Linezolid for drug‐resistant pulmonary tuberculosis

Description of the condition

Tuberculosis is caused by infection with bacteria of the Mycobacterium tuberculosis complex. It remains one of the leading infectious causes of death worldwide; there were 1.4 million deaths from tuberculosis worldwide in 2015, with an additional 0.4 million deaths from tuberculosis amongst people living with HIV (WHO 2016b). Pulmonary tuberculosis is the most common form of tuberculosis, and the most important from a public health perspective because tuberculosis is transmitted by aerosolized droplets from people with active pulmonary tuberculosis when they cough (Vashishtha 2013). It is estimated that around one third of the world's population are infected with tuberculosis, although of these only one in ten will develop active tuberculosis disease (WHO 2009).

Most people with tuberculosis are infected with strains of M tuberculosis that are treatable with the standard first‐line drugs recommended by the World Health Organization (WHO) guidelines: rifampicin, isoniazid, pyrazinamide, and ethambutol (WHO 2010). Early diagnosis and treatment with effective drugs is a mainstay of tuberculosis disease control, as well as being a life‐saving intervention for people with tuberculosis.

Multidrug‐resistant tuberculosis (MDR‐TB) is tuberculosis disease that is caused by M tuberculosis strains that have acquired resistance to two important drugs in the first‐line regimen: rifampicin and isoniazid (Sharma 2006). Rifampicin‐monoresistant tuberculosis is often managed as MDR‐TB (WHO 2016a). Extensively drug‐resistant tuberculosis (XDR‐TB), occurs when M tuberculosis strains are resistant to rifampicin, isoniazid, and any of the antibiotics in the fluoroquinolone class, as well as any of the three injectable drugs used in the second‐line treatment of tuberculosis: amikacin, kanamycin, and capreomycin (WHO 2016a).

The WHO estimates that 480,000 cases of MDR‐TB occurred in 2015, with 190,000 deaths worldwide, and an estimated 9.5% of people with MDR‐TB actually having XDR‐TB (WHO 2016b). Detection of drug‐resistant tuberculosis is challenging and currently requires costly laboratory services. Access to effective treatment is far from universal. Despite rapid progress, only 12% of new tuberculosis cases were tested for drug resistance in 2014, with case detection at only 41% (WHO 2015a). Over the last decade treatment success rates have remained static at around 50% (WHO 2015a), and the international tuberculosis community has recognized that new drugs and regimens with improved efficacy are urgently needed to improve cure rates. The WHO End TB Strategy outlines measures for post‐2015 tuberculosis control; these include a goal to detect and treat everyone with drug‐resistant tuberculosis, which will require significant scaling up of resources and efforts (WHO 2014).

Constructing drug‐resistant tuberculosis chemotherapy regimens is difficult; several of the available agents are expensive and toxic, and efficacy is uncertain because data from clinical studies are limited (Chang 2013a). This is especially true for XDR‐TB. Treatment for drug‐resistant tuberculosis is long: conventional regimens are administered for a total of 20 months for most patients, with an initial intensive phase of around eight months, dependent upon response to therapy (WHO 2016a). This has led to efforts being channeled towards investigation of new and existing drugs and regimens, with a drive to shorten treatment duration, standardize study design and reporting (Mitnick 2015), and focus on low‐resource settings that are disproportionately affected by tuberculosis and MDR‐TB globally (Sloan 2016).

Description of the intervention

Linezolid was categorized as a ‘Group 5' drug in the 2011 WHO drug‐resistant tuberculosis guidelines (WHO 2011). Medications assigned to this group were not recommended for use as core drugs, due to insufficient evidence detailing their safety or efficacy. However, the 2016 WHO update re‐allocated it as a ‘Group C: other core second‐line agent', prioritizing its use over some more traditional agents (WHO 2016a). The number of linezolid‐treated patients included in reviews of evidence informing both the 2011 and 2016 WHO guidance was insufficient to provide efficacy and safety estimates (WHO 2011, Fox 2017). In 2018, in a rapid communication from the WHO on treatment of MDR‐TB and rifampicin‐monoresistant‐TB, linezolid’s position was further upgraded to a ‘group A: Medicines to be prioritised' (WHO 2018). A summary of evidence for the 2018 recommendation has been published (Ahmad 2018).

