PLoS MedPLoS MedPLoSplosmedPLoS Medicine1549-12771549-1676Public Library of ScienceSan Francisco, USA229524393429397PMEDICINE-D-11-0235610.1371/journal.pmed.1001300Research ArticleMedicineGlobal HealthInfectious DiseasesPublic HealthMultidrug Resistant Pulmonary Tuberculosis Treatment Regimens and Patient Outcomes: An Individual Patient Data Meta-analysis of 9,153 PatientsIPD Meta-analysis of MDR-TB TreatmentAhujaShama D.1AshkinDavid2AvendanoMonika3BanerjeeRita4BauerMelissa5BayonaJamie N.6BecerraMercedes C.78BenedettiAndrea5BurgosMarcos9CentisRosella10ChanEward D.11ChiangChen-Yuan12CoxHelen13D'AmbrosioLia10DeRiemerKathy14DungNguyen Huy15EnarsonDonald16FalzonDennis17FlanaganKatherine18FloodJennifer19Garcia-GarciaMaria L.20GandhiNeel21GranichReuben M.17Hollm-DelgadoMaria G.5HoltzTimothy H.22IsemanMichael D.23JarlsbergLeah G.24KeshavjeeSalmaan7KimHye-Ryoun25KohWon-Jung26LancasterJoey27LangeChristophe28de LangeWiel C. M.29LeimaneVaira30LeungChi Chiu31LiJiehui32MenziesDick5*MiglioriGiovanni B.10MishustinSergey P.33MitnickCarole D.7NaritaMasa34O'RiordanPhilly35PaiMadhukar5PalmeroDomingo36ParkSeung-kyu37PasvolGeoffrey38PeñaJose39Pérez-GuzmánCarlos40QuelapioMaria I. D.41Ponce-de-LeonAlfredo42RiekstinaVija30RobertJerome43RoyceSarah24SchaafH. Simon44SeungKwonjune J.45ShahLena5ShimTae Sun46ShinSonya S.45ShiraishiYuji47Sifuentes-OsornioJosé48SotgiuGiovanni49StrandMatthew J.23TabarsiPayam50TupasiThelma E.41van AltenaRobert29Van der WaltMartie27Van der WerfTjip S.29VargasMario H.51ViikleppPirett52WestenhouseJanice53YewWing Wai54YimJae-Joon55on behalf of the Collaborative Group for Meta-Analysis of Individual Patient Data in MDR-TBCollaborative Group for Meta-Analysis of Individual Patient Data in MDR-TBAhujaShama D.Bureau of Tuberculosis, New York, New York, United States of AmericaAshkinDavidA.G. Holley Hospital, Lantana, Florida, United States of AmericaAvendanoMonikaUniversity of Toronto, Toronto, Canada,BanerjeeRitaMayo Clinic, Rochester, Minnesota, United States of AmericaBauerMelissaMontreal Chest Institute, McGill University, Montreal, CanadaBayonaJamie N.The Dartmouth Center for Health Care Delivery Science, Hanover, New Hampshire, United States of AmericaBecerraMercedes C.Harvard Medical School, Boston, Massachusetts, United States of America; Partners in Health, Boston, Massachusetts, United States of AmericaBenedettiAndreaMontreal Chest Institute, McGill University, Montreal, CanadaBurgosMarcosUniversity of New Mexico School of Medicine, Albuquerque, New Mexico, United States of AmericaCentisRosellaWHO Collaborating Centre for TB and Lung Diseases, Care and Research Institute, Fondazione S. Maugeri, Tradate, ItalyChanEward D.Denver Veterans Affair Medical Center, Denver, Colorado, United States of AmericaChiangChen-YuanWan Fang Hospital, School of Medicine-Taipei Medical University, TaiwanCoxHelenMédecins Sans Frontières, Capetown, South AfricaD'AmbrosioLiaWHO Collaborating Centre for TB and Lung Diseases, Care and Research Institute, Fondazione S. Maugeri, Tradate, ItalyDeRiemerKathyUC Davis School of Medicine, Davis, California, United States of AmericaDungNguyen HuyNational TB Control Program, Hanoi, VietnamEnarsonDonaldInternational Union against Tuberculosis and Lung Disease, Paris, FranceFalzonDennisWorld Health Organization, Geneva, SwitzerlandFlanaganKatherineMRC Laboratories, Banjul, The GambiaFloodJenniferCalifornia Department of Public Health, Sacramento, California, United States of AmericaGarcia-GarciaMaria L.Instituto Nacional de Salud Pública, Mexico, MexicoGandhiNeelAlbert Einstein College of Medicine, Bronx, New York, United States of AmericaGranichReuben M.World Health Organization, Geneva, SwitzerlandHollm-DelgadoMaria G.Montreal Chest Institute, McGill University, Montreal, CanadaHoltzTimothy H.Thailand MOPH & US CDC Collaboration, Bangkok, ThailandIsemanMichael D.National Jewish Health, Denver, Colorado, United States of AmericaJarlsbergLeah G.University of California, San Francisco, San Francisco, United States of AmericaKeshavjeeSalmaanHarvard Medical School, Boston, Massachusetts, United States of AmericaKimHye-RyounKorea Cancer Center Hospital, Seoul, KoreaKohWon-JungSamsung Medical Center, Seoul, KoreaLancasterJoeySouth African Medical Research Council, Pretoria, South AfricaLangeChristopheMedical Clinic, Tuberculosis Center Borstel, Borstel, Germanyde LangeWiel C. M.University Medical Center Groningen, Groningen, The NetherlandsLeimaneVairaClinic of Tuberculosis and Lung Diseases, Riga, LatviaLeungChi ChiuTuberculosis and Chest Services, Hong KongLiJiehuiNew York City Health and Mental Hygiene, New York, New York, United States of AmericaMenziesDickMontreal Chest Institute, McGill University, Montreal, CanadaMiglioriGiovanni B.WHO Collaborating Centre for TB and Lung Diseases, Care and Research Institute, Fondazione S. Maugeri, Tradate, ItalyMishustinSergey P.Tomsk Oblast Tuberculosis Dispensary, Tomsk, RussiaMitnickCarole D.Harvard Medical School, Boston, Massachusetts, United States of AmericaNaritaMasaUniversity of Washington, Seattle, Washington, United States of AmericaO'RiordanPhillyCity Road Medical Centre, London, United KingdomPaiMadhukarMontreal Chest Institute, McGill University, Montreal, CanadaPalmeroDomingoHospital F.J. Muñiz, Buenos Aires, ArgentinaParkSeung-kyuTB Center, Seoul, KoreaPasvolGeoffreyImperial College London, London, United KingdomPeñaJoseUniversidad Autonoma Madrid, Madrid, SpainPérez-GuzmánCarlosInstituto de Salud del Estado de Aguascalientes, Mexico, MexicoQuelapioMaria I. D.Tropical Disease Foundation, Makati City, PhilippinesPonce-de-LeonAlfredoInstituto Nacional de Ciencias Médicas y de Nutrición “Salvador Zubirán”, Mexico, MexicoRiekstinaVijaClinic of Tuberculosis and Lung Diseases, Riga, LatviaRobertJeromeBactériologie-Hygiène – UPMC, Paris, FranceRoyceSarahUniversity of California, San Francisco, San Francisco, United States of AmericaSchaafH. SimonStellenbosch University, Stellenbosch, South AfricaSeungKwonjune J.Brigham and Women's Hospital, Boston, Massachusetts, United States of AmericaShahLenaMontreal Chest Institute, McGill University, Montreal, CanadaShimTae SunUniversity of Ulsan College of Medicine, Seoul, KoreaShinSonya S.Brigham and Women's Hospital, Boston, Massachusetts, United States of AmericaShiraishiYujiFukujuji Hospital, Tokyo, JapanSifuentes-OsornioJoséInstituto Nacional de Ciencias Médicas y de Nutrición “Salvador Zubirán”, Mexico, MexicoSotgiuGiovanniUniversity of Sassari, Sassari, ItalyStrandMatthew J.National Jewish Health, Denver, Colorado, United States of AmericaTabarsiPayamShaheed Beheshti Medical University, Tehran, IranTupasiThelma E.Tropical Disease Foundation, Makati City, Philippinesvan AltenaRobertUniversity Medical Center Groningen, Groningen, The NetherlandsVan der WaltMartieSouth African Medical Research Council, Pretoria, South AfricaVan der WerfTjip S.University Medical Center Groningen, Groningen, The NetherlandsVargasMario H.Instituto Nacional de Enfermedades Respiratorias, Mexico, MexicoViikleppPirettNational Institute for Health Development, Tallinn, EstoniaWestenhouseJaniceCenter for Infectious Diseases-California Department of Public Health, Sacramento, California, United States of AmericaYewWing WaiGrantham Hospital, Hong KongBureau of Tuberculosis, New York, New York, United States of AmericaA.G. Holley Hospital, Lantana, Florida, United States of AmericaUniversity of Toronto, Toronto, CanadaMayo Clinic, Rochester, Minnesota, United States of AmericaMontreal Chest Institute, McGill University, Montreal, CanadaThe Dartmouth Center for Health Care Delivery Science, Hanover, New Hampshire, United States of AmericaHarvard Medical School, Boston, Massachusetts, United States of AmericaPartners in Health, Boston, Massachusetts, United States of AmericaUniversity of New Mexico School of Medicine, Albuquerque, New Mexico, United States of AmericaWHO Collaborating Centre for TB and Lung Diseases, Care and Research Institute, Fondazione S. Maugeri, Tradate, ItalyDenver Veterans Affair Medical Center, Denver, Colorado, United States of AmericaWan Fang Hospital, School of Medicine-Taipei Medical University, TaiwanMédecins Sans Frontières, Capetown, South AfricaUC Davis School of Medicine, Davis, California, United States of AmericaNational TB Control Program, Hanoi, VietnamInternational Union against Tuberculosis and Lung Disease, Paris, FranceWorld Health Organization, Geneva, SwitzerlandMRC Laboratories, Banjul, The GambiaCalifornia Department of Public Health, Sacramento, California, United States of AmericaInstituto Nacional de Salud Pública, Mexico, MexicoAlbert Einstein College of Medicine, Bronx, New York, United States of AmericaThailand MOPH & US CDC Collaboration, Bangkok, ThailandNational Jewish Health, Denver, Colorado, United States of AmericaUniversity of California, San Francisco, San Francisco, United States of AmericaKorea Cancer Center Hospital, Seoul, KoreaSamsung Medical Center, Seoul, KoreaSouth African Medical Research Council, Pretoria, South AfricaMedical Clinic, Tuberculosis Center Borstel, Borstel, GermanyUniversity Medical Center Groningen, Groningen, The NetherlandsClinic of Tuberculosis and Lung Diseases, Riga, LatviaTuberculosis and Chest Services, Hong KongNew York City Health and Mental Hygiene, New York, New York, United States of AmericaTomsk Oblast Tuberculosis Dispensary, Tomsk, RussiaUniversity of Washington, Seattle, Washington, United States of AmericaCity Road Medical Centre, London, United KingdomHospital F.J. Muñiz, Buenos Aires, ArgentinaTB Center, Seoul, KoreaImperial College London, London, United KingdomUniversidad Autonoma Madrid, Madrid, SpainInstituto de Salud del Estado de Aguascalientes, Mexico, MexicoTropical Disease Foundation, Makati City, PhilippinesInstituto Nacional de Ciencias Médicas y de Nutrición “Salvador Zubirán”, Mexico, MexicoBactériologie-Hygiène – UPMC, Paris, FranceStellenbosch University, Stellenbosch, South AfricaBrigham and Women's Hospital, Boston, Massachusetts, United States of AmericaUniversity of Ulsan College of Medicine, Seoul, KoreaFukujuji Hospital, Tokyo, JapanInstituto Nacional de Ciencias Médicas y de Nutrición “Salvador Zubirán”, Mexico, MexicoUniversity of Sassari, Sassari, ItalyShaheed Beheshti Medical University, Tehran, IranInstituto Nacional de Enfermedades Respiratorias, Mexico, MexicoNational Institute for Health Development, Tallinn, EstoniaCenter for Infectious Diseases-California Department of Public Health, Sacramento, California, United States of AmericaGrantham Hospital, Hong KongSeoul National University College of Medicine, Seoul, KoreaEvansCarltonAcademic EditorUniversidad Peruana Cayetano Heredia, Peru

