Diazepam 1979, IL.

Study characteristics
Methods	Allocation: randomised. Blinding: double blind. Design: not stated. Duration: 24 hours. Setting: Eitanim Psychiatric Hospital, Israel.
Participants	Diagnosis: paranoid schizophrenia (n = 9), schizoaffective schizophrenia (n = 9), schizophrenia subtypes (n = 5), paranoid states (n = 2), mania (n = 5), no diagnosis (n = 10). n = 40. Age: 20‐49 years (mean 33 years). Gender: 27 men, 13 women. Ethnicity: not stated. Consent: not stated. History: previous hospitalisation (mean 3 instances). Inclusion criteria: newly admitted psychotic people aged 20‐50 years after 2 days of drug‐free observation. Exclusion criteria: physical illness.
Interventions	Benzodiazepines vs antipsychotics. 1. Diazepam 30‐40 mg IV (n = 20). 2. Haloperidol 20‐35 mg IV (n = 20). In the haloperidol group, 10 participants received high‐dose haloperidol and 10 participants received moderately high‐dose haloperidol; these scores were analysed together. Doses were given over a 3‐hour period. All participants were free from neuroleptics for 48 hours before the study began.
Outcomes	Global impression: sedation; mean score (CGI). Mental state: mean score (BPRS).
Notes	Note: assumptions were made regarding missing SDs data ‐ SDs for the CGI‐S and BPRS mean scores were estimated from the range.
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Randomised "patients were randomly assigned" to treatment ‐ no further description provided.
Allocation concealment (selection bias)	Unclear risk	Not stated.
Blinding (performance bias and detection bias) All outcomes	Unclear risk	Observer blind ‐ it is unclear how trial authors remained blinded to alternation and decrease of haloperidol dosage. Rating scales: BPRS and CGI were performed by consensus of 2 physicians (trial authors EL and YL), who were "unaware of the treatment administered."
Incomplete outcome data (attrition bias) All outcomes	High risk	Follow‐up: 50%. Reasons for dropout were reported; 16 due to treatment and 4 due to 'raters' being unavailable.
Selective reporting (reporting bias)	Low risk	All measure outcomes were reported.
Other bias	Low risk	Funding: supported by a grant from the Gralnick Foundation, High Point Hospital, Port Chester, NY. We did not detect any other bias.