The Antidepressant Advisor Study

To address the urgent need for user-friendly decision support for antidepressant choices in UK primary care, the investigators will probe a specifically tailored innovative tool advising GPs on patients who have not responded to first-line treatments. Despite one promising US study, there is no such tool available in the UK. The US study was based in a private GP setting and employed now outdated treatment algorithms. In contrast, the treatment algorithm implemented in the current decision support tool is based on National Institute for Health and Care Excellence (NICE) guidelines including the latest health technology appraisals. The investigators have further gained the support of the leading UK provider of primary care electronic record systems (EMIS group: >3000 GP practice users) who will develop the software implementation of the algorithm as an add-on tool to their widely employed software.

Considering complex intervention guidance, the study is designed as a feasibility study to provide estimates of the following unknown variables needed to plan a subsequent larger trial: lost to follow-up and recruitment rates, GP satisfaction with the tool, impact on health service use, and standard deviations on our planned primary outcome measure on the self-rated Quick Inventory of Depressive Symptomatology (QIDS-SR16) for a subsequent trial. The investigators will use a single-blinded parallel group cluster-randomised controlled trial design with at least 8 GPs across 8 practices (one per practice) being randomised supported by the Clinical Research Network (CRN) to two arms: 1) Usual care of patients with depression with no computerised decision support, 2) Using the novel computerised decision support tool to assist with antidepressant choices.

The investigators will recruit 8-20 GPs in 8-20 different practices to avoid communication between GPs, each contributing approximately 10 study patients. The EMIS eligibility tool (see below) will identify potential participants by screening for antidepressant and problem history. Blinded practice staff/CRN will ask all eligible patients for consent for contact. Patients will fill in an electronic version of the self-report version of the Primary Care Evaluation of Mental Disorders (PRIME-MD) and provide electronic consent for this. Alternatively, participants will be called to answer the questions over the phone and provide oral consent. Other patients will prefer receiving a mailed printed questionnaire and provide written consent. PRIME-MD has been validated against clinical diagnoses of current major depressive and anxiety disorders, as well as alcohol abuse according to DSM-IV criteria and will be modified to screen for drug abuse and anxiety disorders may be dropped from the screening as they are not exclusion criteria and will be assessed in the in-depth assessment. Patients will further complete self-report versions of the WHO Composite International Diagnostic Interview (CIDI) screening scale for bipolar disorder and 3 screening questions validated in the investigators' group to exclude schizophrenia. Further, the investigators will ask about the inclusion and exclusion criteria for the study, because the information recorded on EMIS is not always complete and does not always include a patient's history with previous GP surgery for example. Eligible patients will be seen for an in-depth assessment by the study research associate (RA) where she will take written informed consent.

2.1.2 Data collection and Assessment for patients meeting the pre-screening criteria

As in the investigators' previous work, the RA will be closely supervised and trained, establishing sufficient inter-rater reliability on semi-structured interviews with the PI before carrying out assessments independently. Apart from the outcome measures, the in-depth assessment will include measuring past subthreshold hypomanic symptoms, and perceived credibility of and expectancy towards treatment using a dedicated questionnaire, and a detailed clinical evaluation whilst accessing patients' EMIS records including:

