Treatment Outcomes of Multidrug-Resistant Tuberculosis: A Systematic Review and Meta-Analysis

Background Treatment outcomes for multidrug-resistant Mycobacterium Tuberculosis (MDRTB) are generally poor compared to drug sensitive disease. We sought to estimate treatment outcomes and identify risk factors associated with poor outcomes in patients with MDRTB. Methodology/Principal Findings We performed a systematic search (to December 2008) to identify trials describing outcomes of patients treated for MDRTB. We pooled appropriate data to estimate WHO-defined outcomes at the end of treatment and follow-up. Where appropriate, pooled covariates were analyzed to identify factors associated with worse outcomes. Among articles identified, 36 met our inclusion criteria, representing 31 treatment programmes from 21 countries. In a pooled analysis, 62% [95% CI 57–67] of patients had successful outcomes, while 13% [9]–[17] defaulted, 11% [9]–[13] died, and 2% [1]–[4] were transferred out. Factors associated with worse outcome included male gender 0.61 (OR for successful outcome) [0.46–0.82], alcohol abuse 0.49 [0.39–0.63], low BMI 0.41[0.23–0.72], smear positivity at diagnosis 0.53 [0.31–0.91], fluoroquinolone resistance 0.45 [0.22–0.91] and the presence of an XDR resistance pattern 0.57 [0.41–0.80]. Factors associated with successful outcome were surgical intervention 1.91 [1.44–2.53], no previous treatment 1.42 [1.05–1.94], and fluoroquinolone use 2.20 [1.19–4.09]. Conclusions/Significance We have identified several factors associated with poor outcomes where interventions may be targeted. In addition, we have identified high rates of default, which likely contributes to the development and spread of MDRTB.


Introduction
Multidrug-Resistant Tuberculosis (MDRTB) refers to Mycobacterium tuberculosis (TB) strains with in vitro resistance to the two most effective anti-tuberculosis drugs, isoniazid (INH) and rifampin (RFP). MDRTB has become a major barrier to achieving successful control of TB, as therapy is less effective, associated with more adverse events and is more costly to treat when compared with standard first line therapy. According to a recent WHO report, approximately 490,000 MDRTB cases occur globally every year, corresponding to approximately 4.8% of the world's TB cases [1,2]. The importance of addressing drug resistant TB is further amplified by more recent reports on extensively drug resistant TB (XDRTB) [3], which represented 7% of MDR isolates referred to supranational reference laboratories from 2000-2004 [1].
Inadequate treatment of MDRTB can lead to worse patient outcomes, while increasing the risk of extensive drug resistance [4][5][6]. Guidelines for the management of MDRTB have been developed over the past decade, but there is little evidence based on randomized controlled trials to support current recommen-dations [7,8]. Moreover, treatment strategies have varied significantly and are difficult to compare between populations [8,9]. This lack of evidence reflects a lack of political and financial will, in part from the perception that MDRTB is of limited epidemiological importance [7]. It also reflects the limited number of second line drugs that are available and the unequal distribution of access depending on local resources. The recent recognition of the increasing magnitude of MDRTB, along with the poor prognosis of XDRTB has created the impetus for a more evidence-based approach to the treatment of MDRTB.
Recently, standardized definitions were established to allow comparison between treatment groups and facilitate the development of a more evidence-based approach [9,10]. We therefore decided to complete a systematic review of MDRTB treatment regimens. Where appropriate, we performed a meta-analysis to explore associations between MDRTB treatment outcomes and the clinical and microbiological factors that influence outcome. We aimed to identify all the published literature and to establish the best possible evidence base of clinical and microbiological predictors of treatment response.

Search strategy
Several search strategies were used to identify potentially relevant studies. Search strategy was developed by the investigators (Johnston and Shahidi) with consultation of a medical librarian.

