|Home | About | Journals | Submit | Contact Us | Français|
It is often necessary to include WHO group 5 drugs in the treatment of extensively drug-resistant tuberculosis (XDR-TB) and fluoroquinolone-resistant multidrug-resistant tuberculosis (MDR-TB). As clinical evidence about the use of group 5 drugs is scarce, we conducted a systematic review using published individual patient data. We searched PubMed and OvidSP through 7 April 2013 for publications in English to assemble a cohort with fluoroquinolone-resistant MDR-TB treated with group 5 drugs. Favorable outcome was defined as sputum culture conversion, cure, or treatment completion in the absence of death, default, treatment failure, or relapse. A cohort of 194 patients was assembled from 20 articles involving 12 geographical regions. In descending order of frequency, linezolid was used in treatment of 162 (84%) patients, macrolides in 84 (43%), clofazimine in 65 (34%), amoxicillin with clavulanate in 56 (29%), thioridazine in 18 (9%), carbapenem in 16 (8%), and high-dose isoniazid in 16 (8%). Cohort analysis with robust Poisson regression models and random-effects meta-analysis similarly suggested that linezolid use significantly increased the probability (95% confidence interval) of favorable outcome by 57% (10% to 124%) and 55% (10% to 121%), respectively. Defining significant associations by risk ratios ≥ 1.2 or ≤ 0.9, neither cohort analysis nor meta-analysis demonstrated any significant add-on benefit from the use of other group 5 drugs with respect to outcome for patients treated with linezolid, although selection bias might have led to underestimation of their effects. Our findings substantiated the use of linezolid in the treatment of XDR-TB or fluoroquinolone-resistant MDR-TB and call for further studies to evaluate the roles of other group 5 drugs.
Multidrug-resistant tuberculosis (MDR-TB), defined as TB with bacillary resistance to at least isoniazid and rifampin, has become a global epidemic. Concern about drug-resistant TB intensified with the emergence of extensively drug-resistant TB (XDR-TB), which is MDR-TB with additional resistance to any fluoroquinolone and at least one of the second-line injectable drugs (SLID). MDR-TB (notably cases with bacillary resistance to fluoroquinolones) and XDR-TB are more difficult to treat than drug-susceptible TB, with substantially worse outcome alongside mounting drug resistance (1–8). It is often necessary to include World Health Organization (WHO) group 5 drugs in the treatment of XDR-TB and fluoroquinolone-resistant MDR-TB. The WHO group 5 drug classification refers to anti-TB drugs with unclear efficacy or an unclear role in MDR-TB treatment (9). These include thiacetazone, linezolid, high-dose isoniazid, clofazimine, amoxicillin with clavulanate, macrolides, carbapenem, and thioridazine. Linezolid and clofazimine have been evaluated by three systematic reviews (10–12), but the researchers provided only pooled estimates of treatment outcome without any controlled comparison. One systematic review with random-effects multivariable logistic meta-regression has evaluated the role of second-line anti-TB drugs among general MDR-TB patients using individual patient data (13). Despite a large sample size of 9,153 patients, that review was not sufficiently powered to evaluate the role of linezolid and high-dose isoniazid. As clinical evidence about WHO group 5 drugs and outcome in the treatment of the most complicated form of XDR-TB is scarce, we conducted a systematic review using published individual patient data.
We searched the published literature through 7 April 2013 for publications in English to assemble a cohort with XDR-TB or fluoroquinolone-resistant MDR-TB treated with WHO group 5 drugs. PubMed and OvidSP were used to search for biomedical articles from MEDLINE, life science journals, and EMBASE using the following key phrases with keywords in titles or abstracts with the help of Boolean operators (“and” plus “or”): (i) tuberculosis or TB and (ii) linezolid or clofazimine or clavulanate or amoxicillin or meropenem or imipenem or thioridazine or phenothiazine or clarithromycin or high-dose isoniazid or thiacetazone and (iii) drug-resistant or multidrug-resistant or extensively or XDR* or MDR*. The asterisk denotes a wild card. To increase the thoroughness of the literature search, the second key phrase with keywords in titles was combined with “tuberculosis” as a Medical Subject Heading (MeSH) and a third key phrase with the following keywords in titles joined by the Boolean operator “or”: efficacy or resistant or drug-resistant or multidrug-resistant. The search algorithm used in PubMed is shown in an appendix in the supplemental material.
