Cognitive Behavioural Therapy for Psychosis

BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from

BMJ Open is committed to open peer review. As part of this commitment we make the peer review history of every article we publish publicly available.

When an article is published we post the peer reviewers’ comments and the authors’ responses online. We also post the versions of the paper that were used during peer review. These are the versions that the peer review comments apply to.

The versions of the paper that follow are the versions that were submitted during the peer review process. They are not the versions of record or the final published versions. They should not be cited or distributed as the published version of this manuscript.

BMJ Open is an open access journal and the full, final, typeset and author-corrected version of record of the manuscript is available on our site with no access controls, subscription charges or pay-per-view fees (http://bmjopen.bmj.com).

If you have any questions on BMJ Open’s open peer review process please email [email protected] http://bmjopen.bmj.com/ on September 26, 2021 by guest. Protected copyright. BMJ Open BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from

For whom is Cognitive Behavioural Therapy (CBT) for psychosis most effective? Protocol for an IPD meta-analysis of randomised control trials comparing CBT versus standard care and other psychosocial interventions (Cognitive Behaviour Therapy for Psychosis: Individual Modifiers of ForPatient peer Response review to Treatment) only

Journal: BMJ Open

Manuscript ID bmjopen-2019-035062

Article Type: Protocol

Date Submitted by the 17-Oct-2019 Author:

Dunn, Graham; University of Manchester, Health Methodology Research http://bmjopen.bmj.com/ Kendall, Tim; National Collaborating Centre for Mental Health Emsley, Richard; King’s College London, Department of Biostatistics and Health Informatics, Institute of Psychiatry, Psychology and Neuroscience Morrison, Anthony; University of Manchester Varese, Filippo; Manchester Academic Health Science Centre,

Keywords: IPD-MA, Treatment Modifier, CBT, Psychosis

on September 26, 2021 by guest. Protected copyright.

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 1 of 24 BMJ Open

1 2 3

4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 8 9 I, the Submitting Author has the right to grant and does grant on behalf of all authors of the Work (as defined 10 in the below author licence), an exclusive licence and/or a non-exclusive licence for contributions from authors 11 who are: i) UK Crown employees; ii) where BMJ has agreed a CC-BY licence shall apply, and/or iii) in accordance 12 with the terms applicable for US Federal Government officers or employees acting as part of their official 13 duties; on a worldwide, perpetual, irrevocable, royalty-free basis to BMJ Publishing Group Ltd (“BMJ”) its 14 licensees and where the relevant Journal is co-owned by BMJ to the co-owners of the Journal, to publish the 15 Work in this journal and any other BMJ products and to exploit all rights, as set out in our licence. 16 17 The Submitting Author accepts and understands that any supply made under these terms is made by BMJ to 18 the Submitting Author Forunless you peer are acting as review an employee on behalf only of your employer or a postgraduate 19 student of an affiliated institution which is paying any applicable article publishing charge (“APC”) for Open 20 Access articles. Where the Submitting Author wishes to make the Work available on an Open Access basis (and 21 intends to pay the relevant APC), the terms of reuse of such Open Access shall be governed by a Creative 22 Commons licence – details of these licences and which Creative Commons licence will apply to this Work are set 23 out in our licence referred to above. 24 25 Other than as permitted in any relevant BMJ Author’s Self Archiving Policies, I confirm this Work has not been 26 accepted for publication elsewhere, is not being considered for publication elsewhere and does not duplicate 27 material already published. I confirm all authors consent to publication of this Work and authorise the granting 28 of this licence. 29 30 31 32 33 34 35

36 http://bmjopen.bmj.com/ 37 38 39 40 41 42 43

44 on September 26, 2021 by guest. Protected copyright. 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 2 of 24

1 2 3

4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 For whom is Cognitive Behavioural Therapy (CBT) for 6 7 8 9 psychosis most effective? Protocol for an IPD meta-analysis of 10 11 12 randomised control trials comparing CBT versus standard care 13 14 15 16 and other psychosocial interventions (Cognitive Behaviour 17 18 For peer review only 19 Therapy for Psychosis: Individual Modifiers of Patient 20 21 22 23 Response to Treatment) 24 25 26 Maria Sudell1,a*, Catrin Tudur Smith1,b, Xiaomeng Liao1,c, Eleanor Longden3,d, Graham Dunn4, 27 28 Tim Kendall5, 6,e, Richard Emsley7,f & Anthony P. Morrison2,3,g Filippo Varese2,3,h 29 30 1Institute of Translational Medicine, University of Liverpool; 2School of Health Sciences, 31 32 Division of Psychology and Mental Health, Faculty of Biology Medicine and Health, University 33 3 34 of Manchester; Greater Manchester Mental Health NHS Mental Health Foundation Trust; 35 4Institute of Population Health, University of Manchester; 5National Collaborating Centre for

36 http://bmjopen.bmj.com/ 37 Mental Health; 6Sheffield Health and Social Care NHS Foundation Trust; 7Department of 38 39 Biostatistics & Health Informatics, Kings College London; [email protected]; 40 41 [email protected]; [email protected]; [email protected]; 42 e f 43 [email protected]; [email protected];

44 g h on September 26, 2021 by guest. Protected copyright. 45 [email protected] ; [email protected]; 46 *Corresponding author 47 48 49 University of Manchester 50 Division of Psychology and Mental Health 51 52 School of Health Sciences 53 Zochonis Building, 2nd floor; room 2.40 54 Manchester 55 M13 9PL 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 3 of 24 BMJ Open

1 2 3 4 Article Summary BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 8 Introduction: Aggregate data meta-analyses have shown heterogeneous treatment effects 9 10 for Cognitive Behavioural Therapy (CBT) for patients with schizophrenia-spectrum 11 12 diagnoses. This heterogeneity could stem from specific intervention or patient 13 characteristics that could influence the clinical effectiveness of CBT, termed treatment 14 15 effect modifiers. This Individual Participant Data Meta-Analysis (IPD-MA) will investigate a 16 17 range of potential treatment effect modifiers of the efficacy CBT. Methods and analysis: We 18 For peer review only 19 will perform a systematic review and meta-analysis of studies investigating CBT vs treatment 20 21 as usual (TAU), or CBT vs other psychosocial interventions, for patients with schizophrenia- 22 spectrum diagnoses. The Cochrane Central Register of Controlled Trials (CENTRAL), PubMed, 23 24 EMBASE and the online clinical trials registers of the US government, European Union, 25 26 World Health Organization and Current Controlled Trials Ltd will be searched. Two 27 28 researchers will screen titles and abstracts identified by the search. Individual participant 29 30 data will be requested for any eligible study, for the primary outcome (overall psychotic 31 32 symptoms), secondary outcomes and treatment effect modifiers. Data will be checked and 33 recoded according to an established Statistical Analysis Plan. One and two-stage random 34 35 effects meta-analyses investigating potential treatment effect modifiers will be conducted.

36 http://bmjopen.bmj.com/ 37 A list of potential treatment effect modifiers for CBT will be produced, motivating future 38 39 research into particular modifiers. Ethics and dissemination: This study does not require 40 41 ethical approval as it is based on data from existing studies, although best ethical practice 42 43 for secondary analysis of clinical data will be followed. The findings will be submitted for

44 publication in peer-reviewed journals, and promoted to relevant stakeholders. on September 26, 2021 by guest. Protected copyright. 45 46 47 48 Prospero registration number 15/187/05 49 50 51 Strengths and Limitations of study 52 53 54  This will be the first published IPD-MA to investigate treatment effect modifiers for 55 56 CBT for patients with schizophrenia-spectrum diagnoses 57  The review will consider the efficacy of CBTp across multiple outcomes of interest in 58 59 addition to psychotic symptoms severity 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 4 of 24

1 2 3  The search will be conducted without geographical, language or time restrictions 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5  A potential limitation of this study will be the degree of heterogeneity between 6 7 studies in recorded measures and scales employed. 8 9 10 11 Introduction 12 13 14 Cognitive behavioural therapy (CBT) is a recommended intervention for the treatment and 15 16 management of psychosis and schizophrenia [2]. Aggregate Data Meta-Analyses (AD-MA) 17 18 suggest that CBT forFor psychosis peer has modest review but considerably only heterogeneous treatment 19 20 effects (for example [3, 4]). This inconsistency partly stems from inter-trial variation in 21 22 several key methodological characteristics of the existing Randomised Controlled Trials 23 24 (RCT; for example, blinding/masking of outcome assessments [3, 4]) but could also reflect 25 the impact of unaccounted clinical heterogeneity i.e. specific intervention and patient 26 27 characteristics that can potentially influence the clinical effectiveness of CBT. For instance, 28 29 previous trials differed widely in terms of intervention characteristics (e.g. number of 30 31 treatment sessions, treatment duration, use of manualised interventions), patients’ baseline 32 33 severity of psychotic and other comorbid symptoms, their demographic characteristics (e.g. 34 35 age, gender and ethnic origin), and illness duration. The identification of moderators of

36 treatment response and/or sub-groups of patients who may particularly benefit from CBT http://bmjopen.bmj.com/ 37 38 would allow optimisation of treatment delivery, with significant implications in terms of 39 40 improved clinical effectiveness, cost savings and maximisation of patients’ informed choice 41 42 of treatment. 43

44 on September 26, 2021 by guest. Protected copyright. 45 46 The impact of these potential treatment effect modifiers, however, remains unaccounted 47 for (or at best poorly estimated) in AD-MA due to their reliance of the reporting quality of 48 49 primary studies and the limited statistical power of “standard” meta-analytic methods for 50 51 testing treatment effect moderators [5]. Additional primary research would be costly and 52 53 impractical given the large sample size required. The only approach suited for this type of 54 55 research is Individual Patient Data Meta-Analysis (IPD-MA), a research synthesis method 56 57 which summarises the evidence on a particular clinical question by considering individual 58 participant level rather than aggregate level data from multiple related studies. IPD-MA 59 60 allow (1) greater ability to examine the impact of multiple individual-level and study-level

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 5 of 24 BMJ Open

1 2 3 factors (and their combination) on the treatment effects considered, (2) standardisation of 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 statistical methods used across studies, and (3) the potential of reduced risk of bias e.g. due 6 7 to selective reporting of outcomes compared to conventional AD-MA [1, 6-8]. This article is 8 9 the protocol for an Individual Patient Data Meta-Analysis to assess the impact of treatment 10 11 effect modifiers of CBT. 12 13 14 Objectives of IPD-MA 15 16 17 18 This research will identifyFor treatment peer effect review modifiers of CBT only in people with diagnoses in the 19 20 schizophrenia-spectrum using IPD-MA methods to combine data from multiple RCTs. 21 22 23 Treatment comparisons 24 25 26 27 Consistent with the analytic approaches in recent AD-MA [13, 14] our IPD-MA will 28 29 distinguish between “pure” CBT interventions as defined by NICE [2] and “CBT+” 30 interventions, where CBT+ is defined as a CBT treatment packages incorporating significant 31 32 elements of other, distinct psychosocial intervention approaches (e.g. mindfulness, 33 34 motivational interviewing, family intervention) alongside core CBT elements. CBT and CBT+ 35 will be compared to Treatment As Usual (TAU) or other psychosocial interventions in 36 http://bmjopen.bmj.com/ 37 38 individuals with a diagnosis of schizophrenia-spectrum disorders. TAU is defined as the level 39 40 of care service users would routinely receive had they not been involved in the trial, other 41 psychosocial interventions is defined as TAU supplemented by additional psychological or 42 43 social interventions e.g. family therapy. Although all trials of CBT against these comparators

