Menzies 2008

Methods	Design: Randomized, two‐arm, parallel group, open‐label, active‐controlled, trial Period of study: 27 April 2004 to 31 January 2007 (the trial was stopped early at the recommendation of the data safety monitoring board after the third planned interim analysis)
Participants	Number randomized: 847 Age: ≥ 18 years Gender: both genders Inclusion criteria Documented TST that met the criteria for a positive test by Canadian standards Recommended treatment for LTBI by the treating physician Exclusion criteria Contacts of INH or rifampicin‐resistant cases Patients allergic to INH or rifampicin or those taking drugs with clinically significant interaction
Interventions	Intervention: Rifampicin, 10 mg/kg (maximum 600 mg/day) daily for 4 months (N = 420) Control: INH, 5 mg/kg (maximum 300 mg/day) daily for 9 months (N = 427)
Outcomes	1. Grade 3 or 4 adverse events resulting in treatment discontinuation 2. On‐time treatment completion defined as taking more than 80% of doses within 150 days for rifampicin and 301 days for INH taken; monitored by medication event monitoring system 3. Serious adverse events 4. Grade 3 or 4 hepatotoxicity defined as ALT more than 5 times the ULN without symptoms or more than 3 times with symptoms, and more than 10 times ULN, respectively 5. Treatment‐limiting adverse events Outcomes reported in supplementary publication but not used in quantitative synthesis in this review: 7. Health care system costs
Notes	Setting: International multicentric trial involving nine university‐affiliated hospitals Country: Brazil (1), Canada (7), Saudi Arabia (1) Duration of follow‐up: 4 to 9 months Funding: Canadian Institutes of Health Research Comment: 804 of 847 randomized subjects had TST ≥ 10 mm HIV‐ positive patients were included in this trial. However, their numbers were small, and they were equal among the two arms: 7 (2%) of 427 in the INH arm and 6 (1%) of 420 participants in the rifampicin arm had HIV infection
Risk of bias
Bias	Authors' judgement	Support for judgement
Random sequence generation (selection bias)	Low risk	"A Web‐based programme verified eligibility and randomly assigned participants (by using a random‐number generator), after they signed informed consent..(to interventions)....in blocks of varying size, stratified by centre. A team at the University of Sherbrooke, Sherbrooke, Quebec, Canada, prepared the web‐based program and allocation sequence."
Allocation concealment (selection bias)	Low risk	The report (see above) indicates that centralised randomization was used and it is likely that the pill containers with electronic monitoring (see below) were serially numbered and linked to the randomization sequence.
Blinding (performance bias and detection bias) Efficacy outcomes	Low risk	"Doses taken were measured with the Medical Event Monitoring System, an electronic device in the pill container cap that recorded the date and time of bottle opening." Comment: This may not be an entirely accurate method of measuring adherence since it does not guarantee drug intake; however, it will not introduce bias in relative estimates of adherence in the two arms.
Blinding (performance bias and detection bias) Adverse events:	High risk	"The treating physician decided whether to discontinue, re‐challenge with, or restart the study therapy, although the protocol specified that participants with grade 3 or 4 adverse events were not to be re‐challenged. When all investigations were complete, and if therapy was permanently discontinued in response to the event, the patient’s clinical course and results of investigations and re‐challenge (if any) were made available to a 3‐member independent review panel who were blinded to study drug." Comment: This particular outcome was independently adjudicated by a three member review panel blinded to treatment allocation; only those patients that permanently discontinued study drug were reviewed. However, treatment discontinuation was at the discretion of the treating physician who was not blinded. "Between 16% and 24% of patients were missing laboratory assessments before treatment or during the first 2 months of treatment." Comment: This could potentially underestimate asymptomatic hepatotoxicity.
Incomplete outcome data (attrition bias) All outcomes	Low risk	"In total, 1 patient taking rifampicin and 6 patients taking INH dropped out, and no information was available regarding their health status when they stopped therapy. In a worst case scenario, if all had developed grade 3 or 4 adverse events, the magnitude of the observed difference in these events would have increased, favouring rifampicin." Comment: In addition to the above, all randomized patients were accounted for in analysis of completion.
Selective reporting (reporting bias)	Low risk	Although the trial protocol or trial registration documents were not available, all pre‐stated outcomes were reported.
Other bias	Low risk	The trial was stopped early for harms but this was done after the third interim analysis by the data safety monitoring board; the trial was not stopped early for apparent benefit.