Patients included in the review were required to have pulmonary XDR-TB or fluoroquinolone-resistant MDR-TB treated with WHO group 5 drugs alongside other anti-TB drugs. In the preliminary screening, we included all original articles with a focus on treatment efficacy of MDR-TB, the availability of fluoroquinolone susceptibility testing results, and individual patient data. Thus, review articles and commentaries, articles with a focus on aspects other than treatment efficacy, those containing only in vitro data, pharmacokinetic data, animal data, or genomic data, and those with no fluoroquinolone susceptibility testing results or individual patient data were excluded. Articles or individual patients were subsequently excluded for one or more of the following reasons: (i) no data on in vitro activity of fluoroquinolone, (ii) bacillary susceptibility to fluoroquinolone in vitro, (iii) disease involving only extrapulmonary sites, (iv) nonuse of WHO group 5 drugs, and (v) lack of evaluable outcome regarding treatment efficacy. Data of included articles were extracted from the published papers and any online repositories by the first author and were randomly checked by the other coauthors.
The assembled patients constituted a retrospective cohort that can be evaluated by cohort analysis with control for confounding factors at the individual patient level. Treatment efficacy of group 5 drugs was evaluated by examining the strength of the association between drug use and favorable outcome, which was defined as sputum culture conversion, cure, or treatment completion in the absence of death, default, treatment failure, or relapse.
As the majority of patients in the assembled cohort received linezolid, we first evaluated the relationship between favorable outcome and linezolid use, with consideration of the following potential confounding factors: gender, age group, HIV coinfection, other comorbidities that may predispose to TB disease, treatment supervision, adjunct surgery in the latest episode of anti-TB treatment involving the use of group 5 drugs, previous second-line TB treatment, additional resistance to any second-line injectable drug (SLID), use of adjunct immunotherapy, and concurrent use of other anti-TB drugs, including other group 5 drugs. If treatment duration was available, use of an anti-TB drug for less than 1 month was considered nonuse. When missing data exceeded 5%, a categorical subgroup for missing data was created to include missing data in the analysis. To optimize inclusion of major factors that may confound the relationship between favorable outcome and linezolid use, we followed basic principles in epidemiological studies with an emphasis on the strength of association rather than statistical significance (14). We screened potential confounders by examining risk ratios in univariate analysis rather than P values. A potential confounder must meet at least two criteria (14). First, it must be at least weakly associated with both outcome and the exposure variable. Risk ratios ≥ 1.2 or ≤ 0.9 were used to denote at least weak association (15). In the presence of a categorical subgroup for missing data, the potential confounder was included when the risk ratio involving the predominant comparison subgroup relative to the reference subgroup met the criteria for at least weak association. Second, a potential confounder must cause at least a 10% change in the coefficient of the exposure term when it is added in the regression model.
Using the approach described above, we ascertained the adjusted risk ratio of favorable outcome from linezolid use in model A with robust Poisson regression analysis, which is probably the best available method (16). To evaluate possible add-on effects of each nonlinezolid group 5 drug, we further constructed model B by force-entering use of each nonlinezolid group 5 drug into model A. Multicolinearity was considered before multivariable analysis and assessed by the variance inflation factor, which was regarded as unacceptable when its value exceeded 5 (17).
Univariate and robust Poisson regression analyses were conducted in R with the packages “gmodels” and “sandwich” (18), respectively.
To estimate the pooled risk ratio of favorable outcome from use of group 5 drugs, random-effects meta-analyses were conducted in R with the package “metafor” (19). To control for the confounding effect of linezolid, the pooled estimate of each nonlinezolid group 5 drug was ascertained after restricting analysis to subjects treated with linezolid-containing regimens. Publication bias was examined using the random-effects version of the Egger's regression test for funnel plot asymmetry (20). Heterogeneity between studies was tested by the chi-square test of heterogeneity (Q-test).
A total of 196 articles were initially identified in the literature search. These included prospective and retrospective observational studies. Figure 1 shows how a cohort of fluoroquinolone-resistant MDR-TB patients treated with WHO group 5 drugs was assembled from 20 articles (6, 21–39). The cohort consisted of 194 patients from 12 countries: 117 from four Asian regions (South Korea, China, Hong Kong, and India), 40 from five European countries (Portugal, Italy, Germany, Belgium, and Spain), 36 from two North and South American countries (the United States and Argentina), and 1 from South Africa.
Data were missing in >5% of the results in seven variables (proportion of missing data): gender (32%), age (32%), status with respect to previous second-line treatment (SLT) (18%), HIV status (12%), comorbidities other than HIV (36%), use of adjunctive surgery (11%), and mode of treatment supervision (29%). Among 131 patients with known gender and age, 77 (59%) were males and 54 (41%) were females. The median age (interquartile range) was 39 (range, 28 to 46) years. A total of 156 (98%) of 160 patients with available information had previous SLT. Among 170 patients with data on HIV status, 3 (2%) were HIV positive. Of 124 patients of known status with respect to comorbidities other than HIV, 23 (19%) had diseases that could predispose to TB disease. A total of 8 (5%) of 172 patients with available data had adjunctive surgery. Of 138 patients with data about the mode of treatment supervision, 80 (58%) had directly observed treatment (DOT).