44 on September 26, 2021 by guest. Protected copyright. 45 will be eligible, these will be synthetized and contrasted in separate analyses for: (1) CBT vs 46 47 TAU, (2) CBT vs other psychosocial interventions, (3) CBT+ vs TAU, (4) CBT+ vs other 48 49 psychosocial interventions. 50 51 52 Treatment Effect Modifiers 53 54 55 56 The selection of treatment effect modifiers was informed by (1) knowledge of variables 57 examined in previous and on-going RCTs by members of our team and the collaborators in 58 59 the CBTp: IMPART Consortium; (2) the findings of studies which examined predictors of 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 6 of 24

1 2 3 outcomes in previous RCTs [e.g. 9, 10-12], and (3) consultation meetings with service-users 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 with psychosis and clinical psychologists and CBT therapists working with clients with 6 7 psychosis in secondary care settings in the UK. Secondary outcomes were selected based on 8 9 (1) outcomes often targeted in CBT for psychosis, (2) measures identified as valuable in our 10 11 PPI consultations 12 13 14 The treatment effect modifiers shown in Table 1 will be investigated for both the primary 15 and secondary outcomes. Treatment effect modifiers have been separated into three main 16 17 groups: (1) Participant Demographic Characteristics, (2) Participant’s Clinical (3) Specific 18 For peer review only 19 Intervention Characteristics 20 21 22 Treatment effect Modifier 23 24 25 Participant’s Demographic Characteristics Age at entry to trial 26 27 28 Gender 29 30 31 Ethnicity 32 33 Participant’s Clinical Characteristics Effect of specific diagnostic sub-groups 34 35

36 Phase of illness (first episode psychosis / http://bmjopen.bmj.com/ 37 38 multiple episode psychosis) 39 40 41 Illness Duration 42 43 Duration of untreated psychosis

44 on September 26, 2021 by guest. Protected copyright. 45 46 47 Initial Severity of Psychotic Symptoms 48 49 (measured by baseline PANSS scores) 50 51 Initial severity of comorbid affective 52 53 symptoms (measured by baseline Anxiety 54 55 scores) 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 7 of 24 BMJ Open

1 2 3 Initial severity of comorbid affective 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 symptoms (measured by baseline 6 7 Depression scores) 8 9 10 Dosage equivalence of baseline anti- 11 12 psychotic medication(s) [15] 13 14 15 Number of anti-psychotic medications 16 received at baseline) 17 18 For peer review only 19 Specific Intervention Characteristics time period over which treatment was 20 21 delivered* 22 23 24 number of therapy sessions offered in the 25 26 study* 27 28 29 number of therapy sessions attended by 30 31 the individual 32 33 Minimum study required level of therapist’s 34 35 training and competence*;

36 http://bmjopen.bmj.com/ 37 38 Measures of therapeutic alliance; 39 40 41 Use of manualised interventions*; 42 43

44 Use of formulation-based interventions*; on September 26, 2021 by guest. Protected copyright. 45 46 47 Indicator for whether the intervention was 48 49 designed to target the outcome under 50 scrutiny*; 51 52 53 Individual versus group interventions* 54 55 56 Length of study follow up 57 58 59 Table 1: Treatment effect modifiers examined in this IPD-MA. Note: treatment effect modifiers marked with * are study 60 level variables, the remaining are individual level variables

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 8 of 24

1 2 3

4 Primary outcome BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 The primary objective is to identify treatment effect modifiers for CBT or CBT+ vs TAU or 8 9 other psychosocial interventions on overall psychotic symptoms severity as measured by 10 11 PANSS scores [16, 17] in patients with schizophrenia-spectrum diagnoses. 12 13 14 15 Secondary outcomes 16 17 18 Treatment effect modifiersFor forpeer CBT or CBT+ review vs TAU or other only psychosocial interventions will 19 20 be examined for the secondary outcomes listed in Table 2. For some secondary outcomes, 21 22 multiple sub-outcomes will be examined, such as Change in severity of affective symptoms, 23 which is examined both for anxiety and depression. Adverse event information will be 24 25 sought specifically for the 4 sub-outcomes mentioned, but additional adverse event 26 27 information will also be tabulated. 28 29 30 31 Minimum clinical important differences (MCIDs) in PANSS scores 32 33 ≥ 11 points 34 35 ≥ 15 points

36 Change in specific symptom clusters http://bmjopen.bmj.com/ 37 38 Positive psychopathology 39 40 Negative psychopathology 41 42 General psychopathology 43

44 Change in specific symptoms often targeted in CBT for psychosis on September 26, 2021 by guest. Protected copyright. 45 46 Hallucinations severity 47 48 Delusions severity 49 50 Hallucinations associated subjective distress 51 Delusions associated subjective distress 52 53 Paranoia severity 54 55 Change in severity of affective symptoms 56 57 Anxiety 58 59 Depression 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 9 of 24 BMJ Open

1 2 3 Subjectively defined recovery 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 Quality of Life 6 7 Social and Occupational functioning 8 9 Early Treatment Discontinuation 10 11 Adverse effects 12 13 Deaths 14 15 Attempts at suicide 16 17 Suicide ideation 18 Serious violentFor incidents peer review only 19 20 Hospital readmissions 21 22 Table 2: Secondary outcomes 23 24 25 Methods 26 27 28 29 30 Protocol 31 32 33 This evidence synthesis will follow state-of-the-art guidelines for IPD-MA, and our outputs 34 35 will comply as a minimum with the PRISMA statement for the reporting of IPD-MA [1]. The

36 http://bmjopen.bmj.com/ 37 search strategy of our IPD-MA builds on the protocol and database searches carried out as 38 39 part of a recent AD-MA carried out by members of our team [13, 14]. Our study selection 40 criteria are consistent with those employed in this recent AD-MA. Similarly, our literature 41 42 searches will update those carried out as part of this review to identify any RCTs that have 43

44 become available since the date of search. on September 26, 2021 by guest. Protected copyright. 45 46 47 Search Strategy 48 49 50 51 We will update the searches already conducted as part of a recent meta-analysis of 52 53 aggregated data [13, 14] to identify any trials that might have been become available for 54 55 research synthesis since the date of last search. Database searches will be conducted on the 56 Cochrane Central Register of Controlled Trials (CENTRAL), PubMed, EMBASE and the online 57 58 clinical trials registers of the US government, European Union, World Health Organization 59 60 and Current Controlled Trials Ltd.

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 10 of 24

1 2 3 In line with the protocol of the AD-MA carried out by members of our team [13, 14], titles, 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 abstracts and keywords will be searched in the publication databases using the adaptations 6 7 of following generic strategy: (schizo$ [exp. schizophrenia + psychosis + schizoaffective]) 8 9 AND (trial [exp. RCT +controlled trial + clinical trial]) AND (cbt [exp. cognitive therapy + 10 11 behaviour therapy +psychotherapy]). 12 13 14 Selection of Studies 15 16 17 18 The project’s PI andFor another peer member of reviewthe research team only will screen titles and abstracts 19 for relevance, and subsequently assess eligibility by examining the full-text reports against 20 21 the above-mentioned criteria. When required, additional information to ascertain eligibility 22 23 will be requested from the RCTs’ authors and through the examination of treatment manual 24 25 when available (e.g. to ascertain that interventions complied with NICE operational 26 27 definitions of CBT). Discrepancies in selection decisions will be discussed, and arbitration by 28 29 other members of the research team sought to achieve consensus. We will include studies 30 based on the following eligibility criteria: 31 32 33 34 Participants 35

36 Trials where > 50% of participants have diagnoses in the schizophrenia-spectrum http://bmjopen.bmj.com/ 37 38 (schizophrenia, schizoaffective disorder or early psychosis) will be eligible. Trials where > 39 40 50% participants have an established diagnosis of bipolar disorder, intellectual disability, 41 42 psychosis secondary to a general medical condition or organic pathology, or a primary 43 diagnosis of substance-induced psychosis will be excluded. No restriction will be placed on

44 on September 26, 2021 by guest. Protected copyright. 45 participants’ age, ethnicity, illness severity and illness duration. 46 47 48 49 Treatment Comparisons 50 51 Included studies must compare treatments that fit into one or more of the four stated 52 53 treatment categories (CBT, CBT+, TAU, other psychosocial interventions). Study 54 55 interventions will be classified into one of these four categories by clinicians involved with 56 57 this project. 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 11 of 24 BMJ Open

1 2 3

4 Outcomes BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 The study must provide data for one or more of the stated primary or secondary outcomes. 7 8 Data can be recorded on any comparable scale, at any time-point. Data will also be sought 9 10 for the treatment effect modifiers listed in Table 1. 11 12 13 Study Designs 14 15 16 Parallel single-blind or open controlled trials with at least two arms utilizing random 17 18 allocation to treatmentFor will bepeer considered review for inclusion. Single only arm or cross-over studies will 19 20 not be eligible for inclusion in the review. Studies employing other research designs (case 21 22 series, cohort analyses) will not be eligible. 23 24 Trials will be eligible if they evaluated CBT or CBT+ interventions versus TAU or other 25 26 psychosocial treatments, eligible treatments as defined in Treatment Comparisons Section. 27 28 29 30 Data Collection and Processes 31 32 33 IPD will be collected from principal investigators of past and on-going RCTs of CBT for 34 35 psychosis in the UK and internationally. Our ability to successfully collect relevant IPD is

36 http://bmjopen.bmj.com/ 37 facilitated by several factors. Our research team includes researchers who have conducted 38 39 some of the largest RCTs in this area. Furthermore, we have established a network of 40 collaborators to support the retrieval of relevant IPD: the CBTp IMPART Consortium. We will 41 42 continue to expand the CBTp IMPART Consortium over the lifetime of the project by sending 43

44 invitation emails to all researchers who have published and/or are currently conducting on September 26, 2021 by guest. Protected copyright. 45 46 randomised controlled trials relevant to this work. Participating researchers will be sent 47 48 specific data request forms outlining variables pertinent to the present IPD-MA. They will be 49 50 asked to fully anonymise the requested dataset and share them with our research team 51 using a safe data transfer system provided by the IT services University of Liverpool. We will 52 53 remain in regular contact with all Consortium members throughout the lifetime of the 54 55 project to clarify queries about their IPD and its integrity. 56 57 58 A statistical analysis plan detailing the data cleaning and coding, and the analyses to be 59 60 conducted has been produced and will be available upon request. Data received will be

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 12 of 24

1 2 3 systematically recoded to ensure common scales or measurements across studies. We will 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 liaise with principal investigators and statisticians of the primary studies to resolve any data 6 7 issues and prepare the dataset for IPD-MA. In addition to primary and secondary outcomes 8 9 data, IPD and relevant supporting material (e.g. trial codebook, therapy manuals, statistical 10 11 analysis plans) will be requested to code the stated treatment groups, treatment effect 12 13 modifiers, primary and secondary outcomes. 14 15 The primary outcome will be analysed on the PANSS scale, with comparable information 16 17 recorded on other scales converted onto the PANSS scale where possible (e.g. Leucht et al 18 For peer review only 19 [18] provide supplemental tables to convert between PANSS and BPRS). Secondary 20 21 outcomes will be transformed onto the most commonly reported scale, where established 22 23 conversion tables exist. If transformation to a common scale is impossible, or there is no 24 25 most common scale, standardised values (calculated by dividing by the between patient 26 variation) will be employed [19]. Transformation onto sub-scales will not be attempted. For 27 28 example, secondary outcome Change in specific symptom clusters: Positive is defined as the 29 30 sum of the positive sub-scale of the PANSS score. If a study records a comparable measure, 31 32 no attempt will be made to transform this measure onto the positive sub-scale of PANSS, 33 34 however the data would contribute to a standardized score analysis e.g. as part of a 35 sensitivity analysis.