¶ All members of the Collaborative Group for Meta-Analysis of Individual Patient Data in MDR-TB are the authors and are listed in the manuscript byline.

* E-mail: Dick.Menzies@McGill.ca

JR is a Consultant for bioMérieux. WWY has been indirectly sponsored to participate in International Conferences by GlaxoSmithKline and Pfizer in the last 3 years. CDM is on the Scientific Advisory Board for Otsuka pharmaceuticals development of OPC67683 (Delaminid), a new anti-TB compound. SK received salary support from the Eli Lilly Foundation as part of funding for the activities of Partners In Health by the Foundation's MDR-TB Partnership. This funder was not involved in the study design; collection, analysis and interpretation of data; writing of the paper; and/or decision to submit for publication. The Partners In Health project in Tomsk received funding from Mr. Tom White, the Open Society Institute, the Bill and Melinda Gates Foundation, and the Global Fund to fight AIDS, Tuberculosis and Malaria. None of these funders were involved in the study design; collection, analysis and interpretation of data; writing of the paper; and/or decision to submit for publication. KD is an unpaid, volunteer member of the New Diagnostics Working Group (NDWG), formed of members of the Stop TB Partnership. The Secretariat of the NDWG is hosted by FIND (Foundation for New Innovative Diagnostics). JB was working as consultant for Otsuka Pharmaceutical for the implementation of clinical trial in Peru. JB was co PI of a NIH grant in Peru, Epidemiology of Tuberculosis. MP and GP are members of the Editorial Board of PLOS Medicine. All other authors have declared that no competing interests exist.