Despite the promoted status of linezolid, concerns about serious adverse effects prompted the 2016 WHO update to caution that where close monitoring for adverse events is unavailable, “linezolid would best be reserved for MDR‐TB patients who have additional drug resistance...or who are intolerant to other components of the core regimen” (WHO 2016a). The 2018 WHO rapid communication, which recommends linezolid for all people with MDR‐TB unless it cannot be used, still states that, "Optimal duration of use of Lzd [linezolid] is not established. Use for at least 6 months was shown to be highly effective, although toxicity may limit its use” (WHO 2018).

Five meta‐analyses have examined the evidence for linezolid in drug‐resistant tuberculosis (Cox 2012; Sotgiu 2012; Chang 2013c; Zhang 2015; Ahmad 2018). They include mostly observational data, much of it retrospective. Few randomized studies have been undertaken. There remains much debate surrounding linezolid, due to the lack of high‐quality evidence. Many suggest it should be more widely used, hence its upgrade in the recent WHO guidance (Caminero 2015; Ahmad 2018; WHO 2018).

Considerable reliance on retrospective data may have exacerbated the effect of confounders in the meta‐analyses of treatment efficacy (Cox 2012; Sotgiu 2012; Chang 2013c; Zhang 2015; Ahmad 2018). These reviews also selected, and focus on, efficacy rather than safety. As highlighted by the WHO documents, safety is a major area of concern with linezolid (Ramachandran 2015).

How the intervention might work

Linezolid is an oxazolidinone antibiotic that disrupts protein synthesis by binding to the 70S initiation complex of bacterial ribosomes (Sloan 2016). It also binds to human mitochondria and inhibits protein synthesis, which is the mechanism of toxicity in clinical use (De Vriese 2006). It is active against most Gram‐positive bacteria, with extensive evidence of in vitro activity against isolates of M tuberculosis, including those resistant to first‐line drugs (Erturan 2005; Huang 2008).

Linezolid can be taken orally or intravenously. Its excellent oral bioavailability is an advantage, avoiding the need for long‐term daily injections (Dryden 2011). Though an adult dose of 600 mg twice daily is commonly used for up to 28 days to treat infections due to Gram‐positive bacteria, a variety of dosing strategies have been used in the context of drug‐resistant tuberculosis, where treatment duration is much longer. These have ranged from 300 mg to 1200 mg daily, with once‐ or twice‐daily administration. Lower doses have been tried in an attempt to increase tolerability and reduce toxicity (Park 2006; Migliori 2009; Yew 2009; Koh 2012). A thrice‐weekly intermittent dosing regimen has also been attempted in limited cohorts to extend the duration of linezolid therapy (Chang 2013b). The optimal dosing and duration of linezolid remains unclear from the perspective of preventing emergence of resistance, as well as efficacy, tolerability, and toxicity.

Adverse effects of linezolid include suppression of the bone marrow causing anaemia and thrombocytopenia, peripheral neuropathy, and optic neuropathy leading to disability and blindness, which is usually irreversible. More commonly, gastrointestinal upset may lead to difficulties with adherence (Ramachandran 2015). Adverse events with courses of linezolid longer than one month appear to be common within antituberculous drug regimens, affecting over 80% of participants in some studies (Lee 2012).

Why it is important to do this review

We set out to perform a systematic review reporting on the efficacy of linezolid for drug‐resistant tuberculosis, balanced against an estimate of the risk of linezolid‐associated adverse events. Such estimates will assist policy makers who are deciding on the place of linezolid in their national and regional drug‐resistant tuberculosis programmes, as well as individual clinicians trying to interpret the wide variety of published data on how effective, safe, and tolerable linezolid is in people being treated for MDR‐TB and XDR‐TB.