36 http://bmjopen.bmj.com/ 37 38 39 Statistical Analysis 40 41 42 43 Data Analysis

44 on September 26, 2021 by guest. Protected copyright. 45 46 All randomised patients will be included, and an intention to treat (ITT) principle will be 47 48 followed throughout. To examine IPD integrity and concordance with original trial analyses, 49 50 all trials will be re-analysed individually and the original authors asked to confirm the 51 52 individual study results and resolve any discrepancies. Throughout, a frequentist approach 53 to analyses will be taken. All analyses will be conducted using R [20]. If the planned 54 55 quantitative analyses cannot be undertaken, the data will be described qualitatively.We will 56 57 examine the pooled treatment effect for each outcome by performing a series of one-step 58 59 IPD meta-analyses (where IPD from individual studies are analysed simultaneously whilst 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 13 of 24 BMJ Open

1 2 3 accounting for study-level clustering), and two-step IPD meta-analyses (where estimates of 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 the treatment effect are initially computed from the IPD of each study, and then aggregated 6 7 using conventional inverse variance meta-analytic approach) [6]. Throughout, due to 8 9 anticipated heterogeneity between studies, a random effects approach will be employed 10 11 (using a Dersimonian and Laird approach for two-stage analyses [21], and including a study 12 13 level random treatment effect, and fixed study membership effect in one-stage approaches 14 15 Both one stage and two stage analyses will be conducted to allow a full investigation of the 16 17 data [10]. The two stage analyses will be conducted initially, to help identify areas of higher 18 For peer review only 19 between study heterogeneity through examination of e.g. forest plots of analyses. One 20 21 stage analyses will then be conducted to allow multiple interactions between treatment 22 23 modifiers to be examined. One and two stage analysis results will be compared, to confirm 24 25 that areas of heterogeneity are identified similarly between each approach. 26 27 Due to considerable variation in the follow-up periods considered in the original trials, 28 29 separate analysis of trials with highly comparable or identical points of assessment (e.g. 3 30 31 month, 6 month etc.) would be unfeasible. In order to maximise the data contributing to the 32 33 analysis, outcomes measured at multiple time points will be modelled longitudinally. As 34 35 dropout may be an issue during study periods, we will employ joint modelling methods (e.g.

36 [22, 23]) to account for study dropout. Outcomes measured at a single time point (e.g. at http://bmjopen.bmj.com/ 37 38 the end of the treatment period) will be analysed using generalised linear models. 39 40 41 In two-stage analyses, treatment-modifier interactions will be estimated within each study, 42 43 and the results pooled. In one-stage analyses, treatment-moderators interactions

44 on September 26, 2021 by guest. Protected copyright. 45 separating out within and between study effects will be examined, while accounting for 46 47 clustering of participants within studies [6, 7, 24]. Any treatment effect modifier found to be 48 significant at a level of 0.05 will be retained in a list of potential treatment effect modifiers. 49 50 Treatment effect modifiers with a significant effect for each outcome will be identified 51 52 through examination of 95% Confidence Intervals (CI). Confidence intervals for both joint 53 54 and GLM analyses will be calculated through bootstrapping, using 200 bootstrap samples. 55 56 Once this list has been compiled, both forward and backward manual selection procedures 57 58 for model parameters will be conducted for the one and two stage analyses of the primary 59 and secondary outcomes, provided sufficient data is available. The overlap between the 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 14 of 24

1 2 3 parameters selected by the forward and backward selection procedures will highlight, from 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 the list of potential treatment effect modifiers, parameters most likely to be true treatment 6 7 effect modifiers. 8 9 10 This investigation involves a large number of planned analyses. However, this analysis of 11 12 treatment modifiers of CBT is, to an extent, exploratory. As such, this investigation aims to 13 14 identify potential relationships, and in so doing motivate future investigations specifically 15 targeting the identified areas of interest. Consequently, and given that there is not a 16 17 standard multiple testing approach currently recommended for IPD-MA, we will not adjust 18 For peer review only 19 for multiple testing, although we reiterated that these analyses are, to an extent, 20 21 exploratory. If methods are developed during the course of this project that are 22 23 recommended as standard to account for multiple testing in IPD-MA, application of the 24 25 methods will be examined. 26 27 The main analyses will be conducted as complete case analyses, i.e. only those contributing 28 29 data for all variables (outcome or explanatory) included in the model will be used in the 30 31 meta-analyses. If the level of missing data for treatment effect modifiers or outcomes is 32 33 large across the studies included in the meta-analysis, if possible, the effect of missing data 34 35 on the conclusions of the analyses will be investigated by re-conducting the analyses of the

36 primary and secondary outcomes based on multiply-imputed datasets, and the results http://bmjopen.bmj.com/ 37 38 compared to those obtained from the complete case analysis. This is in addition to the 39 40 planned sensitivity analyses. 41 42 43 Heterogeneity, bias and study quality

44 on September 26, 2021 by guest. Protected copyright. 45 46 This investigation employs a random effects approach to analyses. As such, in two-stage 47 48 analyses, statistical heterogeneity will be examined using the 휏2 (which provides an estimate 49 50 of between-study variance) and 퐼2 statistics (which provides the proportion of total variance 51 52 that is due to “true” heterogeneity in treatment effects, interpretation as stated in the 53 2 54 Cochrane Handbook [25]). Additionally, the 푝 value for the 휒 test for heterogeneity, along 55 with visual inspection of forest plots, will be assessed. In one-stage analyses, heterogeneity 56 57 can be assessed through the variance of the study level random treatment effect and 58 59 through examination of the coefficients for fixed study membership terms. 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 15 of 24 BMJ Open

1 2 3 If substantial heterogeneity is observed between results from different groups of studies, 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 data across heterogeneous groups of studies will not be pooled, and the demographics and 6 7 characteristics of the differing groups of studies will be compared in an attempt to identify 8 9 differences that could cause the heterogeneity. Potential evidence of substantial 10 11 heterogeneity in two stage analyses is defined in this investigation as some combination of 12 푝 휒2 퐼2 13 (1) significant value for test for heterogeneity in intervention tests at level of 0.10, (2) 14 statistic greater than 50% (representing substantial heterogeneity), (3) visual inspection of 15 16 the forest plot to identify heterogeneity (Cochrane Handbook, section 9.5.2 [26]). Potential 17 18 evidence of substantialFor heterogeneity peer in reviewone stage analyses only will be given by comparison of 19 20 models with and without the study level random treatment effect and fixed study 21 22 membership terms. If evidence exists of heterogeneity not accounted for through the 23 24 proposed model structure, use of additional terms (fixed effects or study level random 25 effects) will be examined and noted. 26 27 28 Risk of bias in each study will be assed using the Cochrane Collaboration Risk of Bias tool . 29 30 Analyses and results will be interpreted in light of the risk of bias of the studies included in 31 32 the meta-analysis. 33 34 35 Publication bias (and other selection bias/small study effects) will be investigated through

36 inspection of contour-enhanced funnel plots and appropriate statistical tests for funnel plot http://bmjopen.bmj.com/ 37 38 asymmetry [11-13]. Assessment of publication bias will only be undertaken for analyses 39 40 containing 10 or more trials (due to the low power of the assessments for analyses 41 42 containing small numbers of trials). 43

44 on September 26, 2021 by guest. Protected copyright. 45 Analyses of the data will clearly report the proportion of individuals within each study for 46 47 which IPD could be obtained, as well as the numbers of studies which were deemed eligible 48 to be included in the meta-analysis, but which did not supply any data. Interpretation of 49 50 overall strength of the evidence regarding modifiers of patient response to treatment 51 52 examined in this evidence synthesis will be appraised in light of relevant assessment of IPD 53 54 integrity, availability and risk of bias. 55 56 57 Subgroup and Sensitivity Analysis 58 59 60 Sensitivity analyses

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 16 of 24

1 2 3 For both primary and secondary outcomes, where conversion between scales has occurred, 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 sensitivity analyses will be conducted that (1) remove studies with transformed data, 6 7 analysing only data recorded on the main scale, and (2) use standardised scores across 8 9 studies [19] (unless standardised scores have been used in the main analysis). 10 11 12 There were a range of ways in which some treatment effect modifiers could be coded in this 13 14 investigation. As such, sensitivity analyses will also be conducted investigating whether the 15 method of coding particular modifiers, effected the results. Sensitivity analyses will also be 16 17 conducted to investigate unavailability of IPD (comparing results from two-stage MA that 18 For peer review only 19 combine results from study specific IPD analyses, with AD extracted from study reports, to 20 21 the main analysis based only on IPD), and to investigate changes in treatment effect over 22 23 time (primary and longitudinally measured secondary outcome one and two-stage analyses 24 25 will be re-conducted including an interaction term between treatment and time). 26 27 28 Subgroup Analyses 29 30 We will conduct subgroup analyses contrasting non-blind versus single-blind trials to 31 32 examine the effect of masking of outcome assessments. Any analyses where outcomes have 33 34 been transformed onto a common scale will also be sub-grouped as data originally recorded 35

36 on the common scale versus data transformed onto the common scale. In one stage http://bmjopen.bmj.com/ 37 38 analyses, sub-grouping will be achieved by interacting the coefficient of interest with the 39 40 grouping variable. In two stage analyses, results will be pooled from studies belonging to 41 each sub-group. 42 43

44 on September 26, 2021 by guest. Protected copyright. 45 Discussion 46 47 48 49 NICE and other clinical guidelines worldwide recommends CBT as an intervention for the 50 51 treatment and management of psychosis and schizophrenia [2], however AD-MA to date 52 have reported heterogeneous treatment effects between studies. This heterogeneity may 53 54 be attributable to certain as yet unidentified patient or intervention specific characteristics 55 56 that influence the clinical effectiveness of CBT. This IPD-MA examines a range of potential 57 58 treatment effect modifiers for CBT for patients with patients with schizophrenia-spectrum 59 60 diagnoses, for the primary outcome overall psychotic symptoms severity, as well as a range

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 17 of 24 BMJ Open

1 2 3 of secondary outcomes commonly targeted by CBT for psychosis. This investigation is 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 required to establish the effectiveness of CBT for psychosis across a range of different 6 7 populations, and will provide evidence to improve healthcare for patients with 8 9 schizophrenia-spectrum diagnoses. 10 11 12 Author Contributions 13 14 15 16 FV, CTS and APM conceived the idea for the review, and are project managing the 17 18 systematic review andFor meta-analysis. peer GD review provided input inonly the initial stages of the 19 development and design of the project. EL provided patient and public involvement of the 20 21 project. TK will be involved in future dissemination of research. CTS and MS drafted the 22 23 Statistical Analysis Plan for the project with help from RE, and (along with XL) input into data 24 25 collection, cleaning and analysis. All authors contributed to the development of the idea and 26 27 drafting and revision of the manuscript. All authors gave approval for the manuscript to be 28 29 submitted. 30 31 32 Funding 33 34 35 This project was funded by the NIHR Health Technology Assessment Programme (project

36 http://bmjopen.bmj.com/ 37 number 15/187) 38 39 40 41 Competing interests 42 43

44 We recognise that one member of our team may be regarded as having a vested interest in on September 26, 2021 by guest. Protected copyright. 45 46 CBT (Morrison is the only co-applicant actively involved in CBT training, leading of trial 47 48 grants and receiving royalties from CBT texts or books). The other members of the team do 49 50 not have conflicts of interest. 51 52 53 Patient consent 54 55 56 57 Not required as secondary analysis of data 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 18 of 24

1 2 3

4 Provenance and peer review BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 This project has been externally peer reviewed as part of the funding competition that 8 9 supported this work. A protocol consistent with the information in this report has been sent 10 11 to and approved by the HTA. 12 13 14 15 Author note 16 17 18 If there is need to amendFor this peer protocol, thereview date of each amendmentonly and the reason for 19 20 change will be noted. Each change will be scrutinised by the funder, with the date of 21 22 approval recorded, and version number updated accordingly. 23 We used the PRISMA-P checklist when writing our report [27] 24 25 26 27 References 28 29 30 31 32 1. Stewart, L.A., et al., Preferred reporting items for a 33 34 systematic review and meta-analysis of individual 35