Selection of Studies
Studies obtained from the literature search were checked by title and citation. If an article appeared relevant, the abstract was reviewed. Relevant abstracts were examined in full text. Inclusion criteria were as follows: an original study; reported in English; reported treatment outcomes in a population of adult, cultureconfirmed MDRTB patients; reported outcomes presented in a format allowing for comparison with other studies. Exclusion criteria were as follows: exclusive surgical series; exclusive use of first-line therapy in the treatment protocol.

Validity assessment
Studies were assessed for quality, with only high quality studies included for analysis. High quality studies reported outcomes on at least 10 patients; were prospective cohort, retrospective consecutive cohort, consecutive case control or randomized control in design; reported an average treatment duration of $12 months within an average follow-up duration $18 months; reported basic demographic data; reported less than 1/3 default or lost to followup. When study populations overlapped, we included the more recent and larger study population in the analysis. If the smaller population provided data on an outcome or variable not reported in the larger study, results were included for that specific variable.

Outcome measures
Measured outcomes reflect the definitions proposed by Laserson et al., and published in recent WHO guidelines [9,10]. Successful outcomes included patients meeting the definition of Cure or Treatment Completed. Unsuccessful outcomes included patients meeting the definition of Death, Defaulted, Failed or Transferred Out. When follow-up data was used, relapse was included as an unsuccessful outcome. To homogenize data, end-of-treatment (EOT) and follow-up (FUP) outcomes were separated for analysis. FUP outcomes refer to post-treatment follow-up, with follow-up duration measured in months. If studies were unable to meet WHO definitions, reviewers established outcomes to reflect these definitions. Certain studies reported data that precluded the use of WHO outcome definitions, and were not included for analysis. Several variables-including co-morbidities, demographic variables, microbiological profile, disease presentation and disease characteristics-were collected for outcome analysis. Analysis of these variables pooled both EOT and FUP outcomes given the paucity of homogenous data for individual variables. Where appropriate we contacted investigators for additional data as well as for clarification of their findings.

Data extraction and statistical analysis
Data abstraction was performed by two reviewers using a standardized abstraction form. When there was disagreement, the relevant paper was reviewed and differences were resolved by consensus. Microsoft Excel (version 12.0), STATSDirect version 2.7.6 (StatsDirect Ltd, Cheshire, UK), and STATA version 10.0 (STATA Corporation, College Station, Texas) software were used for data entry and analysis. Study characteristics and treatment outcomes were summarized in tables. Data related to treatment outcomes were pooled from published studies as described above. We pooled the proportion of successful outcome, death, default, transfer of care, and failure across studies using a random-effects meta-analysis. Heterogeneity of studies was estimated by calculating I 2 and Cochrane p values. The heterogeneity of binary covariates was estimated through p values. The overall effect of the dichotomous data was carried out by using a random effects analysis because of study heterogeneity (I 2 .50%, p,0.05) and measured by odds ratio with 95% confidence intervals [95%CI].

Results
We identified 9835 citations from our initial electronic database search. Of these, 116 articles were identified for full text review. A further 17 articles were identified through manual search, bibliographic search, and reviewer suggestions. Of the 133 articles reviewed in full text, 36 articles, representing 31 distinct patient populations were analyzed based on our a priori inclusion-exclusion criteria and quality assessment . Ninety-four studies were removed prior to analysis: 36 lacked adequate data on the patient population or outcomes; 17 evaluated an inappropriate population or overlapped with a larger study population; 5 reported outcomes based on first line therapy alone; 22 did not meet criteria for treatment duration or follow-up; 9 reported more than one-third default or loss to follow-up; 5 reported outcomes on ,10 patients; and 3 were excluded based on study design ( Figure 1).
The 36 articles chosen for review reported treatment outcomes on 31 study populations from 21 countries on 5 continents ( Table 1). The study periods ranged from 1973 to 2006. The majority were retrospective chart reviews, with five prospective cohorts [17,28,37,42,43] and one retrospective case control [32]. The average study population (range) was 210 . The majority of patients were male, with a mean of 68% (48-92) across all studies. In 30 studies with available data, the mean age was 40 years. All studies reported on patients with a prior history of treatment. In studies with available data, 84% (32-100) of patients had previously received anti-tuberculous therapy. The average number of drugs to which resistance was shown was 4.1 for studies reporting resistance. For those reporting first and second-line resistance, the average number of drugs to which there was resistance was 4.4.