• Treatment history
• Medical history
• Age at onset, episode duration and number, illness duration
• International Neuropsychiatric Interview (MINI) suicidality screen
• DSM5 Structured Clinical Diagnostic Interview
• Addenbrooke's Cognitive Examination-R in patients over 50 to detect early Alzheimer's disease
• Young Mania Rating Scale The results of the assessment will be shared with GPs and patients. This will also allow the investigators to use the Maudsley Staging Method to compare the level of treatment resistance between treatment groups prior to entry into the study as a potential confounder. Patients who meet the inclusion/exclusion criteria will subsequently be seen by their GP who initiates and monitors treatment using EMIS. The decision whom to include will not be made by the GPs to ensure against different selection biases in the different treatment arms. The RA will be blinded to the intervention arm that a GP was randomised to, in order to prevent the RA from making biased inclusion/exclusion decisions. The RA will communicate inclusion/exclusion decisions to patients and will draft a short report to their GP. The RA will seek the Chief Investigator's advice on inclusion/exclusion questions without telling him which GP the patient was referred from. Should the RA become unblinded accidentally or for some other reason, this would be noted in the case record file, but the patient would be retained in the study. On the second and final RA visit 15-18 weeks after the baseline assessment, outcome measures and Young Mania Rating Scale will be repeated and the Longitudinal Interval Follow-up Interview will be used to determine remission, its psychiatric status rating scale will be used at baseline for comparison. After being enrolled, patients, who have no smartphone, receive a mobile smart phone to access a secure patient app (developed as part of the proposal which will communicate via secure email with GPs) to prompt them to enter weekly (changing the usual two week time interval to one week and using the Maudsley modified version) PHQ-9 ratings (chosen also because it has no licencing restrictions), to report hypomanic symptoms, fill in a validated self-report side effects scale (Frequency, Intensity, and Burden of Side Effects Rating (FIBSER), and enter medication (including opting in for a reminder function), as well as asking about self-blame-related action tendencies such as hiding. The app will be based on an adaptation of similar mood tracking apps developed by our BRC. Alternatively, ratings and medication compliances will be asked about at weekly intervals via post or phone which will be recorded in the case file.

Main objectives of this study

1. To develop the first computerised decision support tool for antidepressant treatment in UK primary care
2. To probe the feasibility of a clinical trial to assess the tool's efficacy by
3. estimating lost to follow-up rates
4. estimating GP and patient adherence to prescribed medications
5. determining the number of GP practices willing to recruit patients for the study (determined by the CRN who will approach all practices in the participating CCGs)
6. estimating the recruitment rates per GP
7. estimating GP satisfaction
8. To provide standard deviation estimates and intra-class correlation coefficients per GP for computing effect size estimates for improving treatment outcomes in preparation of larger subsequent trials
9. To collect health economic estimates of changes in service use associated with the new tool (including psychiatric referrals to mental health teams and/or the study psychiatrist).

As soon as possible after their baseline in-depth assessment, eligible patients will undergo treatment over 14 weeks with their GPs which allows 3 weeks for determining sufficient treatment response at low and high dose of each of the two different recommended medications and 1 week for cross-tapering before step 1 and step 3. The final assessment by the RA will take place 15-18 weeks after the baseline assessment. Patients' participation will therefore last for approximately 15-18 weeks. Deviations from this time will occur due to scheduling difficulties and will be recorded. Time in the study will be used as a covariate in secondary data analyses.

GPs will be randomised into two groups, asking each GP to enroll approximately 8-11 participants each, adding more GPs if necessary. The study aims to recruit 86 participants assuming the same lost to follow-up rate as in the US trial (18%), resulting in a final sample size of 70 (35 in each group as recommended). This will enable the investigators to estimate the lost to follow-up rate within a 95% confidence interval of +/-8%. Because the comparable previous trial did not provide effect size estimates, this study has been designed to provide means and standard deviations, as well as confidence intervals for measures of change on the primary outcome measure (QIDS-SR16) in n=35 per group as recommended for feasibility trials.

Double data entry will be employed. Categorical outcomes (e.g. lost to follow-up rate) will be described using appropriate summary statistics. The QIDS-SR16 and other continuous outcomes will be summarised at baseline and final assessment time points to obtain means and standard deviations for a larger trial sample size calculation, with the GP intra-class correlation calculated for the outcome variable using one-way random effects analysis of covariance (adjusted for baseline). The investigators plan to do a preliminary analysis of the difference between the groups, as far as possible using the intention to treat principle. This analysis will be identified as preliminary and underpowered when published. The outcomes measured at baseline and follow-up only, such as the QIDS-SR16, will be analysed using robust linear regression to account for clustering within GP. Continuous outcomes measured weekly will be analysed using mixed linear regression models with an intercept for GP to account for clustering. Both types of models will include treatment group as a covariate, in order to estimate differences between the two intervention groups, and will adjust for the baseline measure of the outcome where appropriate. The investigators do not expect missing baseline data, however, any such missing data will be imputed using mean imputation. Missing data in the weekly outcomes will be accounted for under the missing at random assumption by using the maximum likelihood algorithm to fit the mixed models. The investigators may consider multiple imputation for outcomes measured only at follow-up if post-randomisation adherence variables can be quantified and are related to having missing outcomes.