Performed the experiments. Analyzed the data: SDA DA MA RB MBa JB MCB AB MBu RC EDC CYC HC LD KD NHD DE DF KF JF MLGG NRG RG MGHD THH MDI LGJ SK HRK WJK JLL CL WCMD VL CCL JL DM GBM SM CDM MN PO MP DP SKP GP JMP CPG MIDQ AP VR JR SR HSS KJS LS TSS SSS YS JSO GS MS PT TEP RVA MLV TSV MHV PV JW WWY JJY. Contributed reagents/materials/analysis tools: SDA DA MA RB MBa JB MCB AB MBu RC EDC CYC HC LD KD NHD DE DF KF JF MLGG NRG RG MGHD THH MDI LGJ SK HRK WJK JLL CL WCMD VL CCL JL DM GBM SM CDM MN PO MP DP SKP GP JMP CPG MIDQ AP VR JR SR HSS KJS LS TSS SSS YS JSO GS MS PT TEP RVA MLV TSV MHV PV JW WWY JJY. Wrote the first draft of the manuscript: DM. Contributed to the writing of the manuscript: SDA DA MA RB MBa JB MCB AB MBu RC EDC CYC HC LD KD NHD DE DF KF JF MLGG NRG RG MGHD THH MDI LGJ SK HRK WJK JLL CL WCMD VL CCL JL DM GBM SM CDM MN PO MP DP SKP GP JMP CPG MIDQ AP VR JR SR HSS KJS LS TSS SSS YS JSO GS MS PT TEP RVA MLV TSV MHV PV JW WWY JJY. ICMJE criteria for authorship read and met: SDA DA MA RB MBa JB MCB AB MBu RC EDC CYC HC LD KD NHD DE DF KF JF MLGG NRG RG MGHD THH MDI LGJ SK HRK WJK JLL CL WCMD VL CCL JL DM GBM SM CDM MN PO MP DP SKP GP JMP CPG MIDQ AP VR JR SR HSS KJS LS TSS SSS YS JSO GS MS PT TEP RVA MLV TSV MHV PV JW WWY JJY. Agree with manuscript results and conclusions: SDA DA MA RB MBa JB MCB AB MBu RC EDC CYC HC LD KD NHD DE DF KF JF MLGG NRG RG MGHD THH MDI LGJ SK HRK WJK JLL CL WCMD VL CCL JL DM GBM SM CDM MN PO MP DP SKP GP JMP CPG MIDQ AP VR JR SR HSS KJS LS TSS SSS YS JSO GS MS PT TEP RVA MLV TSV MHV PV JW WWY JJY.

8201282012288201298e1001300279201117720122012Ahuja et alThis is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Dick Menzies and colleagues report findings from a collaborative, individual patient-level meta-analysis of treatment outcomes among patients with multidrug-resistant tuberculosis.

Background

Treatment of multidrug resistant tuberculosis (MDR-TB) is lengthy, toxic, expensive, and has generally poor outcomes. We undertook an individual patient data meta-analysis to assess the impact on outcomes of the type, number, and duration of drugs used to treat MDR-TB.

Methods and Findings

Three recent systematic reviews were used to identify studies reporting treatment outcomes of microbiologically confirmed MDR-TB. Study authors were contacted to solicit individual patient data including clinical characteristics, treatment given, and outcomes. Random effects multivariable logistic meta-regression was used to estimate adjusted odds of treatment success. Adequate treatment and outcome data were provided for 9,153 patients with MDR-TB from 32 observational studies. Treatment success, compared to failure/relapse, was associated with use of: later generation quinolones, (adjusted odds ratio [aOR]: 2.5 [95% CI 1.1–6.0]), ofloxacin (aOR: 2.5 [1.6–3.9]), ethionamide or prothionamide (aOR: 1.7 [1.3–2.3]), use of four or more likely effective drugs in the initial intensive phase (aOR: 2.3 [1.3–3.9]), and three or more likely effective drugs in the continuation phase (aOR: 2.7 [1.7–4.1]). Similar results were seen for the association of treatment success compared to failure/relapse or death: later generation quinolones, (aOR: 2.7 [1.7–4.3]), ofloxacin (aOR: 2.3 [1.3–3.8]), ethionamide or prothionamide (aOR: 1.7 [1.4–2.1]), use of four or more likely effective drugs in the initial intensive phase (aOR: 2.7 [1.9–3.9]), and three or more likely effective drugs in the continuation phase (aOR: 4.5 [3.4–6.0]).

Conclusions

In this individual patient data meta-analysis of observational data, improved MDR-TB treatment success and survival were associated with use of certain fluoroquinolones, ethionamide, or prothionamide, and greater total number of effective drugs. However, randomized trials are urgently needed to optimize MDR-TB treatment.

Please see later in the article for the Editors' Summary.

Editors' SummaryBackground

In 2010, 8.8 million people developed tuberculosis—a contagious bacterial infection—and 1.4 million people died from the disease. Mycobacterium tuberculosis, the bacterium that causes tuberculosis, is spread in airborne droplets when people with the disease cough or sneeze and usually infects the lungs (pulmonary tuberculosis). The characteristic symptoms of tuberculosis are a persistent cough, weight loss, and night sweats. Tuberculosis can be cured by taking several powerful antibiotics regularly for at least 6 months. The standard treatment for tuberculosis comprises an initial intensive phase lasting 2 months during which four antibiotics are taken daily followed by a 4-month continuation phase during which two antibiotics are taken. However, global efforts to control tuberculosis are now being thwarted by the emergence of M. tuberculosis strains that are resistant to several antibiotics, including isoniazid and rifampicin, the two most powerful, first-line (standard) anti-tuberculosis drugs.

Why Was This Study Done?

Although multi-drug resistant tuberculosis (MDR-TB) can be cured using second-line anti-tuberculosis drugs, these are more expensive and more toxic than first-line drugs and optimal treatment regimens for MDR-TB have not been determined. Notably, there have been no randomized controlled trials of treatments for MDR-TB. Such trials, which compare outcomes (cure, treatment failure, relapse, and death) among patients who have been randomly assigned to receive different treatments, are the best way to compare different anti-tuberculosis drug regimens. It is possible, however, to get useful information about the association of various treatments for MDR-TB with outcomes from observational studies using a statistical approach called “individual patient data meta-analysis.” In observational studies, because patients are not randomly assigned to different treatments, differences in outcomes between treatment groups may not be caused by the different drugs they receive but may be due to other differences between the groups. An individual patient data meta-analysis uses statistical methods to combine original patient data from several different studies. Here, the researchers use this approach to investigate the association of specific drugs, numbers of drugs and treatment duration with the clinical outcomes of patients with pulmonary MDR-TB.

What Did the Researchers Do and Find?