Data were complete in the other variables, including the use of anti-TB drugs. A total of seven WHO group 5 drugs were included in the treatment of this cohort: linezolid, high-dose isoniazid (at least 10 mg/kg/dose), clofazimine, amoxicillin with clavulanate, macrolides (clarithromycin or azithromycin), carbapenem with or without clavulanate, and thioridazine. In descending order of frequency, linezolid was used in treatment of 162 (84%) patients, macrolides in 84 (43%), clofazimine in 65 (34%), amoxicillin with clavulanate in 56 (29%), thioridazine in 18 (9%), carbapenem with or without clavulanate in 16 (8%), and high-dose isoniazid in 16 (8%). None received thiacetazone.
Univariate analysis showed eight variables that might possibly associate with favorable outcome by consideration of risk ratios (see Table 1): directly observed treatment, previous second-line treatment, use of gamma interferon, use of rifabutin, use of linezolid, use of high-dose isoniazid, use of clofazimine, and use of thioridazine. Table 2 shows the results of univariate analysis of linezolid use and each of the other seven variables. Although each risk ratio denotes at least weak association, each variable caused <10% change in the coefficient for linezolid use upon inclusion in the robust Poisson regression model (findings not shown). Table 3 shows two robust Poisson regression models regarding favorable outcome and use of group 5 drugs. Model A shows that linezolid use substantially increased the probability (95% confidence interval [CI]) of favorable outcome by 57% (10% to 124%). Model B shows that adding all nonlinezolid group 5 drugs into model A hardly changes the risk ratio of favorable outcome from linezolid use. Furthermore, using risk ratios ≥ 1.2 or ≤ 0.9 to denote at least weak association, none of the nonlinezolid group 5 drugs confers any significant add-on benefit.
Table 4 summarizes the results of random-effects meta-analysis regarding favorable outcome and use of group 5 drugs. The Q-test showed no evidence of heterogeneity. The pooled estimate of risk ratio (95% CI) of favorable outcome from linezolid use relative to nonuse was 1.55 (1.10 to 2.21). Among patients given linezolid, corresponding pooled estimates were 0.95 (0.67 to 1.33) for high-dose isoniazid, 0.99 (0.76 to 1.31) for clofazimine, 1.01 (0.78 to 1.30) for amoxicillin plus clavulanate, 0.96 (0.76 to 1.22) for macrolides, 0.76 (0.48 to 1.22) for carbapenem with or without clavulanate, and 0.78 (0.54 to 1.13) for thioridazine. Forest plots of risk ratios are shown in the supplemental material. The random-effects version of the Egger's regression test for funnel plot asymmetry showed nonsignificant findings for each group 5 drug.
To our knowledge, this is the first systematic review with cohort analysis and meta-analysis that has evaluated the role of WHO group 5 drugs in the treatment of fluoroquinolone-resistant MDR-TB and XDR-TB and with emphasis on the strength of association rather than statistical significance. We have systematically collected individual patient data from 20 articles in English to assemble a cohort of 194 fluoroquinolone-resistant MDR-TB patients treated with WHO group 5 drugs. Statistical analyses showed neither between-study heterogeneity nor publication bias. Both cohort analysis using robust Poisson regression models and meta-analysis using random-effects models showed that use of linezolid substantially and significantly increased the probability of favorable outcome by 50% to 60%. Defining clinically significant improvement by risk ratios ≥ 1.2 or ≤ 0.9, neither cohort analysis nor meta-analysis demonstrated any add-on benefit from the use of the other group 5 drugs (high-dose isoniazid, clofazimine, amoxicillin with clavulanate, macrolides, carbapenem, and thioridazine) with respect to outcome for XDR-TB or fluoroquinolone-resistant MDR-TB patients treated with linezolid. Our findings further corroborate the important role of linezolid in the treatment of the most complicated forms of MDR-TB (9, 40) and call for further studies to evaluate the clinical roles of other group 5 drugs.