Existing evidence, while of low quality, has concluded that linezolid is efficacious in MDR‐TB, leading to its inclusion as a drug to be prioritized in the latest WHO guidance (WHO 2018). However, evidence regarding appropriate dosing and duration is lacking. Importantly, as linezolid is rolled out for wider use, closer interrogation of the adverse events data is desirable.

Criteria for considering studies for this review

Search methods for identification of studies

We attempted to identify all relevant studies regardless of language or publication status (published, unpublished, in press, and in progress).

Data collection and analysis

Description of studies

Risk of bias in included studies

We assessed risk of bias for the included RCTs using the Cochrane ‘Risk of bias' assessment tool (Higgins 2011a). We assessed the risk of bias in the cohort studies using ROBINS‐I tool (Sterne 2016). See the ‘Characteristics of included studies' section, which includes a ‘Risk of bias' table for each included study. We summarized the results of the ‘Risk of bias' assessments across all included RCTs in Figure 2 and non‐randomized studies in Figure 3.

Most of the subheadings that follow address risk of bias in RCTs; for non‐randomized studies, see the subheading Other potential sources of bias.

Effects of interventions

See: Table 1

Summary of main results

Table 3 and Table 4 summarize findings from the two RCTs for which we were able to assess intervention effect (104 participants), Lee 2012 and Tang 2015, respectively. Table 6 and Table 7 include a summary of adverse events findings from the 14 non‐randomized studies (1678 participants; 2 prospective, 12 retrospective). We were unable to generate pooled effect estimates using meta‐analysis due to heterogeneity of outcomes studied and reported. Table 1 provides a GRADE assessment of outcomes from Tang 2015.

Settings varied: the RCTs were based in the Republic of Korea (Lee 2012), and China (Tang 2015); three cohort studies recruited in the Republic of Korea, five in Europe (one included a centre in a former Soviet Union country), two from South Africa, one each in China and India, and two from multiple heterogeneous centres. Tang 2015, and seven of the 14 non‐randomized studies, commenced recruitment in 2009 or later.

Dosing and duration of linezolid in studies were variable, but also reported incompletely. Five studies did not report dosing at all. In the majority of the remainder it was not clear how many participants received each reported dose. Lee 2012 used 600 mg daily until a second planned randomization to continuing 600 mg or reducing to 300 mg daily. Tang 2015 used 1200 mg daily, then at four to six weeks, all were reduced to 300 mg or 600 mg, until sputum culture conversion. Only eight of the 14 non‐randomized cohorts stated a mean or median duration, all of which were reported to be longer than 90 days. Incompleteness of these data precluded comment on the effect of dose and duration of linezolid on outcomes. Follow‐up duration was variable, when reported; nine of the 14 non‐randomized studies did not report follow‐up duration.

Lee 2012 did not report data in a manner that permitted reporting of the primary outcomes of this review. However, their reporting of sputum culture conversion did permit comparison between those receiving linezolid immediately versus those starting it two months after randomization. Tang 2015 reported all of the review's primary and secondary outcomes, but reported sputum culture conversion in a way that made it difficult to compare directly with the data for that outcome reported by Lee 2012. In both studies, the group randomized to receive linezolid from the outset achieved a significantly higher proportion of sputum culture conversion from positive to negative at the time points specified by the study authors than the comparator group, who either started linezolid late or were not given it at all.

Tang 2015 reported significantly higher cure and lower failure amongst the linezolid‐treated group than controls, with no other significant differences in death, treatment completed and treatment interruption. The differences in cure and failure became insignificant when we performed worst‐case sensitivity analysis, though this method produces extreme effect estimates. Our level of certainty in the evidence was very low for cure and failure, following downgrading for risk of bias, indirectness and imprecision, as presented in Table 1.