36 participant data: The prisma-ipd statement. JAMA, 2015. http://bmjopen.bmj.com/ 37 38 39 313(16): p. 1657-1665. 40 41 2. National Institute for Health and Care Excellence, NICE 42 43 guidelines CG178 - Psychosis and schizophrenia in

44 on September 26, 2021 by guest. Protected copyright. 45 46 adults: Treatment and management. 2014, London: 47 48 National Institute for Health and Care Excellence. 49 50 3. Wykes, T., et al., Cognitive Behavior Therapy for 51 52 53 Schizophrenia: Effect Sizes, Clinical Models, and 54 55 Methodological Rigor. Schizophrenia Bulletin, 2008. 34(3): 56 57 p. 523-537. 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 19 of 24 BMJ Open

1 2 3 4 4. Jauhar, S., et al., Cognitive-behavioural therapy for the BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 symptoms of schizophrenia: systematic review and meta- 8 9 analysis with examination of potential bias. The British 10 11 Journal of Psychiatry, 2014. 204(1): p. 20-29. 12 13 14 5. Lambert, P.C., et al., A comparison of summary patient- 15 16 level covariates in meta-regression with individual patient 17 18 data meta-analysis.For peer Journal review of Clinical only Epidemiology, 19 20 2002. 55(1): p. 86-94. 21 22 23 6. Riley, R.D., P.C. Lambert, and G. Abo-Zaid, Meta-analysis 24 25 of individual participant data: rationale, conduct, and 26 27 reporting. BMJ, 2010. 340. 28 29 30 7. Riley, R.D., et al., Meta-analysis of continuous outcomes 31 32 combining individual patient data and aggregate data. 33 34 Statistics in Medicine, 2008. 27(11): p. 1870-1893. 35

36 http://bmjopen.bmj.com/ 37 8. Smith, C.T., et al., Individual participant data meta- 38 39 analyses compared with meta-analyses based on 40 41 aggregate data. Trials, 2011. 12(Suppl 1): p. A57-A57. 42 43 9. Brabban, A., S. Tai, and D. Turkington, Predictors of

44 on September 26, 2021 by guest. Protected copyright. 45 46 Outcome in Brief Cognitive Behavior Therapy for 47 48 Schizophrenia. Schizophrenia Bulletin, 2009. 35(5): p. 49 50 51 859-864. 52 53 10. Rathod, S., et al., Insight into schizophrenia: the effects of 54 55 cognitive behavioural therapy on the components of 56 57 58 insight and association with sociodemographics—data on 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 20 of 24

1 2 3 4 a previously published randomised controlled trial. BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 Schizophrenia Research, 2005. 74(2–3): p. 211-219. 8 9 11. Allott, K., et al., Patient predictors of symptom and 10 11 functional outcome following cognitive behaviour therapy 12 13 14 or befriending in first-episode psychosis. Schizophrenia 15 16 Research, 2011. 132(2–3): p. 125-130. 17 18 12. Lincoln, ForT.M., etpeer al., Who review stays, who only benefits? Predicting 19 20 dropout and change in cognitive behaviour therapy for 21 22 23 psychosis. Psychiatry Research, 2014. 216(2): p. 198- 24 25 205. 26 27 13. Hutton, P., et al., Cognitive behavioural therapy for 28 29 30 psychosis: A systematic review and meta-analysis 31 32 PROSPERO reference: CRD42013003911, 2013. 33 34 14. Hutton, P., et al., Cognitive behavioural therapy for 35

36 http://bmjopen.bmj.com/ 37 psychosis: rationale and protocol for a systematic review 38 39 and meta-analysis. Psychosis, 2014. 6(3): p. 220-230. 40 41 15. Leucht, S., et al., Dose equivalents for second-generation 42 43 antipsychotics: The minimum effective dose method.

44 on September 26, 2021 by guest. Protected copyright. 45 46 Schizophrenia Bulletin, 2014. 40(2): p. 314-326. 47 48 16. Hermes, E.D.A., et al., Minimum Clinically Important 49 50 51 Difference in the Positive and Negative Syndrome Scale 52 53 With Data From the Clinical Antipsychotic Trials of 54 55 Intervention Effectiveness (CATIE). JOURNAL OF 56 57 58 CLINICAL PSYCHIATRY, 2012. 73(4): p. 526-532. 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 21 of 24 BMJ Open

1 2 3 4 17. Leucht, S., et al., Linking the PANSS, BPRS, and CGI: BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 clinical implications. Neuropsychopharmacology: Official 8 9 Publication Of The American College Of 10 11 Neuropsychopharmacology, 2006. 31(10): p. 2318-2325. 12 13 14 18. Leucht, S., et al., Equipercentile linking of the BPRS and 15 16 the PANSS. European Neuropsychopharmacology, 2013. 17 18 23(8): p.For 956-959. peer review only 19 20 19. Higgins, J., Meta-analysis of continuous outcome data 21 22 23 from individual patients. STATISTICS IN MEDICINE, 24 25 2001. 20(15): p. 2219-2241. 26 27 20. R Core Team, R: A Language and Environment for 28 29 30 Statistical Computing. 2015, R Foundation for Statistical 31 32 Computing: Vienna, Austria. 33 34 21. DerSimonian, R. and N. Laird, Meta-analysis in clinical 35

36 http://bmjopen.bmj.com/ 37 trials. Controlled Clinical Trials, 1986. 7(3): p. 177-188. 38 39 22. Wulfsohn, M.S. and A.A. Tsiatis, A Joint Model for Survival 40 41 and Longitudinal Data Measured with Error. International 42 43 Biometric Society, 1997(1): p. 330.

44 on September 26, 2021 by guest. Protected copyright. 45 46 23. Sudell, M., et al., Investigation of 2-stage meta-analysis 47 48 methods for joint longitudinal and time-to-event data 49 50 51 through simulation and real data application. Statistics in 52 53 Medicine, 2017. 54 55 24. Hua, H., et al., One-stage individual participant data meta- 56 57 58 analysis models: estimation of treatment-covariate 59 60 interactions must avoid ecological bias by separating out

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 22 of 24

1 2 3 4 within-trial and across-trial information. Statistics in BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 Medicine, 2017. 36(5): p. 772-789. 8 9 25. Higgins, J., S. Green, and (editors). Cochrane Handbook 10 11 for Systematic Reviews of Interventions Version 5.1.0 12 13 14 (updated March 2011). The Cochrane Collaboration, 15 16 2011. Available from www.cochrane-handbook.org. 17 18 26. Deeks JJ,For Higgins peer JPT, review Altman DG only (editors). Chapter 9: 19 20 Analysing data and undertaking meta-analyses. In: 21 22 23 Higgins JPT, Green S (editors). Cochrane Handbook for 24 25 Systematic Reviews of Interventions Version 5.1.0 26 27 (updated March 2011). The Cochrane Collaboration, 28 29 30 2011. Available from www.cochrane-handbook.org. 31 32 33 27. Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, 34 35 Petticrew M, Shekelle P, Stewart LA. Preferred Reporting

36 http://bmjopen.bmj.com/ 37 38 Items for Systematic Review and Meta-Analysis Protocols 39 40 (PRISMA-P) 2015 statement. Syst Rev. 2015;4(1):1.. 41 42 43

44 on September 26, 2021 by guest. Protected copyright. 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 23 of 24 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 1 2 3 Reporting checklist for protocol of a systematic review. 4 5 6 Based on the PRISMA-P guidelines. 7 8 9 Instructions to authors 10 11 Complete this checklist by entering the page numbers from your manuscript where readers will find each of the 12 items listed below. 13 14 15 Your article may not currently address all the items on the checklist. Please modify your text to include the 16 missing information. If youFor are certain peer that an item review does not apply, pleaseonly write "n/a" and provide a short 17 18 explanation. 19 20 Upload your completed checklist as an extra file when you submit to a journal. 21 22 In your methods section, say that you used the PRISMA-Preporting guidelines, and cite them as: 23 24 25 Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, Petticrew M, Shekelle P, Stewart LA. Preferred 26 Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) 2015 statement. Syst Rev. 27 2015;4(1):1. 28 29 30 Page 31 Reporting Item Number

32 http://bmjopen.bmj.com/ 33 34 Title 35 36 Identification #1a Identify the report as a protocol of a systematic review 1 37 38 Update #1b If the protocol is for an update of a previous systematic review, n/a 39

40 identify as such on September 26, 2021 by guest. Protected copyright. 41 42 Registration 43 44 #2 If registered, provide the name of the registry (such as PROSPERO) 2 45 46 and registration number 47 48 Authors 49 50 51 Contact #3a Provide name, institutional affiliation, e-mail address of all protocol 1 52 authors; provide physical mailing address of corresponding author 53 54 Contribution #3b Describe contributions of protocol authors and identify the guarantor 15 55 56 of the review 57 58 Amendments 59 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 24 of 24 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 1 #4 If the protocol represents an amendment of a previously completed or n/a 2 3 published protocol, identify as such and list changes; otherwise, state 4 plan for documenting important protocol amendments 5 6 Support 7 8 9 Sources #5a Indicate sources of financial or other support for the review 16 10 11 Sponsor #5b Provide name for the review funder and / or sponsor 16 12 13 Role of sponsor or #5c Describe roles of funder(s), sponsor(s), and / or institution(s), if any, n/a 14 15 funder in developing the protocol 16 For peer review only 17 Introduction 18 19 20 Rationale #6 Describe the rationale for the review in the context of what is already 2-3 21 known 22 23 Objectives #7 Provide an explicit statement of the question(s) the review will 3-7 24 25 address with reference to participants, interventions, comparators, and 26 outcomes (PICO) 27 28 29 Methods 30 31 Eligibility criteria #8 Specify the study characteristics (such as PICO, study design, setting, 8-9

32 http://bmjopen.bmj.com/ 33 time frame) and report characteristics (such as years considered, 34 language, publication status) to be used as criteria for eligibility for 35 36 the review 37 38 Information sources #9 Describe all intended information sources (such as electronic 8 39 databases, contact with study authors, trial registers or other grey

40 on September 26, 2021 by guest. Protected copyright. 41 literature sources) with planned dates of coverage 42 43 Search strategy #10 Present draft of search strategy to be used for at least one electronic 8 44 45 database, including planned limits, such that it could be repeated 46 47 Study records - data #11a Describe the mechanism(s) that will be used to manage records and 10-11 48 49 management data throughout the review 50 51 Study records - #11b State the process that will be used for selecting studies (such as two 8-9 52 selection process independent reviewers) through each phase of the review (that is, 53 54 screening, eligibility and inclusion in meta-analysis) 55 56 Study records - data #11c Describe planned method of extracting data from reports (such as 10-11 57 58 collection process piloting forms, done independently, in duplicate), any processes for 59 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 25 of 24 BMJ Open

obtaining and confirming data from investigators BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 1 2 3 Data items #12 List and define all variables for which data will be sought (such as 4-7 4 PICO items, funding sources), any pre-planned data assumptions and 5 6 simplifications 7 8 Outcomes and #13 List and define all outcomes for which data will be sought, including 4-7 9 prioritization prioritization of main and additional outcomes, with rationale 10 11 12 Risk of bias in #14 Describe anticipated methods for assessing risk of bias of individual 13-14 13 individual studies studies, including whether this will be done at the outcome or study 14 15 level, or both; state how this information will be used in data synthesis 16 For peer review only 17 Data synthesis #15a Describe criteria under which study data will be quantitatively 11-15 18 19 synthesised 20 21 Data synthesis #15b If data are appropriate for quantitative synthesis, describe planned 11-15 22 summary measures, methods of handling data and methods of 23 24 combining data from studies, including any planned exploration of 25 consistency (such as I2, Kendall’s τ) 26 27 28 Data synthesis #15c Describe any proposed additional analyses (such as sensitivity or 14-15 29 subgroup analyses, meta-regression) 30 31 Data synthesis #15d If quantitative synthesis is not appropriate, describe the type of 11