The researchers used three recent systematic reviews (studies that use predefined criteria to identify all the research on a given topic) to identify studies reporting treatment outcomes of microbiologically confirmed MDR-TB. They obtained individual patient data from the authors of these studies and estimated adjusted odds (chances) of treatment success from the treatment and outcome data of 9,153 patients with MDR-TB provided by 32 centers. The use of later generation quinolones, ofloxacin, and ethionamide/prothionamide as part of multi-drug regimens were all associated with treatment success compared to failure, relapse or death, as were the use of four or more likely effective drugs (based on drug susceptibility testing of mycobacteria isolated from study participants) during the initial intensive treatment phase and the use of three or more likely effective drugs during the continuation phase. The researchers also report that among patients who did not die or stop treatment, the chances of treatment success increased with the duration of the initial treatment phase up to 7.1–8.5 months and with the total duration of treatment up to 18.6–21.5 months.

What Do These Findings Mean?

These findings suggest that the use of specific drugs, the use of a greater number of effective drugs, and longer treatments may be associated with treatment success and the survival of patients with MDR-TR. However, these findings need to be interpreted with caution because of limitations in this study that may have affected the accuracy of its findings. For example, the researchers did not include all the studies they found through the systematic reviews in their meta-analysis (some authors did not respond to requests for individual patient data, for example), which may have introduced bias. Moreover, because the patients included in the meta-analysis were treated at 32 centers, there were many differences in their management, some of which may have affected the accuracy of the findings. Because of these and other limitations, the researchers note that, although their findings highlight several important questions about the treatment of MDR-TB, randomized controlled trials are urgently needed to determine the optimal treatment for MDR-TB.

Additional Information

Please access these Web sites via the online version of this summary at http://dx.doi.org/10.1371/journal.pmed.1001300.

The World Health Organization provides information on all aspects of tuberculosis, including MDR-TB; its guidelines for the programmatic management of drug-resistant tuberculosis are available

The US Centers for Disease Control and Prevention has information about tuberculosis, including information on the treatment of tuberculosis and on MDR-TB

The US National Institute of Allergy and Infectious Diseases also has information on all aspects of tuberculosis, including a drug-resistant tuberculosis visual tour

MedlinePlus has links to further information about tuberculosis (in English and Spanish)

TB & ME, a collaborative blogging project run by patients being treated for multidrug-resistant tuberculosis and Medecins sans Frontieres, provides information about MDR-TB and patient stories about treatment for MDR-TB

The Tuberculosis Survival Project, which aims to raise awareness of tuberculosis and provide support for people with tuberculosis, also provides personal stories about treatment for tuberculosis

Partial funding for the assembly of individual patient data and meta-analysis was provided from the Stop TB Department of the World Health Organization, through a grant from USAID. Funding for data gathering at participating centres came from the following sources: in the State of California from the US Centers for Disease Control Cooperative Agreement Funds; in Italy from the European Community's Seventh Framework Programme (FP7/2007–2013) under grant agreement FP7- 223681; in Mexico (Veracruz) from the Mexican Secretariat of Health, the National Institutes of Health of the United States (A135969 and K01TW000001), the Wellcome Trust (176W009), the Howard Hughes Medical Institute (55000632), and the Mexican Council of Science and Technology: SEP 2004-C01-47499, FOSSIS 2005-2 (14475), (87332); in South Africa from the South African Medical Research Council funding. Funding was provided to the following investigators: M Bauer and D Menzies were supported by salary awards from the Fonds de Recherche en Sante de Quebec, L Shah was supported by CIHR (Canada Graduate Scholarship), Neel Gandhi received a Doris Duke Charitable Foundation Clinical Scientist Development Award. The funders had no role in study design, data collection and analysis, decisions to publish,or preparation of the manuscript.
Introduction

The increasing incidence of multidrug resistant tuberculosis (MDR-TB), defined as resistance to at least isoniazid and rifampin, is a major concern for TB control programs worldwide. MDR-TB treatment requires prolonged use of multiple second-line anti-TB drugs, which are more expensive and toxic than first-line drugs, yet less efficacious [1]. As a result of these problems, administration of MDR-TB treatment imposes substantial operational challenges in resource constrained settings. Further, the optimal composition and duration of MDR-TB treatment regimens is uncertain [1],[2].

Three systematic reviews have recently examined determinants of treatment outcomes in MDR-TB [3][5]. However, these three reviews identified no randomized trials, and the majority of the observational studies identified reported results with individualized treatment. There were considerable differences between studies in the diagnostic methods used, treatment regimens given, and clinical characteristics of the patient populations. As a result, these meta-analyses could only analyze pooled odds of treatment success associated with proportions of patients with specific clinical characteristics or receiving specific treatments. This approach has considerable limitations for a clinical problem of this complexity.

Even in the absence of randomized trials, an individual patient data meta-analysis of observational data offers potential benefits. Detailed patient level information can be used to estimate associations of treatment factors with outcomes, stratified by important covariates, within restricted sub-groups, or adjusted for covariates in meta-regression. We conducted an individual patient data meta-analysis using patient level data combined from different centers, using methods suggested by the Cochrane group [6]. We addressed several questions formulated by an expert committee of the World Health Organization (WHO) responsible for revision of guidelines for treatment of MDR-TB [7]. These questions included the impact of specific drugs, number of drugs, and duration of treatment on clinical outcomes of patients with pulmonary MDR-TB.

MethodsSelection of Studies

The studies considered for this individual patient data meta-analysis were identified from published original studies included in three recent systematic reviews of MDR-TB treatment outcomes [3][5]. All three reviews included studies published after 1970 that reported original data of treatment of patients with microbiologically confirmed MDR-TB. Additional specific criteria for this meta-analysis were: the study authors could be contacted and were willing to share their data; the cohort included at least 25 subjects treated for MDR-TB; and, at least treatment success, as defined below, was reported. Patients within these datasets were excluded if they had only extra-pulmonary TB, had extensive drug resistance (XDR-TB, as defined elsewhere [8]), or were missing treatment information.

Data Sharing, Extraction, and Verification

Letters describing the meta-analysis were communicated to all corresponding authors of eligible studies. The McGill investigators signed formal data sharing agreements with all collaborating investigators regarding sharing of results, publications, and “ownership” of the data. This project was approved by the Research Ethics Board of the Montreal Chest Institute, McGill University Health Centre, and when deemed necessary by local ethics boards of originally approved studies.

