Birchwood 2014.

Methods	Allocation: randomised Blinding: assessor blind Location: three UK centres, UK Length of follow‐up: 18 months
Participants	Diagnosis: schizophrenia, schizoaffective or mood disorders (schizophrenia (n = 98); schizoaffective disorder (n = 29); paranoid schizophrenia (n = 17); psychosis (n = 50); bipolar disorder (n = 3) (ICD‐10)) N = 197 Sex: 113 M, 84 F Age: > 16 years, mean ˜ 37.4 years, SD ˜ 12.1 years Included: length of illness: not stated, had a history of harmful command hallucinations for at least 6 months with recent (< 9 months) history of harm to self or others, or major social transgressions as a result of the commands (full or incomplete compliance); or had harmful command hallucinations whereby the individual was distressed and appeasing the powerful voice Excluded: organic impairment or addictive disorder considered to be the primary diagnosis and insufficient command of the English language
Interventions	1. CBT group*: N = 98 Content: Cognitive behavioural therapy for command hallucinations (CTCH): behavioural therapy techniques were used to assess and modify conviction in four beliefs linked to the construct of voice power. Protocol for cognitive therapy for command hallucinations were developed by the author and details were provided in our casebook manuals. Delivered by: cognitive therapists who were supervised in each centre by a lead clinician with expertise in cognitive behaviour therapy for psychosis Frequency: a maximum of 9 months (about 25 sessions of therapy) Treatment duration: 6 months 2. Standard care* group: N = 99 Content: treatment‐as‐usual was provided by community mental health and assertive outreach and early intervention teams. Treatment‐as‐usual included antipsychotic medication. Delivered by: not reported Frequency: not reported Treatment duration: 6 months
Outcomes	Mental state: general, positive symptoms, negative symptoms, hallucinations, delusions, affective symptoms (PANSS scores); general, distress, (PSYRATS scores); hopelessness (BHS scores) Adverse events: death Satisfaction with treatment: leaving the study early Unable to use: Mental state: depression (CDS scores) ‐ skewed data Behavioural responses to voices (not predefined in protocol)
Notes	*The term "Treatment‐as‐usual (TAU)" was used in this paper. Participants in CBT group also received the standard care intervention. This trial shared the same intervention protocol as Trower 2004 but reported data from different participants.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: "...using an allocation sequence generated with OpenCDMS.25 and were stratified by the centre with permuted blocks with a randomly varying block size after stratification by centre." (p.24) Comments:The investigators described a random component in the sequence generation process.
Allocation concealment (selection bias)	Low risk	Quote: "...using an allocation sequence generated with OpenCDMS.25 and were stratified by the centre." (p.24) Comments:The outcome assessor could not foresee assignment, however the participants and therapists were informed of the allocation assignment.
Blinding of participants and personnel (performance bias) All outcomes	High risk	Quote: "After randomisation, an email notification about group allocation was sent to the trial manager, trial administrator, and therapists. An email notification confirming that the participant had been randomly assigned to treatment (with no information about group allocation) was sent to the centre research assistant. The trial administrator then sent a letter to the participant and the care coordinator informing them about the outcome of the randomisation." (p.25) Comments: The participants and therapists were informed of the allocation assignment
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Quote: "Randomization was masked from the assessors." "When masking was broken, another rater, masked to group assignment, assessed and rated the participant for all subsequent assessments; accordingly, all final ratings were masked..." (p.25) Comments: The outcome assessor could not foresee assignment.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Quote: "Analysis was by intention to treat." (p.26) Comments: All randomised participants were analysed.
Selective reporting (reporting bias)	High risk	Comments: We looked through the protocol of this trial and found that the author did not report the childhood trauma, the quality of life, and costs of the interventions.
Other bias	Low risk	Comments: This trial was registered, number ISRCTN62304114. The funder did not play a role in data collection, analysis, or interpretation. Other bias was not obvious.