Tang 2015 reported more anaemia, nausea, and vomiting, and neuropathy events amongst participants in the linezolid group compared with controls. Lee 2012 did not provide comparative adverse event data for those receiving linezolid versus those who did not. Linezolid was discontinued in seven out of 39 (17.9%) participants in Lee 2012 and two out of 33 (6.1%) participants in Tang 2015.

Where reported within the cohort studies, 141 out of 624 (22.6%; 11 cohorts), discontinued linezolid. We could not reliably compare total adverse events, serious adverse events, and overall and specific linezolid‐attributed adverse events, but we have shown these outcomes descriptively in Table 7. This is due to a lack of data on follow‐up duration and numbers of participants experiencing events.

Overall completeness and applicability of evidence

Settings of the studies, in terms of tuberculosis incidence and drug‐resistant tuberculosis prevalence, were diverse. India was the only lower‐middle‐income country (Udwadia 2010; Ferlazzo 2018), with the remainder being upper‐middle‐ or high‐income countries. Children were included in three studies, two of which exclusively recruited children (Seddon 2014; Galli 2016). Four non‐randomized studies did not report ages of their participants.

Reporting of linezolid dose and duration, and follow‐up was variable, as described in the Summary of main results and Table 2. Seven cohort studies included participants with HIV, of which two reported that most were taking antiretroviral therapy. Four did not report HIV status, and three reported that no participants were known to have HIV. The RCTs excluded people with known HIV (Lee 2012; Tang 2015).

All participants in the Lee 2012 and Tang 2015 RCTs had XDR‐TB. Amongst the 12 cohorts contributing adverse events outcome data disaggregated for linezolid receipt, 38.7% had XDR‐TB. Background regimens, where reported, were individualized to drug susceptibility results, as per WHO guidance, in all of the non‐randomized studies, but one RCT, Tang 2015, used a specified universal regimen, whilst Lee 2012 reported a variety of background regimens. Thoracic surgical interventions were undertaken in a minority (< 25%) of participants, and proportions appeared balanced between those who received linezolid and those who did not, where reported.

This review highlights the lack of RCT evidence, with only one, with no placebo or blinding, being suitable for analysis for primary and secondary outcomes (Tang 2015), and the Lee 2012 RCT providing limited comparative data for participants according to receipt of linezolid. Outcome reporting was poor overall in the non‐randomized studies, which were included for adverse events outcomes only. This means the evidence is neither complete, nor widely applicable.

Potential biases in the review process

We took measures to limit bias in the review process, by following procedures outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011b). The Cochrane Infectious Diseases Group (CIDG) Information Specialist conducted the literature search. It is unlikely that the search missed major studies, but some small unpublished studies may have been missed. We did not make a funnel plot, as included studies did not provide data suitable for meta‐analysis. Two of the review authors examined the search results, determined study selection, and extracted data independently, to minimize bias in study selection and data extraction.

Agreements and disagreements with other studies or reviews

We found five, previously published systematic reviews. These reviews are summarized in Table 8. All of these reviews conclude that linezolid is efficacious in the treatment of drug‐resistant tuberculosis, although authors comment on the high likelihood of adverse effects (Cox 2012; Sotgiu 2012; Chang 2013c; Zhang 2015; Ahmad 2018). Three reviews used studies as the unit of analysis, while two were individual patient data analyses. One review, Ahmad 2018, reported a risk of bias assessment. Only one of the reviews included a RCT (Ahmad 2018), and only two included a comparator group of people who did not receive linezolid (Chang 2013c; Ahmad 2018).

Cox 2012, Sotgiu 2012, and Zhang 2015 assessed treatment outcomes and adverse events in 11 (148 participants), 12 (121 participants), and 15 (367 participants) studies, respectively. Most of these studies were case series in which all participants received linezolid. Risk ratios could not be calculated due to the lack of comparative adverse events data on participants who did not receive linezolid. Cox 2012 and Sotgiu 2012 concluded that linezolid was efficacious for drug‐resistant tuberculosis, though both advised caution in its use due to high incidence of adverse events. Zhang 2015 suggested that linezolid was a "promising option as treatment of MDR/XDR TB", but advised randomized studies to define dosing.