32 http://bmjopen.bmj.com/ 33 summary planned 34 35 Meta-bias(es) #16 Specify any planned assessment of meta-bias(es) (such as publication 13-14 36 37 bias across studies, selective reporting within studies) 38 39 Confidence in #17 Describe how the strength of the body of evidence will be assessed 14

40 on September 26, 2021 by guest. Protected copyright. 41 cumulative (such as GRADE) 42 evidence 43 44 45 The PRISMA-P checklist is distributed under the terms of the Creative Commons Attribution License CC-BY 46 4.0. This checklist was completed on 23. September 2019 using https://www.goodreports.org/, a tool made by 47 the EQUATOR Network in collaboration with Penelope.ai 48 49 50 51 52 53 54 55 56 57 58 59 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from

Protocol for Individual Participant Data Meta Analysis of randomised control trials of patients with psychosis, to investigate treatment effect modifiers for CBT versus treatment as usual or other psychosocial interventions

ForJournal: peerBMJ Open review only Manuscript ID bmjopen-2019-035062.R1

Article Type: Protocol

Date Submitted by the 31-Jan-2020 Author:

Complete List of Authors: Sudell, Maria; University of Liverpool, Department of Biostatistics, Institute of Translational Medicine Tudur-Smith, Catrin; University of Liverpool, Biostatistics Liao, Xiaomeng; University of Liverpool, Department of Biostatistics, Institute of Translational Medicine Longden, Eleanor; Greater Manchester West Mental Health NHS Foundation Trust Dunn, Graham; University of Manchester, Health Methodology Research Kendall, Tim; National Collaborating Centre for Mental Health Emsley, Richard; King’s College London, Department of Biostatistics and

Health Informatics, Institute of Psychiatry, Psychology and Neuroscience http://bmjopen.bmj.com/ Morrison, Anthony; University of Manchester Varese, Filippo; Manchester Academic Health Science Centre,

Primary Subject Mental health Heading:

Secondary Subject Heading: Research methods

Keywords: IPD-MA, Treatment Modifier, CBT, Psychosis on September 26, 2021 by guest. Protected copyright.

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 1 of 26 BMJ Open

1 2 3

36 http://bmjopen.bmj.com/ 37 38 39 40 41 42 43

44 on September 26, 2021 by guest. Protected copyright. 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 2 of 26

1 2 3

4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 Protocol for Individual Participant Data Meta Analysis of 6 7 8 9 randomised control trials of patients with psychosis, to 10 11 12 investigate treatment effect modifiers for CBT versus treatment 13 14 15 16 as usual or other psychosocial interventions 17 18 For peer review only 19 Maria Sudell1,a*, Catrin Tudur Smith1,b, Xiaomeng Liao1,c, Eleanor Longden3,d, Graham Dunn4, 20 21 Tim Kendall5, 6,e, Richard Emsley7,f & Anthony P. Morrison2,3,g Filippo Varese2,3,h 22 23 1Institute of Translational Medicine, University of Liverpool; 2School of Health Sciences, 24 25 Division of Psychology and Mental Health, Faculty of Biology Medicine and Health, University 26 3 27 of Manchester; Greater Manchester Mental Health NHS Mental Health Foundation Trust; 28 4Institute of Population Health, University of Manchester; 5National Collaborating Centre for 29 30 Mental Health; 6Sheffield Health and Social Care NHS Foundation Trust; 7Department of 31 32 Biostatistics & Health Informatics, Kings College London; [email protected]; 33 34 [email protected]; [email protected]; [email protected]; 35 [email protected]; [email protected];

36 http://bmjopen.bmj.com/ 37 g h 38 [email protected] ; [email protected]; 39 *Corresponding author 40 41 42 University of Manchester 43 Division of Psychology and Mental Health

44 on September 26, 2021 by guest. Protected copyright. 45 School of Health Sciences 46 Zochonis Building, 2nd floor; room 2.40 47 Manchester 48 M13 9PL 49 50 51 52 53 Article Summary 54 55 56 57 Introduction: Aggregate data meta-analyses have shown heterogeneous treatment effects 58 59 for Cognitive Behavioural Therapy (CBT) for patients with schizophrenia-spectrum 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 3 of 26 BMJ Open

1 2 3 diagnoses. This heterogeneity could stem from specific intervention or patient 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 characteristics that could influence the clinical effectiveness of CBT, termed treatment 6 7 effect modifiers. This Individual Participant Data Meta-Analysis (IPD-MA) will investigate a 8 9 range of potential treatment effect modifiers of the efficacy CBT. Methods and analysis: We 10 11 will perform a systematic review and meta-analysis of studies investigating CBT vs treatment 12 13 as usual (TAU), or CBT vs other psychosocial interventions, for patients with schizophrenia- 14 spectrum diagnoses. The Cochrane Central Register of Controlled Trials (CENTRAL), PubMed, 15 16 EMBASE and the online clinical trials registers of the US government, European Union, 17 18 World Health OrganizationFor andpeer Current Controlledreview Trials Ltdonly will be searched. Two 19 20 researchers will screen titles and abstracts identified by the search. Individual participant 21 22 data will be requested for any eligible study, for the primary outcome (overall psychotic 23 24 symptoms), secondary outcomes and treatment effect modifiers. Data will be checked and 25 recoded according to an established Statistical Analysis Plan. One and two-stage random 26 27 effects meta-analyses investigating potential treatment effect modifiers will be conducted. 28 29 A list of potential treatment effect modifiers for CBT will be produced, motivating future 30 31 research into particular modifiers. Ethics and dissemination: This study does not require 32 33 ethical approval as it is based on data from existing studies, although best ethical practice 34 for secondary analysis of clinical data will be followed. The findings will be submitted for 35

36 publication in peer-reviewed journals, and promoted to relevant stakeholders. http://bmjopen.bmj.com/ 37 38 39 40 Prospero registration number CRD42017060068 41 42 43 Strengths and Limitations of study

44 on September 26, 2021 by guest. Protected copyright. 45 46  This will be the first published IPD-MA to investigate treatment effect modifiers for 47 CBT for patients with schizophrenia-spectrum diagnoses 48 49  The review will consider the efficacy of CBT across multiple outcomes of interest in 50 51 addition to psychotic symptoms severity 52 53  The search will be conducted without geographical, language or time restrictions 54 55  A potential limitation of this study will be the degree of heterogeneity between 56 57 studies in recorded measures and scales employed. 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 4 of 26

1 2 3 4 Introduction BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 8 Cognitive behavioural therapy (CBT) is a recommended intervention for the treatment and 9 10 management of psychosis and schizophrenia [1]. Aggregate Data Meta-Analyses (AD-MA) 11 12 suggest that CBT for psychosis has modest but considerably heterogeneous treatment 13 effects (for example [2, 3]). This inconsistency partly stems from inter-trial variation in 14 15 several key methodological characteristics of the existing Randomised Controlled Trials 16 17 (RCT; for example, blinding/masking of outcome assessments [2, 3]) but could also reflect 18 For peer review only 19 the impact of unaccounted clinical heterogeneity i.e. specific intervention and patient 20 21 characteristics that can potentially influence the clinical effectiveness of CBT. For instance, 22 previous trials differed widely in terms of intervention characteristics (e.g. number of 23 24 treatment sessions, treatment duration, use of manualised interventions), patients’ baseline 25 26 severity of psychotic and other comorbid symptoms, their demographic characteristics (e.g. 27 28 age, gender and ethnic origin), and illness duration. The identification of moderators of 29 30 treatment response and/or sub-groups of patients who may particularly benefit from CBT 31 32 would allow optimisation of treatment delivery, with significant implications in terms of 33 improved clinical effectiveness, cost savings and maximisation of patients’ informed choice 34 35 of treatment.

36 http://bmjopen.bmj.com/ 37 38 39 The impact of these potential treatment effect modifiers, however, remains unaccounted 40 41 for (or at best poorly estimated) in AD-MA due to their reliance of the reporting quality of 42 43 primary studies and the limited statistical power of “standard” meta-analytic methods for

44 testing treatment effect moderators [4]. Additional primary research would be costly and on September 26, 2021 by guest. Protected copyright. 45 46 impractical given the large sample size required. The only approach suited for this type of 47 48 research is Individual Patient Data Meta-Analysis (IPD-MA), a research synthesis method 49 50 which summarises the evidence on a particular clinical question by considering individual 51 52 participant level rather than aggregate level data from multiple related studies. IPD-MA 53 54 allow (1) greater ability to examine the impact of multiple individual-level and study-level 55 factors (and their combination) on the treatment effects considered, (2) standardisation of 56 57 statistical methods used across studies, and (3) the potential of reduced risk of bias e.g. due 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 5 of 26 BMJ Open

1 2 3 to selective reporting of outcomes compared to conventional AD-MA [5-8]. This article is 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 the protocol for an IPD-MA to assess the impact of treatment effect modifiers of CBT. 6 7 8 9 Methods 10 11 12 13 Objectives of IPD-MA 14 15 16 17 Primary outcome 18 For peer review only 19 20 The primary objective of the IPD-MA of RCTs is to identify treatment effect modifiers for CBT 21 or CBT+ vs Treatment As Usual (TAU) or other psychosocial interventions on overall 22 23 psychotic symptoms severity as measured by PANSS scores [9, 10] in patients with 24 25 schizophrenia-spectrum diagnoses. 26 27 28 Secondary outcomes 29 30 31 In this IPD-MA of RCTs, treatment effect modifiers for CBT or CBT+ vs TAU or other 32 33 psychosocial interventions will also be examined for patients with schizophrenia-spectrum 34 35 diagnoses for the secondary outcomes listed in Table 1Error! Reference source not found..

