Tohen 2008.

Methods	Randomised double‐blind trial
Participants	Number of total participants/type of patients: 521 participants randomised. Both inpatients and outpatients Inclusion criteria Sex: male and female Age: 18 ‐ 65 years Diagnosis: a diagnosis of DSM‐IV‐TR acute bipolar manic or mixed episode without psychotic features. In addition, participants were required to score ≥ 20 on the YMRS and ≤ 30 (mild to moderate) and a Clinical Global Impression for Bipolar Disorder‐Severity of Illness scale (CGI‐BP‐S) mania subscore of 3 or 4 at screening (week 1) and at randomisation (week 0). Women had to test negative for pregnancy and be using effective contraception. Exclusion criteria Rapid cycling course or Psychotic features in DSM‐IV‐TR
Interventions	Location: USA, Lithuania, Pueto Rico, Romania, and Russia Date of study: October 2004 ‐ December 2006 Treatment groups: 1. Olanzapine (5 ‐ 20 mg) administered orally once in the evening. Mean dose 11.4 g (SD 2.49 g) 2. Valproate (500 ‐ 2500 mg) administered orally 3 times a day if ≥ 750 mg, twice a day if not Mean dose 848.4 mg (SD 135.62 mg)/Plasma levels 61.3 mg/L (SD 32.04 mg/L) 3. Placebo Placebo capsules used to balance all medications out to 3 doses a day. Concomitant medication: Lorazepam ≤ 2 mg a day was allowed as long as > 8 hours before psychiatric assessment. Anticholinergics and ongoing thyroids supplementation therapy were permitted Length of study: 21 days Randomisation: 215 randomised to olanzapine 201 randomised to valproate 105 randomised to placebo
Outcomes	Primary outcome: Change in scores on the YMRS from baseline to endpoint Secondary outcome: CGI‐BP, MADRS Safety was assessed using adverse events measures (using Abnormal Involuntary Movement Scale, Barnes Akaesthesia Rating Scale, Simpson Angus Rating Scale) and by monitoring other laboratory test values (e.g. ECG results, weight changes) Study withdrawals were reported
Funding	"This study was supported by Eli Lilly and Co., Indianapolis, Ind."
Conflict of interest	Not stated
Notes
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	Quote: "A computer generated random sequence randomly assigned patients to treatment group within each study site."
Allocation concealment (selection bias)	Unclear risk	Quote: "A computer generated random sequence randomly assigned patients to treatment group within each study site." Comment: We contacted the lead author who confirmed that the random allocation was determined centrally (not on site). However, the method of concealment is still unclear and there is a chance that involved participants and staff would be able to anticipate allocation.
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Quote: "Double‐blind" Quote: "To keep investigators blind to treatment‐assignment, all study drugs were dispensed by an interactive voice response/web/fax tool [...] dose adjustments were conducted via the interactive voice response/web/fax tool. To maintain blinding, every time the interactive voice response/web/fax tool send a message to alter the dose of a valproate treated patient, a dummy message was send to alter the dose of an olanzapine or placebo treated patient." Quote: "To maintain blinding, all patients had blood collected for assessing valproate concentration, irrespective of whether they received valproate." Quote: "All study medication appeared identical."
Blinding of outcome assessment (detection bias) Efficacy	Low risk	Quote: "To keep investigators blind to treatment‐assignment, all study drugs were dispensed by an interactive voice response/web/fax tool [...] dose adjustments were conducted via the interactive voice response/web/fax tool." Comment: Although blinding procedures appear robust, there is no explicit mention of who conducted outcome assessments and if they were blind. We contacted the author who confirmed that the rating was conducted by site raters who were blind.
Incomplete outcome data (attrition bias) All outcomes	Low risk	Comment: In all groups dropout rate was below 30%.
Selective reporting (reporting bias)	Low risk	Comment: Protocol found. Primary outcome as stated on protocol.
Other bias	Low risk	Comment: None identified