Chang 2013c assembled a cohort from 20 studies reporting on the then‐named "group 5" anti‐tuberculous drugs, including 194 participants, of whom 162 received linezolid. They used a composite "favorable outcome" as the primary outcome, defined as "sputum culture conversion, cure, or treatment completion in the absence of death, treatment interruption, treatment failure, or relapse." Random‐effects meta‐analysis of "favorable outcome" according to linezolid use resulted in a pooled RR of 1.55 (95% CI 1.10 to 2.21), favouring linezolid. The outcomes in our review were not reported separately by Chang 2013c; in particular, there was no summary or meta‐analysis of adverse events outcomes.

Ahmad and colleagues conducted an individual patient data meta‐analysis of 50 studies reporting treatment outcomes in drug‐resistant tuberculosis, including 39 studies reporting use of linezolid (Ahmad 2018), of which one was a RCT included in our review (Lee 2012). Their primary outcomes were treatment success and death, with no summary of adverse events outcomes due to heterogeneity in measuring and reporting. The data were in favour of treatment success with linezolid use (722/799) versus without (5066/5864), with a crude odds ratio of 1.5 (95% CI 1.2 to 1.9), adjusted odds ratio 3.4 (95% CI 2.6 to 4.5), and adjusted risk difference 0.15 (0.11 to 0.18). Mortality was lower with linezolid use (84/883) versus without (1456/7320), with a crude odds ratio of 0.4 (95% CI 0.3 to 0.5), adjusted odds ratio 0.3 (95% CI 0.2 to 0.3), and adjusted risk difference −0.20 (95% CI −0.23 to −0.16). However, they found high heterogeneity (> 50%) in the studies overall. When XDR‐TB patients’ outcomes were meta‐analysed separately, the effect estimates remained in favour of linezolid use, with low heterogeneity amongst these studies (< 10%). Similar to our review, the authors highlighted a lack of data from RCTs, prospective studies, and low‐ and middle‐income settings (Ahmad 2018).

Our proportion of linezolid discontinuation (22.6%) was lower than the 36% pooled discontinuation found by Cox 2012, and 35% reported by Zhang 2015. The other three previous systematic reviews did not report discontinuation specifically.

When evidence for the use of linezolid was reviewed for the 2016 WHO guidelines (Annex 4 of WHO 2016a), the GRADE assessment for Tang 2015 concluded moderate certainty in the evidence for their comparison of treatment success versus a composite outcome of failure/relapse/death in patients with XDR‐TB. They downgraded for serious risk of bias and imprecision, but upgraded for a strong association. This is methodologically incorrect: upgrading for strong association is only for observational studies where GRADE starts as very low, and is not applicable to RCTs, where GRADE starts as high (Guyatt 2011). The WHO 2016a assessments for treatment success versus failure/relapse/death, and death versus all other outcomes in patients with both MDR‐ and XDR‐TB, when Tang 2015 and six non‐randomized studies were combined, resulted in very low certainty. Our GRADE assessment, with a population of MDR‐ and XDR‐TB in mind, was very low for all outcomes. This was in part due to downgrading by one level for indirectness (the population in Tang 2015 was limited to adults, with XDR‐TB, without HIV co‐infection, in one country), and two levels each for risk of bias (no blinding, no placebo, unclear randomization and allocation methods), and imprecision (small sample size, and for most outcomes, low number of events and wide CIs). We did not upgrade for a large effect size (Table 1).

Characteristics of included studies [author‐defined order]

Characteristics of excluded studies [ordered by study ID]

Characteristics of studies awaiting assessment [ordered by study ID]

Characteristics of ongoing studies [ordered by study ID]

Internal sources

• Liverpool School of Tropical Medicine, UK.

External sources

• Department for International Development, UK.

  Project number 300342‐104