36 http://bmjopen.bmj.com/ 37 38 For some secondary outcomes, multiple sub-outcomes will be examined, such as Change in 39 severity of affective symptoms, which is examined both for anxiety and depression. Adverse 40 41 event information will be sought specifically for the 4 sub-outcomes mentioned, but 42 43 additional adverse event information will also be tabulated. Secondary outcomes were

44 on September 26, 2021 by guest. Protected copyright. 45 selected based on (1) outcomes often targeted in CBT for psychosis, (2) measures identified 46 47 as valuable in our PPI consultations. 48 49 50 Minimum clinical important differences (MCIDs) in PANSS scores [9-12] 51 Reduction of ≥ 11 points 52 53 Reduction of ≥ 15 points 54 55 Clinically Significant Deterioration (CSDs) in PANSS scores [9-12] 56 57 Increase of ≥ 11 points 58 59 Increase of ≥ 15 points 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 6 of 26

1 2 3 Change in specific symptom clusters 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 Positive psychopathology 6 7 Negative psychopathology 8 9 General psychopathology 10 11 Change in specific symptoms often targeted in CBT for psychosis 12 13 Hallucinations severity 14 15 Delusions severity 16 17 Hallucinations associated subjective distress 18 Delusions associatedFor peer subjective reviewdistress only 19 20 Paranoia severity 21 22 Change in severity of affective symptoms 23 24 Anxiety 25 26 Depression 27 28 Subjectively defined recovery 29 30 Quality of Life 31 Social and Occupational functioning 32 33 Early Treatment Discontinuation 34 35 Adverse effects

36 http://bmjopen.bmj.com/ 37 Deaths 38 39 Attempts at suicide 40 41 Suicide ideation 42 43 Serious violent incidents

44 on September 26, 2021 by guest. Protected copyright. 45 Hospital readmissions 46 Table 1: Secondary outcomes 47 48 49 Treatment comparisons 50 51 52 Consistent with the analytic approaches in recent AD-MA [13, 14] our IPD-MA will 53 54 distinguish between “pure” CBT interventions as defined by NICE [1] and “CBT+” 55 56 interventions, where CBT+ is defined as a CBT treatment packages incorporating significant 57 elements of other, distinct psychosocial intervention approaches (e.g. mindfulness, 58 59 motivational interviewing, family intervention) alongside core CBT elements. CBT and CBT+ 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 7 of 26 BMJ Open

1 2 3 will be compared to Treatment As Usual (TAU) or other psychosocial interventions in 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 individuals with a diagnosis of schizophrenia-spectrum disorders. TAU is defined as the level 6 7 of care service users would routinely receive had they not been involved in the trial, other 8 9 psychosocial interventions is defined as TAU supplemented by additional psychological or 10 11 social interventions e.g. family therapy. Although all trials of CBT against these comparators 12 13 will be eligible, these will be synthetized and contrasted in separate analyses for: (1) CBT vs 14 TAU, (2) CBT vs other psychosocial interventions, (3) CBT+ vs TAU, (4) CBT+ vs other 15 16 psychosocial interventions. 17 18 For peer review only 19 Treatment Effect Modifiers 20 21 22 The selection of treatment effect modifiers was informed by (1) knowledge of variables 23 24 examined in previous and on-going RCTs by members of our team and the collaborators in 25 26 the CBTp: IMPART Consortium; (2) the findings of studies which examined predictors of 27 28 outcomes in previous RCTs [e.g. 11, 15-17], and (3) consultation meetings with service-users 29 with psychosis and clinical psychologists and CBT therapists working with clients with 30 31 psychosis in secondary care settings in the UK. 32 33 34 The treatment effect modifiers shown in Table 2Error! Reference source not found. will be 35

36 investigated for both the primary and secondary outcomes. Treatment effect modifiers http://bmjopen.bmj.com/ 37 38 have been separated into three main groups: (1) Participant Demographic Characteristics, 39 40 (2) Participant’s Clinical (3) Specific Intervention Characteristics 41 42 43

44 on September 26, 2021 by guest. Protected copyright. 45 Treatment effect Modifier 46 47 48 Participant’s Demographic Characteristics Age at entry to trial 49 50 51 Gender 52 53 54 Ethnicity 55 56 57 Participant’s Clinical Characteristics Effect of specific diagnostic sub-groups 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 8 of 26

1 2 3 Phase of illness (first episode psychosis / 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 multiple episode psychosis) 6 7 8 Illness Duration 9 10 11 Duration of untreated psychosis 12 13 14 Initial Severity of Psychotic Symptoms 15 16 (measured by baseline PANSS scores) 17 18 For peer review only 19 Initial severity of comorbid affective 20 symptoms (measured by baseline Anxiety 21 22 scores) 23 24 25 Initial severity of comorbid affective 26 27 symptoms (measured by baseline 28 29 Depression scores) 30 31 32 Dosage equivalence of baseline anti- 33 psychotic medication(s) [12] 34 35

36 Number of anti-psychotic medications http://bmjopen.bmj.com/ 37 38 received at baseline) 39 40 41 Specific Intervention Characteristics time period over which treatment was 42 43 delivered*

44 on September 26, 2021 by guest. Protected copyright. 45 46 number of therapy sessions offered in the 47 48 study* 49 50 number of therapy sessions attended by 51 52 the individual 53 54 55 Minimum study required level of therapist’s 56 57 training and competence*; 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 9 of 26 BMJ Open

1 2 3 Measures of therapeutic alliance; 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 Use of manualised interventions*; 7 8 9 Use of formulation-based interventions*; 10 11 12 Indicator for whether the intervention was 13 14 designed to target the outcome under 15 16 scrutiny*; 17 18 For peer review only 19 Individual versus group interventions* 20 21 Length of study follow up 22 23 24 Table 2: Treatment effect modifiers examined in this IPD-MA. Note: treatment effect modifiers marked with * are study 25 level variables, the remaining are individual level variables 26 27 28 29 30 31 Protocol 32 33 34 35 This evidence synthesis will follow state-of-the-art guidelines for IPD-MA, and our outputs

36 http://bmjopen.bmj.com/ 37 will comply as a minimum with the PRISMA statement for the reporting of IPD-MA [8]. The 38 search strategy of our IPD-MA builds on the protocol and database searches carried out as 39 40 part of a recent AD-MA carried out by members of our team [13, 14]. Our study selection 41 42 criteria are consistent with those employed in this recent AD-MA. Similarly, our literature 43

44 searches will update those carried out as part of this review to identify any RCTs that have on September 26, 2021 by guest. Protected copyright. 45 46 become available since the date of search. 47 48 49 Search Strategy 50 51 52 53 We will update the searches already conducted as part of a recent AD-MA [13, 14] to 54 identify any trials that might have been become available for research synthesis since the 55 56 date of last search. Database searches will be conducted on the Cochrane Central Register 57 58 of Controlled Trials (CENTRAL), PubMed, EMBASE and the online clinical trials registers of 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 10 of 26

1 2 3 the US government, European Union, World Health Organization and Current Controlled 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 Trials Ltd. 6 7 8 In line with the protocol of the AD-MA carried out by members of our team [13, 14], titles, 9 10 abstracts and keywords will be searched in the publication databases using the adaptations 11 12 of following generic strategy: (schizo$ [exp. schizophrenia + psychosis + schizoaffective]) 13 14 AND (trial [exp. RCT +controlled trial + clinical trial]) AND (cbt [exp. cognitive therapy + 15 behaviour therapy +psychotherapy]). 16 17 18 For peer review only 19 20 21 22 Selection of Studies 23 24 25 The project’s PI and another member of the research team will screen titles and abstracts 26 27 for relevance, and subsequently assess eligibility by examining the full-text reports against 28 29 the above-mentioned criteria. When required, additional information to ascertain eligibility 30 31 will be requested from the RCTs’ authors and through the examination of treatment manual 32 when available (e.g. to ascertain that interventions complied with NICE operational 33 34 definitions of CBT). Discrepancies in selection decisions will be discussed, and arbitration by 35

36 other members of the research team sought to achieve consensus. We will include studies http://bmjopen.bmj.com/ 37 38 based on the following eligibility criteria: 39 40 41 Participants 42 43 Trials where > 50% of participants have diagnoses in the schizophrenia-spectrum

44 on September 26, 2021 by guest. Protected copyright. 45 46 (schizophrenia, schizoaffective disorder or early psychosis) will be eligible. Trials where > 47 50% participants have an established diagnosis of bipolar disorder, intellectual disability, 48 49 psychosis secondary to a general medical condition or organic pathology, or a primary 50 51 diagnosis of substance-induced psychosis will be excluded. No restriction will be placed on 52 53 participants’ age, ethnicity, illness severity and illness duration. 54 55 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 11 of 26 BMJ Open

1 2 3

4 Treatment Comparisons BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 Included studies must compare treatments that fit into one or more of the four stated 7 8 treatment categories (CBT, CBT+, TAU, other psychosocial interventions). Study 9 10 interventions will be classified into one of these four categories by clinicians involved with 11 12 this project. 13 14 15 Outcomes 16 17 18 The study must provideFor data peer for one or morereview of the stated only primary or secondary outcomes. 19 20 Data can be recorded on any comparable scale, at any time-point. Data will also be sought 21 22 for the treatment effect modifiers listed in Table 2Error! Reference source not found.. 23 24 25 Study Designs 26 27 Parallel single-blind or open controlled trials with at least two arms utilizing random 28 29 allocation to treatment will be considered for inclusion. Single arm or cross-over studies will 30 31 not be eligible for inclusion in the review. Studies employing other research designs (case 32 33 series, cohort analyses) will not be eligible. 34 35 Trials will be eligible if they evaluated CBT or CBT+ interventions versus TAU or other

36 http://bmjopen.bmj.com/ 37 38 psychosocial treatments, eligible treatments as defined in Treatment Comparisons Section. 39 40 41 Data Collection and Processes 42 43

44 IPD will be collected from principal investigators of past and on-going RCTs of CBT for on September 26, 2021 by guest. Protected copyright. 45 46 psychosis in the UK and internationally. Our ability to successfully collect relevant IPD is 47 48 facilitated by several factors. Our research team includes researchers who have conducted 49 50 some of the largest RCTs in this area. Furthermore, we have established a network of 51 52 collaborators to support the retrieval of relevant IPD: the CBTp IMPART Consortium. We will 53 54 continue to expand the CBTp IMPART Consortium over the lifetime of the project by sending 55 invitation emails to all researchers who have published and/or are currently conducting 56 57 randomised controlled trials relevant to this work. Participating researchers will be sent 58 59 specific data request forms outlining variables pertinent to the present IPD-MA. They will be 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 12 of 26

1 2 3 asked to fully anonymise the requested dataset and share them with our research team 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 using a safe data transfer system provided by the IT services University of Liverpool. We will 6 7 remain in regular contact with all Consortium members throughout the lifetime of the 8 9 project to clarify queries about their IPD and its integrity. 10 11 12 In the case of no response to data requests (defined as a minimum of 4 contact attempts 13 14 with no response), details of the study would be stated as “non-acquired data”. Details 15 concerning the number and proportion of studies and individuals for which IPD has been 16 17 obtained will be stated in reports of analyses. Additionally, we have planned sensitivity 18 For peer review only 19 analyses to assess the impact of non-acquired data on results of the project, which will 20 21 combine obtained IPD, with AD from studies not supplying IPD. 22 23 24 A statistical analysis plan detailing the data cleaning and coding, and the analyses to be 25 26 conducted has been produced and will be available upon request. Data received will be 27 systematically recoded to ensure common scales or measurements across studies. We will 28 29 liaise with principal investigators and statisticians of the primary studies to resolve any data 30 31 issues and prepare the dataset for IPD-MA. In addition to primary and secondary outcomes 32 33 data, IPD and relevant supporting material (e.g. trial codebook, therapy manuals, statistical 34 35 analysis plans) will be requested to code the stated treatment groups, treatment effect

36 modifiers, primary and secondary outcomes. http://bmjopen.bmj.com/ 37 38 39 The primary outcome will be analysed on the PANSS scale, with comparable information 40 41 recorded on other scales converted onto the PANSS scale where possible (e.g. Leucht et al 42 43 [18] provide supplemental tables to convert between PANSS and BPRS). Secondary

44 on September 26, 2021 by guest. Protected copyright. 45 outcomes will be transformed onto the most commonly reported scale, where established 46 47 conversion tables exist. If transformation to a common scale is impossible, or there is no 48 most common scale, standardised values (calculated by dividing by the between patient 49 50 variation) will be employed [19]. Transformation onto sub-scales will not be attempted. For 51 52 example, secondary outcome Change in specific symptom clusters: Positive is defined as the 53 54 sum of the positive sub-scale of the PANSS score. If a study records a comparable measure, 55 56 no attempt will be made to transform this measure onto the positive sub-scale of PANSS, 57 58 however the data would contribute to a standardized score analysis e.g. as part of a 59 sensitivity analysis. 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 13 of 26 BMJ Open