Each author provided center-level information such as diagnostic laboratory methods, treatment regimen doses and supervision, and outcome definitions. Regimens were considered individualized if regimens were tailored to individual patients' characteristics such as prior therapy, or drug susceptibility testing (DST) results. Authors also provided de-identified patient level information including age, sex, HIV infection, site of disease, results of chest x-ray, acid fast bacilli (AFB) smear, culture, and DST for first and second-line drugs, drugs used and duration for initial and continuous phases of treatment, surgical resection, and outcomes, including adverse events that required a change in therapy. Treatment outcome definitions provided by each author were compared to the consensus definitions published by Laserson et al. [9], and rated as the same, closely similar, or not similar. These definitions are summarized at the top of Table S3. Relapse was any recurrence of disease after successful treatment completion, and was combined with failure for all analyses. For this analysis we considered the following as part of group 5 drugs: amoxicillin-clavulanate, macrolides (azithromycin, roxithromycin, and clarithromycin), clofazimine, thiacetazone, imipenem, linezolid, high dose isoniazid, and thioridazine.

Authors were contacted to request additional data and clarify variable definitions and coding. Variables from each original dataset were extracted, their meaning and coding verified, then mapped to a common set of variables for all patients. Hence datasets from each center had the same variables for all patients, but each dataset was kept distinct. As a final verification, the clinical characteristics of each study population were compared with the original published papers.

Data Analysis

We considered three types of drug-exposure in our meta-analysis: (i) specific drugs administered (grouped as suggested by WHO [1]), (ii) duration of treatment regimen, and (iii) number of likely effective drugs used. Drugs were considered likely effective if susceptible on drug susceptibility testing, regardless of history of prior use. We estimated odds of treatment success (defined as treatment cure or completion) compared to one of three alternate outcomes: (i) treatment failure or relapse; (ii) treatment failure, relapse or death; and (iii) treatment failure, relapse, death or default. For duration of therapy, comparisons (ii) and (iii) were not analyzed because in studies with individualized therapy the planned duration was not recorded—only the actual duration, which was truncated by default or death during treatment.

We used random effects (random intercept and random slope) multi-variable logistic regression estimated via penalized quasi-likelihood (Proc Glimmix in SAS [10]) in order to estimate the adjusted odds and 95% CIs of treatment success associated with different treatment covariates [11][13]. As a sensitivity analysis, all models for primary analyses were also estimated using adaptive quadrature (QUAD) [14]. Patients were considered as clustered within studies and intercepts and slopes of the main exposure variables were allowed to vary across studies; this is to account for otherwise unmeasured inter-study differences in patient populations, as well as center-specific differences in data ascertainment, measurement, and other factors. The variance of the study specific intercepts (here the baseline log odds of success in each cohort) and slopes (here treatment efficacy) were interpreted to indicate how much these varied across the studies. We report the average estimate of effect across studies from these models and the estimated inter-study variability and standard deviation of that variance, as well as the variance of the intercept and the standard deviation of that variance.

Estimates of effect of each treatment parameter for each dataset were adjusted for five covariates: age, gender, HIV co-infection, extent of disease (considered extensive if AFB smear positive, or if AFB smear information was missing, then if there was cavitation on chest x-ray), and past history of TB treatment (a three level variable—no previous TB treatment, previous TB treatment with first-line drugs, and previous treatment with second-line drugs). Analyses were performed in all patients and in subgroups—stratified or restricted by important covariates. We tested for the interaction between previous treatment with second-line TB drugs and the number of drugs and duration of treatment in the intensive and continuation phases, respectively. In secondary analyses we included more than one treatment parameter (up to four drugs at once), and individual drugs with treatment duration. For the multivariable analyses only, missing values of these five clinical covariates were imputed using means of patients at the same center with non-missing information. Sex was missing in three patients, age was missing in 27, HIV was missing in 1,271(14%), history of past TB treatment missing in 443 (5%), history of past second-line drug use 758 (8%), and extent of disease information missing in 174 (2%). We assessed whether findings were altered when missing values were estimated using a different method of probabilistic imputation, in secondary analysis [15].

Additionally, heterogeneity was explored visually using Forest plots of study specific estimates, and estimated quantitatively via the I2 and its associated 95% CI [13]. For these analyses, estimates of effect were calculated separately for each study, adjusting for relevant patient-level covariates, and pooled using conventional meta-analytic techniques. All analysis was performed using SAS, version 9.2 (SAS Institute).

ResultsSelection of Studies and Description of the Study Population

We identified 67 unique cohorts from the three previous systematic reviews of MDR-TB. Of these, 35 datasets were not used for reasons summarized in Figure 1, leaving 32 datasets with information on 9,898 patients [16][57]. From these we excluded 410 patients with XDR, 208 with inadequate treatment information, and 127 with solely extra-pulmonary TB, leaving 9,153 individual patients analyzed; their clinical characteristics and outcomes are summarized in Table 1.

10.1371/journal.pmed.1001300.g001Study selection.10.1371/journal.pmed.1001300.t001Clinical characteristics and treatment received of patients included in the analysis.
Demographic CharacteristicsDataDataData
Mean age in years (SD) (25 missing)38.7 (13.6)
Male sex (%) (three missing)6,280 (69%)
Clinical characteristicsYesNoMissing
AFB – smear positive (n, %)6,012 (66%)1,878 (21%)1,263 (14%)
Cavities on x-ray (n, %)4,723 (52%)2,019 (22%)2,411 (26%)
Extensive disease (n, %)6,753 (74%)2,226 (24%)174 (2%)
HIV positive (n, %)1,077 (12%)6,805 (74%)1,271 (14%)
Pulmonary TB only (n, %)8,713 (96%)232 (2%)208 (2%)
Prior TB therapy (any)6,683 (73%)2,027 (22%)443 (5%)
Prior therapy with second-line drugs942 (10%)7,455 (82%)756 (8%)
Resistance to other drugsSensitiveResistantNot tested
Ethambutol (n, %)2,736 (30%)4,065 (44%)2,352 (26%)
Pyrazinamide (n, %)2,406 (26%)2,443 (27%)4,304 (47%)
Streptomycin (n, %)2,454 (27%)4,154 (45%)2,545 (28%)
Treatment received
Rifabutin (n, %)130 (1.4%)
Ethambutol (n, %)4,722 (52%)
Pyrazinamide (n, %)6,571 (72%)
Ciprofloxacin (n, %)986 (11%)
Ofloxacin (n, %)6,489 (71%)
Later generation quinolones (n, %)1,258 (14%)
Streptomycin (n, %)1,326 (14%)
Kanamycin (n, %)5,002 (55%)
Amikacin (n, %)428 (5%)
Capreomycin (n, %)1,757 (19%)
Ethionamide (n, %)3,873 (42%)
Prothionamide (n, %)3,709 (41%)
Cycloserine (n, %)5,344 (58%)
Para-aminosalicylic acid (PAS) (n, %)3,196 (33%)
One group 5 drug2,115 (23%)
Two or more group 5 drugs594 (7%)
Outcomes (mutually exclusive)
Success (cure and treatment completed)4,934 (54%)
Failure645 (7%)
Relapse87 (1%)
Default, transfer out, unknown2,095 (23%)
Died during MDR-TB treatment1,392 (15%)

Percentages are of all 9,153 patients. Extensive disease defined as AFB-smear positive, or cavities on chest x-ray if no information about AFB-smear. Prior TB therapy: defined as treatment with any, or second-line TB drugs for 1 mo or more. Later generation quinolones included levofloxacin, moxifloxacin, gatifloxacin, and sparfloxacin. Cycloserine included terizidone—a dimer of D-cycloserine given in some centers. Drugs analysed as group 5 included: amoxicillin-clavulanate, macrolides (azithromycin, roxithromycin, clarithromycin), clofazimine, thiacetazone, imipenem, linezolid, high dose INH, and thioridazine. Relapse ascertained in only 2,261 patients (14 cohorts).

