated outcome measures at each attended therapy session, the researchers will design a direct test of the generalisability of results obtained by trial participants with the wider population of attendees at these services. The researchers will seek permission to download data for each participating service/team for the time period two years preceding the start of each service/ teams' participation until the discharge of the final participant at a service/team.

Patient and Public Involvement and Engagement (PPIE): To ensure genuine, consistent partnership with patients and the public, two experienced PPIE Co-Leads will lead all PPIE input. The PPIE co-leads will conduct all the activities described in funder guidance (e.g., setting and refining the PPIE strategy as the project progresses). The researchers will run regular PPIE groups of up to 12 participants, comprising adult volunteers with lived experience of mental health problems and Talking Therapies services.

Inclusion criteria:

• 18 years of age or older.
• Patients who consent to share their de-identified clinical records for research purposes.
• Patients assessed as eligible for psychological care in Talking Therapies based on clinical guidelines.
• Patients with case-level depression symptoms (PHQ-9 ≥ 10).

Exclusion Criteria:

• Those who are ineligible for treatment in Talking Therapies services according to standard treatment guidelines.

StratCare-2 will share its data using a controlled access, data repository approach. After the closure of the trial, simplified, minimised, anonymised datasets will be made available within a recognised data repository. This will be within a 'controlled access system' (i.e. data access requires approval and compliance with a formal data sharing agreement), in line with United Kingdom Clinical Trials Network (UKCRN) )recommendations. A study-specific Data Sharing Plan will be agreed and approved by the sponsor, Trial Management Group (TMG), Trial Steering Committee (TSC) and Clinical Trials Research Unit (CTRU) Quality Assurance team prior to any data being deposited or shared. This will outline where data is stored, what is stored, and how access to it is requested, reviewed and approved.

Trial statisticians will be blinded to allocation.