1 2 3

4 Statistical Analysis BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 8 Data Analysis 9 10 All randomised patients will be included, and an intention to treat (ITT) principle will be 11 12 followed throughout. To examine IPD integrity and concordance with original trial analyses, 13 14 all trials will be re-analysed individually and the original authors asked to confirm the 15 16 individual study results and resolve any discrepancies. Throughout, a frequentist approach 17 18 to analyses will be taken.For All peer analyses will review be conducted using only R [20]. If the planned 19 20 quantitative analyses cannot be undertaken, the data will be described qualitatively.We will 21 examine the pooled treatment effect for each outcome by performing a series of one-step 22 23 IPD meta-analyses (where IPD from individual studies are analysed simultaneously whilst 24 25 accounting for study-level clustering), and two-step IPD meta-analyses (where estimates of 26 27 the treatment effect are initially computed from the IPD of each study, and then aggregated 28 29 using conventional inverse variance meta-analytic approach) [5]. Throughout, due to 30 31 anticipated heterogeneity between studies, a random effects approach will be employed 32 (using a Dersimonian and Laird approach for two-stage analyses [21], and including a study 33 34 level random treatment effect, and fixed study membership effect in one-stage approaches 35

36 http://bmjopen.bmj.com/ 37 Both one stage and two stage analyses will be conducted to allow a full investigation of the 38 39 data [22]. The two stage analyses will be conducted initially, to help identify areas of higher 40 41 between study heterogeneity through examination of e.g. forest plots of analyses. One 42 43 stage analyses will then be conducted to allow multiple interactions between treatment

44 modifiers to be examined. One and two stage analysis results will be compared, to confirm on September 26, 2021 by guest. Protected copyright. 45 46 that areas of heterogeneity are identified similarly between each approach. 47 48 49 Due to considerable variation in the follow-up periods considered in the original trials, 50 51 separate analysis of trials with highly comparable or identical points of assessment (e.g. 3 52 53 month, 6 month etc.) would be unfeasible. In order to maximise the data contributing to the 54 55 analysis, outcomes measured at multiple time points will be modelled longitudinally. As 56 dropout may be an issue during study periods, we will employ joint modelling methods (e.g. 57 58 [23, 24]) to account for study dropout. Outcomes measured at a single time point (e.g. at 59 60 the end of the treatment period) will be analysed using generalised linear models.

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 14 of 26

1 2 3 In two-stage analyses, treatment-modifier interactions will be estimated within each study, 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 and the results pooled. In one-stage analyses, treatment-moderators interactions 6 7 separating out within and between study effects will be examined, while accounting for 8 9 clustering of participants within studies [5, 6, 25]. Any treatment effect modifier found to be 10 11 significant at a level of 0.05 will be retained in a list of potential treatment effect modifiers. 12 13 Treatment effect modifiers with a significant effect for each outcome will be identified 14 through examination of 95% Confidence Intervals (CI). Confidence intervals for both joint 15 16 and GLM analyses will be calculated through bootstrapping, using 200 bootstrap samples. 17 18 Once this list has beenFor compiled, peer both forward review and backward only manual selection procedures 19 20 for model parameters will be conducted for the one and two stage analyses of the primary 21 22 and secondary outcomes, provided sufficient data is available. The overlap between the 23 24 parameters selected by the forward and backward selection procedures will highlight, from 25 the list of potential treatment effect modifiers, parameters most likely to be true treatment 26 27 effect modifiers. 28 29 30 This investigation involves a large number of planned analyses. However, this analysis of 31 32 treatment modifiers of CBT is, to an extent, exploratory. As such, this investigation aims to 33 34 identify potential relationships, and in so doing motivate future investigations specifically 35 targeting the identified areas of interest. Consequently, and given that there is not a

36 http://bmjopen.bmj.com/ 37 standard multiple testing approach currently recommended for IPD-MA, we will not adjust 38 39 for multiple testing, although we reiterated that these analyses are, to an extent, 40 41 exploratory. If methods are developed during the course of this project that are 42 43 recommended as standard to account for multiple testing in IPD-MA, application of the

44 on September 26, 2021 by guest. Protected copyright. 45 methods will be examined. 46 47 The main analyses will be conducted as complete case analyses, i.e. only those contributing 48 49 data for all variables (outcome or explanatory) included in the model will be used in the 50 51 meta-analyses. If the level of missing data for treatment effect modifiers or outcomes is 52 53 large across the studies included in the meta-analysis, if possible, the effect of missing data 54 55 on the conclusions of the analyses will be investigated by re-conducting the analyses of the 56 57 primary and secondary outcomes based on multiply-imputed datasets, and the results 58 compared to those obtained from the complete case analysis. This is in addition to the 59 60 planned sensitivity analyses.

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 15 of 26 BMJ Open

1 2 3

4 Heterogeneity, bias and study quality BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 This investigation employs a random effects approach to analyses. As such, in two-stage 7 8 analyses, statistical heterogeneity will be examined using the 휏2 (which provides an estimate 9 10 of between-study variance) and 퐼2 statistics (which provides the proportion of total variance 11 12 that is due to “true” heterogeneity in treatment effects, interpretation as stated in the 13 2 14 Cochrane Handbook [26]). Additionally, the 푝 value for the 휒 test for heterogeneity, along 15 16 with visual inspection of forest plots, will be assessed. In one-stage analyses, heterogeneity 17 can be assessed through the variance of the study level random treatment effect and 18 For peer review only 19 through examination of the coefficients for fixed study membership terms. 20 21 22 If substantial heterogeneity is observed between results from different groups of studies, 23 24 data across heterogeneous groups of studies will not be pooled, and the demographics and 25 26 characteristics of the differing groups of studies will be compared in an attempt to identify 27 28 differences that could cause the heterogeneity. Potential evidence of substantial 29 heterogeneity in two stage analyses is defined in this investigation as some combination of 30 31 (1) significant 푝 value for 휒2 test for heterogeneity in intervention tests at level of 0.10, (2) 퐼2 32 33 statistic greater than 50% (representing substantial heterogeneity), (3) visual inspection of 34 35 the forest plot to identify heterogeneity (Cochrane Handbook, section 9.5.2 [27]). Potential

36 http://bmjopen.bmj.com/ 37 evidence of substantial heterogeneity in one stage analyses will be given by comparison of 38 39 models with and without the study level random treatment effect and fixed study 40 membership terms. If evidence exists of heterogeneity not accounted for through the 41 42 proposed model structure, use of additional terms (fixed effects or study level random 43

44 effects) will be examined and noted. on September 26, 2021 by guest. Protected copyright. 45 46 47 Risk of bias in each study will be assed using the Cochrane Collaboration Risk of Bias tool . 48 49 Analyses and results will be interpreted in light of the risk of bias of the studies included in 50 51 the meta-analysis. 52 53 Publication bias (and other selection bias/small study effects) will be investigated through 54 55 inspection of contour-enhanced funnel plots and appropriate statistical tests for funnel plot 56 57 asymmetry [11,13,17]. Assessment of publication bias will only be undertaken for analyses 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 16 of 26

1 2 3 containing 10 or more trials (due to the low power of the assessments for analyses 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 containing small numbers of trials). 6 7 8 Analyses of the data will clearly report the proportion of individuals within each study for 9 10 which IPD could be obtained, as well as the numbers of studies which were deemed eligible 11 12 to be included in the meta-analysis, but which did not supply any data. Interpretation of 13 14 overall strength of the evidence regarding modifiers of patient response to treatment 15 examined in this evidence synthesis will be appraised in light of relevant assessment of IPD 16 17 integrity, availability and risk of bias. 18 For peer review only 19 20 Subgroup and Sensitivity Analysis 21 22 23 24 Sensitivity analyses 25 26 For both primary and secondary outcomes, where conversion between scales has occurred, 27 28 sensitivity analyses will be conducted that (1) remove studies with transformed data, 29 30 analysing only data recorded on the main scale, and (2) use standardised scores across 31 32 studies [19] (unless standardised scores have been used in the main analysis). 33 34 35 There were a range of ways in which some treatment effect modifiers could be coded in this

36 investigation. As such, sensitivity analyses will also be conducted investigating whether the http://bmjopen.bmj.com/ 37 38 method of coding particular modifiers, effected the results. Sensitivity analyses will also be 39 40 conducted to investigate unavailability of IPD (comparing results from two-stage MA that 41 42 combine results from study specific IPD analyses, with AD extracted from study reports, to 43 the main analysis based only on IPD), and to investigate changes in treatment effect over

44 on September 26, 2021 by guest. Protected copyright. 45 46 time (primary and longitudinally measured secondary outcome one and two-stage analyses 47 will be re-conducted including an interaction term between treatment and time). 48 49 50 Subgroup Analyses 51 52 53 We will conduct subgroup analyses contrasting non-blind versus single-blind trials to 54 examine the effect of masking of outcome assessments. Any analyses where outcomes have 55 56 been transformed onto a common scale will also be sub-grouped as data originally recorded 57 58 on the common scale versus data transformed onto the common scale. In one stage 59 60 analyses, sub-grouping will be achieved by interacting the coefficient of interest with the

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 17 of 26 BMJ Open

1 2 3 grouping variable. In two stage analyses, results will be pooled from studies belonging to 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 each sub-group. 6 7 8 9 Patient and Public Involvement 10 11 Throughout this project, we have taken care to involve key stakeholders in the design of the 12 13 research. The research team contains a PPI representative. Additionally, secondary 14 15 outcomes and treatment effect modifiers were identified in part through consultation 16 17 meetings with service-users with psychosis and clinical psychologists and CBT therapists 18 For peer review only 19 working with clients with psychosis in secondary care settings in the UK. 20 21 Results of this research will be disseminated to stakeholders via a range of methods 22 23 including the preparation of an information pack for service users and carers, which will be 24 25 made freely available online. 26 27 28 29 Ethics and Dissemination 30 31 32 This IPD-MA will be conducted in line with current recommendations of secondary analysis 33 34 of IPD data (e.g. [5]). Specific ethics approval for the IPD-MA is not required, as the 35

36 objectives of the IPD-MA are consistent with those of the original trials, and do not violate http://bmjopen.bmj.com/ 37 38 the condition of consent under which the data was collected. Throughout, anonymised data 39 40 will be sought from study authors. Information from data owners should not include any 41 42 personally identifiable information. Data and accompanying documentation will be held on 43 a secure server by the research team. Findings of this research will be submitted for

44 on September 26, 2021 by guest. Protected copyright. 45 publication in high impact peer-reviewed journals, promoted to relevant stakeholders and 46 47 presented to relevant research communities at international conferences. 48 49 50 51 Discussion 52 53 54 NICE and other clinical guidelines worldwide recommends CBT as an intervention for the 55 56 treatment and management of psychosis and schizophrenia [1], however AD-MA to date 57 58 have reported heterogeneous treatment effects between studies. This heterogeneity may 59 60 be attributable to certain as yet unidentified patient or intervention specific characteristics

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 18 of 26

1 2 3 that influence the clinical effectiveness of CBT. This IPD-MA examines a range of potential 4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 treatment effect modifiers for CBT for patients with patients with schizophrenia-spectrum 6 7 diagnoses, for the primary outcome overall psychotic symptoms severity, as well as a range 8 9 of secondary outcomes commonly targeted by CBT for psychosis. The treatment effect 10 11 modifiers and outcomes investigated in this project are wide ranging, however given the 12 13 lack of current research in this area, it is hoped that this exploratory IPD-MA will provide 14 guidance for future focussed research. This investigation is required to establish the 15 16 effectiveness of CBT for psychosis across a range of different populations, and will provide 17 18 evidence to improveFor healthcare peer for patients review with schizophrenia-spectrum only diagnoses. 19 20 21 Author Contributions 22 23 24 25 FV, CTS and APM conceived the idea for the review, and are project managing the 26 27 systematic review and meta-analysis. GD provided input in the initial stages of the 28 29 development and design of the project. EL provided patient and public involvement of the 30 project. TK will be involved in future dissemination of research. CTS and MS drafted the 31 32 Statistical Analysis Plan for the project with help from RE, and (along with XL) input into data 33 34 collection, cleaning and analysis. All authors contributed to the development of the idea and 35 drafting and revision of the manuscript. All authors gave approval for the manuscript to be 36 http://bmjopen.bmj.com/ 37 38 submitted. 39 40 41 Funding 42 43