SD, standard deviation.

The included studies originated from 23 countries, from all WHO health regions. Final sample sizes included in the analysis ranged from 18 to 2,174 patients. In the supplement are summarized: study and center characteristics (Table S1a), excluded studies (Table S1b), doses of drugs commonly used (Table S2), and outcome definitions—the accepted standards and those used in each series (Table S3). Treatment was individualized in 26 studies with 5,985 patients, and standardized in six studies with 2,968 patients. A total of 200 patients in two centers received standardized regimens with first-line drugs only; all remaining patients received second-line drugs. In all but one study, the outcome definitions for treatment success and failure were judged the same or similar to the consensus definitions [9]. Overall 4,934 (54%) of patients were judged to have treatment success, 732 (8%) failed or relapsed, 1,392 (15%) died, and 2,095 (23%) defaulted (Table 1). Pooled treatment success, compared to failure/relapse/death, was 75% but varied widely between studies (Figure 2), while pooled success, compared to fail/relapse/death and default was only 54% (Figure 3). Adverse drug reactions that resulted in changed therapy occurred in 1932 (21%), and 499 (5%) patients underwent pulmonary resection surgery. Patients who died were significantly older, more likely to be HIV co-infected, with more extensive disease, and/or had prior therapy—with first-line and particularly with second-line TB drugs. Those who defaulted were significantly older, and more likely to have HIV co-infection (data not shown in tabular form).

10.1371/journal.pmed.1001300.g002Treatment success versus failure and relapse and death.

Crude treatment success versus failure or relapse or death by study with exact 95% CI, as well as number of subjects with success and number of subjects treated. Fixed and der Simonian and Laird random effects pooled estimates are given (purple dots). Two studies that used only first-line TB drugs are indicated by a red square.

10.1371/journal.pmed.1001300.g003Treatment success versus failure and relapse and death.

Fixed and der Simonian and Laird random effects pooled estimates are given (purple dots). Two studies that used only first-line TB drugs are indicated by a red square.

Treatment Correlates of Outcomes

As seen in Table 2, the use of later generation quinolones, or ofloxacin, or ethionamide/prothionamide, as part of multidrug regimens, was significantly associated with treatment success compared to failure or relapse. The use of these same drugs, as well as pyrazinamide or cycloserine were significantly associated with treatment success compared to failure, relapse, or death, while the use of later generation quinolones, ofloxacin, ethionamide/prothionamide or kanamycin were significantly associated with treatment success compared to failure, relapse, death, or default. Treatment outcomes of the 594 patients who received two or more group 5 drugs were significantly worse than the 2,115 patients who received one group 5 drug, whose outcomes were, in turn, worse than those of the 6,444 patients who received none of these drugs. Since this finding likely reflected confounding by indication for use of these drugs (i.e., group 5 drugs were used more often in patients with more severe disease or worse resistance), the analysis of the four most commonly used group 5 drugs—amoxicillin-clavulanate, macrolides, clofazimine, and thiacetazone—was restricted to patients who received only one group 5 drug. This revealed that none of these four drugs, compared to the others, was associated with significantly superior treatment outcomes. There was often significant heterogeneity in baseline odds of treatment success and in the treatment effect (as seen in the large variance relative to its standard deviation for the intercept and slope, respectively; these are provided in detail in Table S9.)

10.1371/journal.pmed.1001300.t002Summary of association of use of individual drugs with treatment success.
Drug UsednaaORbSuccess Versus Failure/Relapse (95% CI)naaORbSuccess Versus Failure/Relapse/Death (95% CI)naaORbSuccess Versus Failure/Relapse/Death/Default (95% CI)
Group 1 drugs
Pyrazinamide3,9851.2(0.9–1.7)c5,0961.3(1.1–1.6)c6,5711.1(0.9–1.4)c
Ethambutol2,8190.9(0.7–1.1)d3,7400.8(0.7–0.9)c4,7190.9(0.8–1.2)c
Group 2: injectables
Kanamycin only2,8603,4374,457
Versus no injectable1.1(0.5–2.3)e1.3(0.7–2.6)e1.3(0.7–2.5)e
Versus capreomycin1.3(0.7–2.7)c1.6(1.1–2.4)1.3(1.1–1.6)d
Versus streptomycin1.1(0.6–2.2)e1.0(0.6–1.6)e1.0(0.8–1.3)d
Amikacin only192248307
Versus no injectable1.5(0.6–4.1)d1.7(0.8–3.3)1.3(0.5–3.6)e
Capreomycin only7699401,127
Versus no injectable1.1(0.5–2.6)c1.3(0.5–3.7)e1.1(0.4–3.2)e
Group 3: quinolones
Later gen. quinolones751829974
Versus no quinolones2.6(0.6,10.5)c2.5(1.0–5.9)2.8(1.3–6.1)c
Versus ofloxacin1.6(0.5–5.3)e1.9(1.0–3.6)2.1(1.2–3.9)c
Versus ciprofloxacin0.7(0.1,3.8)e1.5(0.5–4.6)e1.7(0.6–4.9)d
Ofloxacin3,8324,5776,102
Versus no quinolones2.5(1.8–3.9)c2.5(1.6–3.9)2.0(1.2–3.3)e
Versus ciprofloxacin1.1(0.5–2.5)c1.4(0.7–2.6)1.3(0.7–2.5)e
Ciprofloxacin335553644
Versus no quinolones1.5(0.6–4.1)d2.0(0.8–5.2)1.6(0.6–4.3)e
Group 4 drugs
Ethionamide/prothionamide4,6081.6(1.2–2.3)c5,5941.9(1.6–2.2)d7,3291.7(1.5–2.0)d
Cycloserine/terizidone3,5471.1(0.8–1.7)c4,1941.5(1.0–2.3)5,3581.5(0.9–2.2)e
PAS2,4591.0(0.8–1.3)2,8601.0(1.0–1.4)d3,7121.2(1.0–1.5)c
Group 5 drugs
Any 1 group 5 versus none1,5380.6(0.4–0.9)c1,7250.7(0.6–0.8)c2,0291.0(0.8–1.2)d
2+ group 5 versus one group 54470.4(0.3–0.6)c5740.5(0.3–0.6)c6540.7(0.5–0.9)d
Amox.-clavulanate onlyf2321.0(0.4–2.5)c2551.2(0.6–2.6)c2901.4(0.8–2.5)c
Clofazimine onlyf6512.7(0.6–12.1)e7642.3(0.4–12.4)e8961.0(0.5–2.1)e
Macrolide onlyf3330.4(0.3–0.6)d3960.5(0.4–0.7)c4590.8(0.6–1.1)c
Thiacetazone onlyf5540.8(0.5–1.5)d5761.0(0.6–1.7)d6681.0(0.7–1.4)d

Bold, estimates are significantly different from the reference group.

n shown, the number of patients that received the drug in question and were included in the specific analysis.

aOR for use of drug, with non-use as the reference category. Adjusted for age, sex, HIV, past TB treatment, past MDR treatment (treatment for more than 1 mo with two or more second-line drugs), and extent of disease. Missing information was imputed for the following parameters in the following number of patients: Sex was missing in three, age was missing in 27, HIV was missing in 1,271 (14%), history of past TB treatment missing in 443 (5%), history of past second-line drug use 758 (8%), and extent of disease information missing in 174 (2%).