Discussion
Drug resistant TB ultimately develops from the inadequate treatment of active pulmonary TB. There are multiple reasons for inadequate therapy; poor prescribing practices with insufficient treatment duration and poor drug selection are well-recognized contributors [5,47]. Systemic problems, through inadequate public health resources and unpredictable drug supplies also play a role [4]. In addition, irregular medication intake-whether from insufficient patient education, adverse events, or socioeconomic determinants-contribute to resistance. There are also a significant proportion of patients who acquire drug resistant disease because they live in an environment with a high prevalence of drug resistant disease. Despite these well-known causes of drug resistance, our review of 36 observational studies revealed very high rates of default and transfer of care. The majority of unsuccessful outcomes at end of treatment were the result of these factors, with relapse accounting for only 2% of patients in follow-up. Moreover, default appears to be a global phenomenon, with rates over 15% in several countries, including Korea (32%) [25], Taiwan (29%) [15], Russia (20%) [23], Italy (17%) [18], Spain (16%) [17], South Africa (29%) [35], Argentina (20%) [33], and Peru (19%) [29]. With the advent of DOTS and DOTS-Plus strategies, MDRTB rates have improved in some high prevalence countries, such as Latvia where new MDR cases have decreased more than two-fold between 2000 and 2007 (V. Leimane, unpublished data). Yet under these improved circumstances, published default rates were above 10% [27].
The consequences of inadequate therapy are also apparent in our data. On pooled analysis we have shown that prior treatment, fluoroquinolone resistance, and XDRTB were all associated with poor outcomes. While resistance to fluoroquinolones was associated with worse outcomes, fluoroquinolone use was associated with successful outcomes. This finding is consistent with several observational studies showing that fluoroquinolones are highly effective in MDRTB treatment [15,38,40,45,46,48,49]. The WHO Green Light Committee recommends that MDR patients receive a fluoroquinolone when their culture results reveal a susceptible isolate [10]. But judicious use of fluoroquinolones is crucial in suspected and confirmed cases of TB if we hope to impede the development of further resistance [29]. What is less clear is the role of newer generation fluoroquinolones in the treatment of fluoroquinolone resistant MDRTB. The findings of Yew suggest that newer generation fluoroquinolones may be effective against ofloxacin resistant strains [46]. Further studies will be necessary to clarify the role of fluoroquinolones in this population. The emerging XDRTB literature may be a resource to explore this issue.
An obvious target for comprehensive interventions is in the management of co-morbidities. The most commonly reported comorbidities in our selected studies were alcohol abuse, low BMI, and diabetes. The association between poor outcome with BMI and alcohol abuse is not unexpected. Body mass index, as a symptom of severe disease and low socioeconomic status requires aggressive intervention [29]. To our knowledge, there are no comparative studies examining the optimal nutritional intervention in an MDRTB population. Moreover, there are no evidence-based guidelines on nutritional supplementation in adults or children [54]. Further studies on MDRTB could not only include a baseline nutritional intervention, but also study the impact of various nutritional interventions on MDRTB outcomes. Meanwhile, alcohol abuse has been associated with poor TB outcome in several studies [16,23,35,36,55] and measures to improve care for alcohol abuse should be undertaken. A recently published study in Tomsk, Table 1. Cont. Russia, describes successful adoption of a program to address alcohol use disorders, with the intent to measure outcomes [56]. Diabetes mellitus is also a well-described co-morbidity in MDRTB patients. Not only are diabetics prone to reactivation TB [57], they may also be at increased risk of developing MDRTB [58,59]. The reasons for this are unclear, but based on our pooled data, diabetics do not appear to have worse outcomes with MDRTB. These results stand in contrast to previously published literature on drug sensitive TB [60], and may relate to our small sample size of 68 diabetics.
