In this review of 57 trials with rifampin-containing regimens, use of rifampin only initially rather than throughout treatment was associated with worse treatment outcomes (higher rates of failure, relapse, and acquired drug resistance). Thrice-weekly intermittent dosing schedules during the initial treatment phase were associated with increased adjusted risk of acquired drug resistance, but not relapse or failure. Initial drug resistance was strongly associated with increased risk of poor treatment outcomes, particularly if rifampin was used only in the initial intensive phase. These findings have important implications for TB treatment.
The most important finding of this review is that all three treatment outcomes were significantly worse with regimens that used rifampin for the first 1–2 mo rather than throughout therapy. This finding adds considerable weight to similar findings by Jindani and colleagues, who compared regimens containing 2 mo versus 6 mo of rifampin 
. This review includes many more studies with a variety of regimens, making these results more robust and generalizable. In this review, the failure and relapse rates progressively declined with long duration of rifampin; such a dose–response relationship strengthens the conclusions that a longer duration of rifampin treatment is responsible for better outcomes. Finally, this review included studies where drug sensitivity testing was performed, which permitted us to detect an increased risk of acquired drug resistance with shorter rifampin duration and also permitted stratified analysis by underlying drug resistance, which proved to be a very important determinant of treatment outcomes. According to the most recent information from WHO 
, the 8-mo regimen was the recommended initial therapy in 24 high-incidence countries. Based on the pooled risk differences from within trial analyses, we estimate that treatment of 100 patients with the regimen of 2HRZE/6HE (the “8-mo” regimen) would result in 13 more failures and relapses than if they received 2HRZE/4HR (the “6-mo” regimen). As a result, forthcoming recommendations by WHO will recommend only the 6-mo (rifampin throughout) regimen, and the 8-mo regimen will no longer be recommended 
. Results of this review suggest that the public health benefits of switching from the 8-mo to the 6-mo regimen should be very considerable.
The lower risk of relapse with regimens using rifampin for at least 8 mo is consistent with subgroup analyses of other trials 
and a recent cohort report from Hong Kong 
. These have shown that patients with extensive cavitary pulmonary disease have increased risk of relapse with 6-mo regimens. Taken together these findings support recommendations to extend therapy for patients at high risk of relapse 
. However, accurate identification of high-risk patients is imprecise, and provision of extended therapy may be challenging in high-burden settings.
The lack of effect of intermittency is interesting but has several caveats. The timing of intermittent dosing may be quite important, as suggested by the finding of increased risk of acquired resistance associated with thrice-weekly therapy throughout. A cohort study from New York City reported that patients with HIV–TB coinfection had an increased risk of acquired rifamycin resistance if they were treated with twice-weekly therapy during the initial intensive phase but not if they were treated with intermittent dosing only during the continuation phase 
Te other important finding is the previously underestimated impact of primary isoniazid resistance on failure, relapse, and acquired resistance. This important effect is a powerful argument for widespread availability of rapid, inexpensive testing for resistance to isoniazid (as well as for rifampin), or for regimens that do not require optimal activity from isoniazid. The influence of primary streptomycin resistance is likely to be less important, since streptomycin has been replaced by ethambutol in most settings.
The primary objective of this review was to compare the efficacy of different durations and dosing schedule of rifampin. To accomplish this, we have analyzed the per-protocol results from each trial, using standardized microbiological definitions. All studies reviewed reported adverse events, dropouts, and defaulters separately, facilitating our approach. However, we did not include these outcomes because they are not as well defined nor standardized, potentially creating greater between-study variability. As well, inter-study differences in providers and populations could have very important influences on these outcomes—even greater than any biologic differences in disease response. These would be balanced within each trial but could have created substantial bias with our analytic approach. If dropout, default, or side effects were associated with the same characteristics as failure or relapse, then excluding these outcomes could underestimate the poor outcomes associated with shorter rifampin exposures. But if not, then including these outcomes would simply reduce all differences between regimens.
This review had several limitations. First, we could identify few trials with direct head-to-head comparisons of rifampin duration, and even fewer directly comparing intermittent regimens. Hence, we had to pool results across studies; this increases potential confounding from differences in treatment, patients' disease severity, or other differences in the study populations, since the studies were conducted in many different countries. The advantage of this approach is that we are able to include many more trials, thereby increasing the precision and avoiding selection bias 
. However, the disadvantage is the greater potential for bias due to between-trial differences in participant characteristics, treatment regimens, as well as the differential impact of dropouts and other losses to follow-up 
. Concern about this latter problem should be alleviated by the consistent results from three analytic approaches—within the smaller set of trials with head-to-head comparisons, across all 57 trials, and the multivariate analysis. Also, the dispersion estimates from multivariable analysis suggest that treatment factors and underlying drug resistance accounted for almost all the differences in outcomes observed.
Most trials were initiated before 1980, limiting the number of participants with HIV infection and drug resistance. The lack of trials in HIV infected persons with active TB meant that the question of rifampin duration in treatment of HIV-TB could not be answered, due to insufficient power. This underscores the paucity of recent TB treatment trials and the urgent need for trials in drug resistant and/or HIV infected populations. There were no trials in children, reflecting a lack of rigorous trials in this population and the difficulties of microbiologic confirmation in this population. Death was not analyzed, because most TB-related deaths occur soon after diagnosis and are determined by comorbidity, age, severity of illness, and delay in diagnosis 
. Deaths later in treatment are often from other causes 
. Hence, differences in the TB treatment regimen may have relatively little impact on mortality. We endeavored to minimize language bias by including studies published in French and Spanish as well as English. Interestingly, this yielded only three additional trials, or 5% of all trials included. In a recent review 
, of all TB related papers listed in PubMed over 10 y, papers published in English, French, and Spanish represented 84% of all published literature worldwide. Hence, this review can be considered reasonably representative of publications in this field. However, in some fields, such as mental health, PubMed will fail to list a substantial proportion of relevant publications from low- to middle-income countries 
, so we may have missed some important trials.
Finally, we were not able to distinguish between relapse of the same strain of M. tuberculosis
that caused the initial infection and reinfection with a new strain of the bacillus. In settings with high rates of ongoing exposure to M. tuberculosis
, particularly if HIV seroprevalence is also high, a relatively high proportion of cases of recurrent TB following initial apparent cure may be due to reinfection 
. However, very few participants had HIV coinfection in the studies reviewed, and in studies with longer follow-up, the great majority of relapses occurred in the first 1–2 y, with very few occurring in the third to fifth years. This suggests that reinfection should have accounted for very few of the disease recurrences. Because follow-up was adequate in almost all studies—only four studies had less than 1-y follow-up—unequal follow-up should not have affected results—supported by the finding that duration of post-treatment follow-up was not associated with relapse rates in multivariable analysis.
This review provides evidence against continued use of regimens that utilize rifampin for the first 2 mo only, as they are significantly and substantially inferior to regimens that use rifampin for at least 6 mo. This review also has identified an important need for adequately powered clinical trials that address dosing schedules, management of isoniazid mono-resistance, and the optimal duration of treatment to prevent relapse.