Variance of the random intercepts and slopes low, so heterogeneity not likely to be important.

Variance of the random intercepts and slopes could not be estimated.

Variance of the random intercepts and slopes high, so heterogeneity likely important.

Group 5 individual drugs: Analysis restricted to patients who received only one group 5 drug. Each single drug comparison made between patients who received only that group 5 agent with patients who received any other single group 5 drug. Drugs included in this analysis as group 5 drugs were: amoxicillin-clavulanate, macrolides (azithromycin, roxithromycin, and clarithromycin), clofazimine, thiacetazone, imipenem, linezolid, high dose INH, and thioridazine. Later generation quinolones included levofloxacin, moxifloxacin, gatifloxacin, and sparfloxacin. Cycloserine included terizidone—a dimer of D-cycloserine given in some centers.

Patients with prior treatment with second-line drugs were significantly less likely to have HIV co-infection, but were more likely to have cavitary disease, and strains with resistance to other first-line drugs (Table S4). In these patients the odds of treatment success with the individual drugs were similar, although CIs were broad (Table S5).

As shown in Table 3, compared to use of three or fewer likely effective drugs in the initial intensive phase, the odds of success were significantly greater with use of four drugs, and were very similar with use of five, six, or more drugs (Table 3). In the continuation phase, compared to use of two or fewer likely effective drugs, use of three drugs was associated with significantly superior odds of success, which were similar to the odds of success with use of four, or five or more likely effective drugs (Table 3). Fewer patients were included in these analyses because only a subset of studies provided information on the number of likely effective drugs used in the initial intensive phase (18 studies) or the continuation phase (15 studies). In patients with prior treatment with second-line drugs the maximal odds of success was seen with five likely effective drugs in the initial intensive phase (Table 4), and five drugs in the continuation phase (Table 4). There was substantial heterogeneity and a statistically significant interaction between prior treatment with second-line drugs and number of drugs used in the continuation phase (p = 0.01), and in the initial intensive phase (p = 0.05) (Tables S10). In further exploratory analyses, there was no association between the number of drugs used in the initial phase with default, but default was more frequent in patients who received more drugs in the continuation phase (Table S11).

10.1371/journal.pmed.1001300.t003Association of number of likely effective drugs with treatment success—during different phases of treatment.
n Likely Effective Drugs – All Patients – Three AnalysesAll Patients – Success Versus Fail/RelapseAll Patients – Success Versus Fail/Relapse/DeathAll Patients – Success Versus Fail/Relapse/Death/Default
naOR (95% CI)naOR (95% CI)naOR (95% CI)
Initial intensive phase
0–21181.0 (reference)2771.0 (reference)3221.0 (reference)
31611.1 (0.5–2.4)a2501.7 (1.2–2.5)a3161.2 (0.8–1.8)a
44682.0 (1.1–3.6)a5422.7 (1.9–3.9)a6711.9 (1.3–2.9)a
58142.0 (1.1–3.6)a9002.8 (1.7–4.6)a1,1141.9 (1.2–3.0)a
6+8112.4 (1.0–5.4)a9772.1 (1.4–3.1)a1,1851.4 (1.0–2.1)a
Continuation phase
0–22541.0 (reference)5311.0 (reference)6331.0 (reference)
35522.5 (1.6–4.0)b6355.7 (3.4–9.7)a7594.9 (2.7–8.7)a
45982.8 (1.6–4.9)a6635.7 (3.2–10.0)a7794.2 (2.6–6.7)a
5+5602.0 (0.9–4.2)a6087.0 (5.1–9.7)b6564.9 (2.5–9.5)a

Likely effective, drugs to which isolate susceptible in laboratory testing. n, number of patients in subgroup of interest. aOR, adjusted for age, sex, HIV, past TB treatment, past MDR treatment (treatment for more than 1 mo with two or more second-line drugs), and extent of disease. Missing information was imputed for the following parameters in the following number of patients: Sex was missing in three, age was missing in 27, HIV was missing in 1,271(14%), history of past TB treatment missing in 443 (5%), history of past second-line drug use 758 (8%), and extent of disease information missing in 174 (2%). Success, defined as cure or treatment completion; see Methods for definitions. Initial intensive phase, period when injectable given. Continuation phase, period when no injectable given. Only 18 studies provided information regarding drug susceptibility testing and the number of drugs in the initial phase, while only 15 of these described the number of drugs in the continuation phase. Bold, estimates are significantly different from the reference group.

Variance of the random intercepts and slopes was low—so heterogeneity not likely to be important.

Variance of the random intercepts and slopes could not be estimated.

10.1371/journal.pmed.1001300.t004Effect of previous treatment on association of number of likely effective drugs with treatment success—during different phases of treatment.
n Likely Effective Drugs – Patients Stratified by Treatment HistoryAll Patients Success Versus Fail/Relapse/DeathNo Prior MDR Treatment Success Versus Fail/Relapse/DeathPrior MDR Treatment Success Versus Fail/Relapse/Death
naOR(95% CI)naOR(95% CI)naOR(95% CI)
Initial phase
0–22771.0(reference)2461.0(reference)61.0(reference)
32501.7(1.2–2.5)1862.3(1.4–2.9)16
45422.7(1.9–3.9)3853.4(2.2–5.2)661.4(0.5–3.8)
59002.8(1.7–4.6)6502.6(1.7–4.0)1113.4(1.2–9.8)
6+9772.1(1.4–3.1)5153.1(2.0–4.9)3271.6(0.3–9.1)
Continuation phase
0–25311.0(reference)4671.0(reference)321.0(reference)
36355.7(3.4–9.7)5325.5(2.2–13.6)465.5(2.3–13.1)
46635.7(3.2–10.0)4303.3(1.8–6.3)899.1(4.4–18.8)
5+6087.0(5.1–9.7)2654.6(1.5–14.0)21113.7(8.2–23.0)

Likely effective, drugs to which isolate susceptible in laboratory testing. n, number of patients in subgroup of interest. aOR, adjustment described in footnotes for Table 3. Success, defined as cure or treatment completion; see Methods for definitions. Initial intensive phase, period when injectable given. Continuation phase, period when no injectable given. Only 18 studies provided information regarding drug susceptibility testing and the number of drugs in the initial phase, while only 15 of these described the number of drugs in the continuation phase. Bold, estimates are significantly different from the reference group.