Although it may be unusual to conduct cohort analysis alongside meta-analysis, this is probably appropriate, as we have indeed assembled a cohort by well-defined criteria, and perhaps also necessary for the following reasons. Unlike meta-analysis based on quality randomized controlled trials, meta-analysis based on observational studies and case reports is inevitably prone to errors from selection bias and confounding. While meta-regression may help identify causes of heterogeneity between studies, it does not control for confounding factors at the individual patient level. Although cohort analysis by the standard methods of univariate and multivariable analyses cannot mitigate selection bias, it addresses major confounding factors and allows better estimation of the actual strength of association between drug use and outcome. To evaluate the impact of drug use on outcome, we have used robust Poisson regression analysis instead of logistic regression analysis because adjusted risk ratios (from the former) may be more direct and comprehensible than adjusted odds ratios (from the latter).
A number of uncontrolled clinical studies have demonstrated a beneficial role of linezolid in the treatment of complicated MDR-TB (10, 11). The need of giving linezolid for a prolonged period with its associated side effects has generally restricted its use to the most complicated MDR-TB (40). The prominent role of linezolid may be partly related to its high penetrability into sputum (39, 41, 42) and partly related to a very low MIC relative to achievable serum drug levels (21, 43). Despite a low mutant prevention concentration (44), acquired resistance to linezolid in Mycobacterium tuberculosis has emerged (21, 45). To ensure treatment success and prevent amplification of drug resistance, it is necessary to optimize the dosing schedule of linezolid and give as many likely effective companion drugs as possible. Although nonlinezolid group 5 drugs may not add much to linezolid activity, it is possible that one or more of them used alongside newer-generation fluoroquinolone and other second-line drugs may help protect linezolid from acquiring bacillary resistance and vice versa.
Selection bias might have led to underestimation of the therapeutic effects of nonlinezolid group 5 drugs, especially high-dose isoniazid, carbapenem, and thioridazine, as used in treatment of <10% of patients in the assembled cohort. With an excellent drug concentration in lung and high achievable serum drug levels, high-dose isoniazid may be beneficial when the isoniazid MIC is below 1 mg/liter and possibly below 5 mg/liter. In contrast to our findings, a recent case-control study has suggested that meropenem with clavulanate may nonsignificantly improve sputum culture conversion in linezolid-based treatment of MDR-TB (46). The add-on effect became statistically significant after excluding XDR-TB patients (46). Further studies may be required to better delineate the role of high-dose isoniazid and meropenem with clavulanate in the treatment of the most complicated form of MDR-TB. It has been suggested for over a decade that amoxicillin with clavulanate (47–49), macrolides (50, 51), and thioridazine (52, 53) may have a role in the treatment of MDR-TB. More data are required to address this controversy.
Clofazimine has been used as part of a 9-month gatifloxacin-containing regimen in the treatment of MDR-TB patients (54), but the targeted patients were largely SLT naive. Any observed benefit from the use of clofazimine could have been partly or entirely attributable to the coadministration of high-dose gatifloxacin. In a retrospective study involving predominantly MDR-TB patients with previous SLT (55), the role of clofazimine also appeared less impressive. With inclusion of this study through another article (34), our systematic review suggested that clofazimine might have only a limited role in the treatment of XDR-TB or fluoroquinolone-resistant MDR-TB treated with linezolid-containing regimens.
There are several limitations in this review. First and foremost is the relatively small sample size and possible selection bias. We have included only articles published in English in our systematic review. Owing to substantial difficulty in obtaining unpublished data from different investigators within a reasonably short time frame, we have excluded studies that have not released individual patient data in press or via online repositories. This approach could have introduced selection bias, but it is impossible to tell whether data from excluded studies might have ameliorated the bias. Although the Egger's regression test for funnel plot asymmetry showed nonsignificant findings for each group 5 drug, concerns about publication bias and other causes of selection bias cannot be totally excluded. Second, a substantial proportion of data were missing in seven variables. In the absence of an obvious systematic factor that may cause selection bias, these missing data were assumed to cause nondifferential rather than differential information bias (14). The drawback has probably been partly addressed by inclusion of missing data as a subcategory in the analysis. Third, we did not control for drug susceptibility testing results, but they would probably be closely related to drug use. Finally, most of the included studies either provided no definition regarding sputum culture conversion and cure or used different definitions (see the supplemental material). This is unfortunately a reality in the majority of the published literature on this subject.
In conclusion, our systematic review with meta-analysis and cohort analysis of individual patient data substantiates the use of linezolid in the treatment of XDR-TB or fluoroquinolone-resistant MDR-TB and calls for further studies to evaluate the roles of other group 5 drugs.
This study was conducted without any funding.
W.-W. Yew has received sponsorship from Pfizer to participate in international conferences in the past 3 years.
Published ahead of print 17 June 2013
Supplemental material for this article may be found at http://dx.doi.org/10.1128/AAC.00120-13.