44 on September 26, 2021 by guest. Protected copyright. 45 This project is funded by the National Institute for Health Research (NIHR) Health 46 Technology Assessment Programme (project number 15/187/05). The views expressed are 47 48 those of the author(s) and not necessarily those of the NIHR or the Department of Health 49 50 and Social Care. 51 52 53 54 Acknowledgements 55 56 57 We thank all those involved in the PPI consultations that helped to shape this research 58 59 project. 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 19 of 26 BMJ Open

1 2 3

4 Competing interests BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 We recognise that one member of our team may be regarded as having a vested interest in 8 9 CBT (Morrison is the only co-applicant actively involved in CBT training, leading of trial 10 11 grants and receiving royalties from CBT texts or books). The other members of the team do 12 13 not have conflicts of interest. 14 15 16 Patient consent 17 18 For peer review only 19 20 Not required as secondary analysis of data 21 22 23 Provenance and peer review 24 25 26 27 This project has been externally peer reviewed as part of the funding competition that 28 supported this work. A protocol consistent with the information in this report has been sent 29 30 to and approved by the HTA. 31 32 33 34 Author note 35

36 http://bmjopen.bmj.com/ 37 If there is need to amend this protocol, the date of each amendment and the reason for 38 39 change will be noted. Each change will be scrutinised by the funder, with the date of 40 41 approval recorded, and version number updated accordingly. 42 43 We used the PRISMA-P checklist when writing our report

44 on September 26, 2021 by guest. Protected copyright. 45 46 References 47 48 49 50 51 1. National Institute for Health and Care Excellence, NICE 52 53 guidelines CG178 - Psychosis and schizophrenia in 54 55 56 adults: Treatment and management. 2014, London: 57 58 National Institute for Health and Care Excellence. 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 20 of 26

1 2 3 4 2. Wykes, T., et al., Cognitive Behavior Therapy for BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 Schizophrenia: Effect Sizes, Clinical Models, and 8 9 Methodological Rigor. Schizophrenia Bulletin, 2008. 34(3): 10 11 p. 523-537. 12 13 14 3. Jauhar, S., et al., Cognitive-behavioural therapy for the 15 16 symptoms of schizophrenia: systematic review and meta- 17 18 analysisFor with examination peer review of potential only bias. The British 19 20 Journal of Psychiatry, 2014. 204(1): p. 20-29. 21 22 23 4. Lambert, P.C., et al., A comparison of summary patient- 24 25 level covariates in meta-regression with individual patient 26 27 data meta-analysis. Journal of Clinical Epidemiology, 28 29 30 2002. 55(1): p. 86-94. 31 32 5. Riley, R.D., P.C. Lambert, and G. Abo-Zaid, Meta-analysis 33 34 of individual participant data: rationale, conduct, and 35

36 http://bmjopen.bmj.com/ 37 reporting. BMJ, 2010. 340. 38 39 6. Riley, R.D., et al., Meta-analysis of continuous outcomes 40 41 combining individual patient data and aggregate data. 42 43 Statistics in Medicine, 2008. 27(11): p. 1870-1893.

44 on September 26, 2021 by guest. Protected copyright. 45 46 7. Smith, C.T., et al., Individual participant data meta- 47 48 analyses compared with meta-analyses based on 49 50 51 aggregate data. Trials, 2011. 12(Suppl 1): p. A57-A57. 52 53 8. Stewart, L.A., et al., Preferred reporting items for a 54 55 systematic review and meta-analysis of individual 56 57 58 participant data: The prisma-ipd statement. JAMA, 2015. 59 60 313(16): p. 1657-1665.

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 21 of 26 BMJ Open

1 2 3 4 9. Hermes, E.D.A., et al., Minimum Clinically Important BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 Difference in the Positive and Negative Syndrome Scale 8 9 With Data From the Clinical Antipsychotic Trials of 10 11 Intervention Effectiveness (CATIE). JOURNAL OF 12 13 14 CLINICAL PSYCHIATRY, 2012. 73(4): p. 526-532. 15 16 10. Leucht, S., et al., Linking the PANSS, BPRS, and CGI: 17 18 clinical implications.For peer Neuropsychopharmacology: review only Official 19 20 Publication Of The American College Of 21 22 23 Neuropsychopharmacology, 2006. 31(10): p. 2318-2325. 24 25 11. Lincoln, T.M., et al., Who stays, who benefits? Predicting 26 27 dropout and change in cognitive behaviour therapy for 28 29 30 psychosis. Psychiatry Research, 2014. 216(2): p. 198- 31 32 205. 33 34 12. Leucht, S., et al., Dose equivalents for second-generation 35

36 http://bmjopen.bmj.com/ 37 antipsychotics: The minimum effective dose method. 38 39 Schizophrenia Bulletin, 2014. 40(2): p. 314-326. 40 41 13. Hutton, P., et al., Cognitive behavioural therapy for 42 43 psychosis: A systematic review and meta-analysis

44 on September 26, 2021 by guest. Protected copyright. 45 46 PROSPERO reference: CRD42013003911, 2013. 47 48 14. Hutton, P., et al., Cognitive behavioural therapy for 49 50 51 psychosis: rationale and protocol for a systematic review 52 53 and meta-analysis. Psychosis, 2014. 6(3): p. 220-230. 54 55 15. Brabban, A., S. Tai, and D. Turkington, Predictors of 56 57 58 Outcome in Brief Cognitive Behavior Therapy for 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 22 of 26

1 2 3 4 Schizophrenia. Schizophrenia Bulletin, 2009. 35(5): p. BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 859-864. 8 9 16. Rathod, S., et al., Insight into schizophrenia: the effects of 10 11 cognitive behavioural therapy on the components of 12 13 14 insight and association with sociodemographics—data on 15 16 a previously published randomised controlled trial. 17 18 SchizophreniaFor peer Research, review 2005. 74 only(2–3): p. 211-219. 19 20 17. Allott, K., et al., Patient predictors of symptom and 21 22 23 functional outcome following cognitive behaviour therapy 24 25 or befriending in first-episode psychosis. Schizophrenia 26 27 Research, 2011. 132(2–3): p. 125-130. 28 29 30 18. Leucht, S., et al., Equipercentile linking of the BPRS and 31 32 the PANSS. European Neuropsychopharmacology, 2013. 33 34 23(8): p. 956-959. 35

36 http://bmjopen.bmj.com/ 37 19. Higgins, J., Meta-analysis of continuous outcome data 38 39 from individual patients. STATISTICS IN MEDICINE, 40 41 2001. 20(15): p. 2219-2241. 42 43 20. R Core Team, R: A Language and Environment for

44 on September 26, 2021 by guest. Protected copyright. 45 46 Statistical Computing. 2015, R Foundation for Statistical 47 48 Computing: Vienna, Austria. 49 50 51 21. DerSimonian, R. and N. Laird, Meta-analysis in clinical 52 53 trials. Controlled Clinical Trials, 1986. 7(3): p. 177-188. 54 55 22. Burke, D.L., J. Ensor, and R.D. Riley, Meta-analysis using 56 57 58 individual participant data: one-stage and two-stage 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 23 of 26 BMJ Open

1 2 3 4 approaches, and why they may differ. Statistics in BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 Medicine, 2017. 36(5): p. 855-875. 8 9 23. Wulfsohn, M.S. and A.A. Tsiatis, A Joint Model for Survival 10 11 and Longitudinal Data Measured with Error. International 12 13 14 Biometric Society, 1997(1): p. 330. 15 16 24. Sudell, M., et al., Investigation of 2-stage meta-analysis 17 18 methodsFor for joint peer longitudinal review and time-to-eventonly data 19 20 through simulation and real data application. Statistics in 21 22 23 Medicine, 2017. 24 25 25. Hua, H., et al., One-stage individual participant data meta- 26 27 analysis models: estimation of treatment-covariate 28 29 30 interactions must avoid ecological bias by separating out 31 32 within-trial and across-trial information. Statistics in 33 34 Medicine, 2017. 36(5): p. 772-789. 35

36 http://bmjopen.bmj.com/ 37 26. Higgins, J., S. Green, and (editors). Cochrane Handbook 38 39 for Systematic Reviews of Interventions Version 5.1.0 40 41 (updated March 2011). The Cochrane Collaboration, 42 43 2011. Available from www.cochrane-handbook.org.

44 on September 26, 2021 by guest. Protected copyright. 45 46 27. Deeks JJ, Higgins JPT, Altman DG (editors). Chapter 9: 47 48 Analysing data and undertaking meta-analyses. In: 49 50 51 Higgins JPT, Green S (editors). Cochrane Handbook for 52 53 Systematic Reviews of Interventions Version 5.1.0 54 55 (updated March 2011). The Cochrane Collaboration, 56 57 58 2011. Available from www.cochrane-handbook.org. 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 24 of 26

1 2 3

4 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 5 6 7 8 9 10 11 12 13 14 15 16 17 18 For peer review only 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35

36 http://bmjopen.bmj.com/ 37 38 39 40 41 42 43

44 on September 26, 2021 by guest. Protected copyright. 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60

For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml Page 25 of 26 BMJ Open BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 1 2 3 Reporting checklist for protocol of a systematic review. 4 5 6 Based on the PRISMA-P guidelines. 7 8 9 Instructions to authors 10 11 Complete this checklist by entering the page numbers from your manuscript where readers will find each of the 12 items listed below. 13 14 15 Your article may not currently address all the items on the checklist. Please modify your text to include the 16 missing information. If youFor are certain peer that an item review does not apply, pleaseonly write "n/a" and provide a short 17 18 explanation. 19 20 Upload your completed checklist as an extra file when you submit to a journal. 21 22 In your methods section, say that you used the PRISMA-Preporting guidelines, and cite them as: 23 24 25 Moher D, Shamseer L, Clarke M, Ghersi D, Liberati A, Petticrew M, Shekelle P, Stewart LA. Preferred 26 Reporting Items for Systematic Review and Meta-Analysis Protocols (PRISMA-P) 2015 statement. Syst Rev. 27 2015;4(1):1. 28 29 30 Page 31 Reporting Item Number

40 identify as such on September 26, 2021 by guest. Protected copyright. 41 42 Registration 43 44 #2 If registered, provide the name of the registry (such as PROSPERO) 2 45 46 and registration number 47 48 Authors 49 50 51 Contact #3a Provide name, institutional affiliation, e-mail address of all protocol 1 52 authors; provide physical mailing address of corresponding author 53 54 Contribution #3b Describe contributions of protocol authors and identify the guarantor 15 55 56 of the review 57 58 Amendments 59 60 For peer review only - http://bmjopen.bmj.com/site/about/guidelines.xhtml BMJ Open Page 26 of 26 BMJ Open: first published as 10.1136/bmjopen-2019-035062 on 28 May 2021. Downloaded from 1 #4 If the protocol represents an amendment of a previously completed or n/a 2 3 published protocol, identify as such and list changes; otherwise, state 4 plan for documenting important protocol amendments 5 6 Support 7 8 9 Sources #5a Indicate sources of financial or other support for the review 16 10 11 Sponsor #5b Provide name for the review funder and / or sponsor 16 12 13 Role of sponsor or #5c Describe roles of funder(s), sponsor(s), and / or institution(s), if any, n/a 14 15 funder in developing the protocol 16 For peer review only 17 Introduction 18 19 20 Rationale #6 Describe the rationale for the review in the context of what is already 2-3 21 known 22 23 Objectives #7 Provide an explicit statement of the question(s) the review will 3-7 24 25 address with reference to participants, interventions, comparators, and 26 outcomes (PICO) 27 28 29 Methods 30 31 Eligibility criteria #8 Specify the study characteristics (such as PICO, study design, setting, 8-9