Levin 2004.

Methods	Double‐blind, randomised, placebo‐controlled trial. 2‐week single‐blind placebo lead‐in phase prior to random allocation. Study included a cross‐over phase which was not included in this review due to substantial dropout (> 30%) in the first 2 weeks.
Participants	Setting: outpatient with 2 clinic visits per week; New York, USA. Scheduled duration 8 weeks (plus subsequent cross‐over phase that was excluded from this review) Participants: 27 enrolled, 25 randomised; cannabis dependent by DSM‐IV, using daily Group sizes: group 1, 13; group 2, 12 Groups similar on demographics and clinical characteristics at baseline, except for somewhat higher (not statistically significant) use of cannabis by group 1 Mean age 32 years 92% male 56% Caucasian, 20% Hispanic, 24% African American Mean (± SD) joints smoked per week at baseline: group 1, 28.3 (SD 23.2); group 2, 19.4 (SD 16.4) Exclusion criteria: dependence on other substances, except caffeine and nicotine, and psychiatric disorder requiring medical intervention
Interventions	Group 1: oral divalproex sodium commenced at 500 mg/day, increasing to maximum of 2 g/day, depending on response Group 2: placebo Medication in 2 doses per day Weekly individual cognitive‐behavioural relapse prevention therapy as adjunct
Outcomes	Outcomes reported for 19 participants who completed 8 weeks of study: frequency and amount of cannabis use and craving score at baseline and weeks 7 and 8; number completing scheduled treatment; number with ≥ 2 weeks of assumed abstinence Urine samples collected and analysed at each visit Participants reported cannabis use and completed a visual analogue scale of intensity and desire for cannabis each week Clinician‐rated global impression assessment for cannabis use completed weekly "Strict abstinence" defined as ≥ 1 negative urine sample and no self‐reported cannabis use for that week. "Assumed abstinence" if patient reported no cannabis use and urine samples had THC‐COOH levels ≥ 50% below the previous week.
Notes	Funding: Research grants (NIDA) Declaration of conflict of interest not published
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Unclear risk	Quote: "Twenty‐seven participants were enrolled and 25 were randomized." Comment: method of sequence generation not reported
Allocation concealment (selection bias)	Unclear risk	Quote: [participants] "… were randomly assigned to receive either divalproex or a matching placebo." Comment: method of allocation concealment not reported
Blinding of participants and personnel (performance bias) All outcomes	Low risk	Quote: "Following randomization, patients received … either divalproex sodium or a placebo using a double‐blind design."
Blinding of outcome assessment (detection bias) All outcomes	Low risk	Study conducted double‐blind. Use of urine screening to support determination of "abstinence" would be expected to help reduce bias in subjective outcomes.
Incomplete outcome data (attrition bias) All outcomes	High risk	Rates of dropout were similar in the 2 groups, but there was no discussion of possible differences between those retained and those who dropped out of the study. Cannabis use outcomes were reported only for those who completed treatment.
Selective reporting (reporting bias)	Low risk	None apparent
Other bias	Low risk	The cross‐over phase of the trial was excluded from analyses and this review due to high rates of dropout in the first 2 weeks