Lastly, our data supports the role of surgery in the management of MDRTB. Surgical patients appear to have superior outcomesthis may reflect the practice of performing surgery on less sick, adherent patients who have responded favourably to initial medical therapy. Although trials in surgery are inherently difficult, they should be undertaken to examine the role of surgery in MDRTB patients. Multiple surgical series have been published on this topic and could be examined for predictors of successful outcomes. In addition, future trials examining the role of surgery on highly resistant strains such as fluoroquinolone and XDR strains may be valuable.
Our study has several limitations. The most important is our inability to identify any randomized control trials (RCTs). Instead we relied exclusively on observational data for treatment outcomes. This likely introduced confounding to our pooled analysis, since crude outcomes, rather than adjusted odds ratios were reported for most trials. Moreover, the analysis of outcomes and pre-determined covariates used heterogeneous data from variable treatment regimens. The trends introduced by this analysis are subject to bias and should be interpreted with caution. Our quality analysis also likely introduced bias since only systems with resources for record keeping, second-line therapy and prolonged treatment and follow-up were included for analysis. These systems likely have improved outcomes compared with the many poorly resourced TB control programs throughout the world. Indeed, a recent editorial noted an ''exceptionality'' bias when commenting on reported XDRTB outcomes [61].
Of note, our treatment outcome definitions were heterogeneous between populations. Of the 36 studies included for analysis, only 12 studies (representing nine patient populations) met the outcome definitions proposed in WHO guidelines [10,16,[22][23][24][25][26][27]30,35,36,40,41]. The effect of this heterogeneity is difficult to assess. We observed, however, that several studies did not use the stringent culture-based outcome definitions such as Cure or Failed. This likely biased results towards more successful outcomes. Despite this, studies that used WHO outcomes showed similar outcomes: 62% [5-69] of patients had successful outcomes with a 15% [9][10][11][12][13][14][15][16][17][18][19][20][21][22] default and 10% [7][8][9][10][11][12][13][14] death rate. Another significant limitation is the lack of data on HIV positive patients. In total 143 of 6359 patients (2%) were known to be HIV positive, with the data from 39 patients available for outcome analysis. This limits the power to detect the effect of the presence of HIV infection. Some studies reported HIV infected populations with treatment and follow-up durations shorter than 12 and 18 months respectively [3,62,63]. Thus, our inclusion criteria for treatment duration and follow-up likely excluded a disproportionate population of HIV positive patients from our analysis. In addition, our criteria eliminated studies examining 'outbreaks' of MDRTB, where a more virulent strain spreads quickly through a community. Such outbreak investigations tend to have limited follow-up and outcome data given the urgency to report findings [64,65].
Recently, a systematic review examining MDRTB treatment outcomes in 33 populations was published by Orenstein et al. [66]. In this review, 62% of patients were classified as successful outcomes. Covariate analysis revealed that the proportion of patients with successful treatment outcomes did not significantly change for several individual covariates. The authors concluded that this likely reflected study heterogeneity. Similarly, our review examined MDRTB treatment outcomes. We have, however, focused on underlying patient and disease characteristics-such as diabetes, alcoholism, and smear positivity-while, Orenstein et al. focused on programmatic characteristics such as length of treatment and the use of Directly Observed Therapy. Overall, our outcomes do not  differ markedly from those of Orenstein et al., and we feel that our covariate analysis contributes to the literature on this topic. Though treatment success remains dismally low in MDRTB populations, there is reason to be hopeful. With the advent of standardized nomenclature, outcomes of MDRTB treatment regimens can be evaluated with more precision. The addition of more comprehensive therapeutic strategies will also help improve outcomes. But, an evidence-based approach, in addition to comprehensive, individualized therapy, is necessary to contain MDRTB [7]. Without high quality evidence to support therapeutic decisions, including data from high-risk populations such as HIV positive and diabetic patients, high mortality, default, and failure rates will persist.