Collins 2005.

Methods	Random allocation in blocks of 12, computer generated. No blinding. Groups similar on demographic and clinical characteristics.
Participants	106 heroin dependent by DSM‐IV. 36% iv users. Group sizes (1) 35 (2) 37 (3) 34. 72% male. Mean age 36. Major psychiatric illness, active medical illness, dependence on other drugs or alcohol exclusion criteria. 36% currently married or cohabiting. 56% currently employed.
Interventions	(1) Nalmefene 4mg iv over 30 minutes, naltrexone 50mg via nasogastric tube, under propofol anaesthesia (4‐6 hours). Various adjunct medications, including octreotide. (2) Buprenorphine, 8mg (single dose) day 1, naltrexone 12.5mg afternoon of day 2, 25mg 12 hours later, then 50mg/day. (3) Clonidine, max 1.2mg/day, discharged day 3, naltrexone 12.5mg day 7, 25mg day 8, then 50mg/day.. Clonidine and various adjunct medications available to all groups. All received 72 hours inpatient care, followed by 12 weeks outpatient naltrexone maintenance with relapse prevention psychotherapy as aftercare.
Outcomes	Graphs of withdrawal severity. Mean weeks in treatment. Number of participants completing inpatient phase, receiving at least one dose of naltrexone, receiving full 50mg dose of naltrexone. Number retained in treatment over 12 weeks. Number retained 12 weeks who provided 2 or less opiate positive urine samples. Number experiencing serious adverse events.
Notes	Withdrawal assessed by Subjective & Objective Opiate Withdrawal Scales, and Clinical Institute Narcotic Assessment, four times a day during inpatient phase. Country: USA
Risk of bias
Bias	Authors' judgement	Support for judgement
Adequate sequence generation?	Low risk	Quote: "...using random, computer‐generated assignments with stratification by sex. ... In addition, the Berger‐Exner test was used to confirm that no selection bias in enrollment occurred."
Allocation concealment?	Low risk	Quote: "All staff remained unaware of the randomization sequence..."
Blinding? Subjective outcomes ‐ intensity of withdrawal, adverse effects	High risk	Patients were not blinded to treatment ‐ sham anaesthesia not ethical. Unclear whether observers were blinded to treatment.
Blinding? Objective outcomes ‐ duration of treatment, completion of treatment	Low risk	These outcomes unlikely to be affected by knowledge of treatment.
Incomplete outcome data addressed? All outcomes	Low risk	Losses to follow‐up similar in three groups.
Free of selective reporting?	Low risk	None apparent
Free of other bias?	Unclear risk	Enrollment stopped at 106 participants (aim 150) "because actual differences in withdrawal severity scores and treatment retention were smaller than anticipated, leading to an impractically large recalculated sample size..." Comment: It seems unlikely that the stopping of enrolment resulted in bias.
Other bias: Selection of comparison cohort	Low risk	Experimental and control groups drawn from same population.
Other bias: Comparability of cohorts	Low risk	No significant differences in demographics of three groups.
Other bias: Representativeness of exposed cohort	Low risk	Participants drawn from population seeking heroin detoxification.
Other bias: Ascertainment of exposure	Low risk	Data collection established by study protocol.