Among those who did not die or default, odds of treatment success increased with longer duration of the initial intensive phase up to the duration of 7.0 to 8.4 mo (Table 5), although CIs were wide for each interval estimate (Figure 4), and variance estimates were also high (Tables S12). There was no statistically significant interaction between intensive phase or total treatment duration and prior treatment with second-line drugs. Similarly odds of success increased progressively with longer total duration of therapy up to 24.6–27.5 mo duration. In patients with prior second-line drug therapy the maximal odds of success was seen with total duration of 27.6–30.5 mo, although there were very few patients in each strata, and CIs were correspondingly wide. As seen in Tables S6S8, additional secondary analyses of optimal duration revealed similar results for: analyses with adjustment for use of four drugs, in addition to adjustment for the five clinical covariates used for all other models (Table S6A), patients who received only one injectable and experienced no injectable-related adverse events (Table S6B), or patients who received regimens containing at least one second-line drug (Table S6C). In patients who received later generation quinolones the maximal odds of success was seen with a shorter duration of total therapy (Table S6D), although CIs were very wide, and included 1.0. There was a significant trend toward more frequent history of prior second-line drug treatment, but lower HIV prevalence with longer duration of initial intensive phase (Table S8a) and total duration (Table S8b) and somewhat greater prevalence of resistance to pyrazinamide, ethambutol, fluoroquinolones, or second-line injectables with longer initial intensive or total duration.

10.1371/journal.pmed.1001300.g004Association of treatment success with duration (adjusted odds and upper bound of CI shown).

(A) Duration of initial intensive phase in all patients (reference group 1.0–2.5 mo). (B) Duration of initial intensive phase—restricted to patients not previously treated with second-line drugs (reference group 1.0–2.5 mo). (C) Total duration of therapy in all patients (reference group is 6.0–12.5 mo). Patients receiving therapy for less than 6 or more than 36 mo excluded from analysis. Note: For duration of 24.6–27.5 mo the upper limit of the CI was 30.2. This is truncated at 21. (D) Total duration of therapy—analysis restricted to patients not previously treated with second-line drugs (reference group is 6.0–12.5 mo. Patients receiving therapy for less than 6 or more than 36 mo excluded from analysis). Note: For duration of 24.6–27.5 mo, the upper limit of the CI was 56.5. This is truncated at 21.

10.1371/journal.pmed.1001300.t005Association of duration of treatment with success versus failure/relapse—patients grouped by treatment history.
Duration of Treatment (mo) InterceptAll PatientsNo Prior Second-Line Drug TreatmentPrior Second-Line Drug Treatment
naOR (95% CI)naOR (95% CI)naOR (95% CI)
Initial
1–2.43081.0 (reference)2711.0 (reference)61.0 (reference)
2.5–3.91,4061.3 (0.5–3.2)a1,2981.5 (0.6–4.2)b234.2 (0.5–34.3)b
4.0–5.44812.3 (1.2–4.2)b4182.2 (0.8–6.5)b1510.9 (1.0–117.8)b
5.5–6.93773.8 (2.0–7.3)b3143.8 (1.8–8.0)b2647.2 (3.0–746.1)b
7.0–8.41724.8 (1.9–11.8)b1244.9 (1.8–13.5)b21
8.5–207922.1 (1.2–3.8)b5171.9 (0.6–5.6)a22826.3 (3.8–183.9)b
Total
6.0–12.57781.0 (reference)6811.0 (reference)331.0 (reference)
12.6–15.54191.5 (0.6–3.6)a3211.4 (0.5–4.2)a340.4 (0.2–1.1)b
15.6–18.51,7003.6 (1.5–8.7)a1,5273.6 (1.1–11.6)a512.2 (0.7–6.8)b
18.6–21.56555.2 (2.0–11.5)a346.2 (1.8–20.6)a401.6 (0.6–4.5)b
21.6–24.55534.9 (2.1–11.5)b4006.3 (2.3–16.8)b1056.5 (2.2–19.7)b
24.6–27.531311.7 (4.5–30.2)b17012.9 (3.0–56.5)b1048.1 (2.1–31.4)b
27.6–30.51602.8 (1.0–7.6)b893.2 (1.0–10.0)b5313.6 (1.6–114.1)b
30.6–36891.2 (0.2–5.8)a362.8 (0.4–19.7)b382.0 (0.6–7.3)b

n, number of patients in subgroup of interest. aOR, adjusted odds ratios—adjustment described in footnotes for Table 3. Success, defined as cure or treatment completion and is compared to failure or relapse (see methods for definitions). Other outcomes of death and default not assessed in this analysis because in some datasets shorter duration that was directly due to death or default could not be identified. Past treatment, prior MDR means past treatment for more than 1 mo with two or more second-line drugs. No prior MDR includes all other treatment history.

Variance of the random intercepts and slopes high—so heterogeneity likely important. (See Tables S1S12 for actual values).

Variance of the random intercepts and slopes low—so heterogeneity not likely to be important. (See also Tables S1S12 for actual values).

As a sensitivity analysis, all analyses reported in Tables 24 were repeated using random effects logistic regression estimated via adaptive quadrature. Results, except where indicated in the tables, were very similar to those estimated using penalized quasi-likelihood. In addition analyses using probabilistic imputation for missing values produced very similar results as results using the imputation methods described above.

Discussion

To our knowledge this is the largest combined analysis of treatment of MDR-TB, and the first individual patient data meta-analysis of treatment outcomes in drug resistant TB. With the detailed individual clinical information for 9,153 patients it was possible to use stratified, restricted, and/or multivariable analyses to control for differences in treatment regimens, drug resistance patterns, prior treatment histories, and other patient characteristics such as HIV co-infection. Overall treatment results were poor—treatment success was achieved in only slightly more than half of all patients. Treatment success was significantly associated with specific durations, number of likely effective drugs for the initial intensive and continuation phases of therapy, and with use of later generation quinolones (levofloxacin, moxifloxacin, gatifloxacin, and sparfloxacin), ofloxacin or ethionamide/prothionamide. These results helped to inform the forthcoming revised MDR-TB treatment guidelines of WHO, and should be useful in planning therapy for individual patients.

We suggest cautious interpretation of these results in light of a number of important limitations. First, we included only 32 studies out of a possible 67 series that had reported outcomes of MDR-TB treatment and were identified in three systematic reviews. This selection may have introduced some bias, although as seen in Table S1b, the characteristics and outcomes of patients in the included and excluded studies were similar. Second, we included two studies in which most patients received only first-line drugs; this may seem obviously inadequate, but earlier reports supported such treatment [58]. However, only 200 patients were treated in this way, and findings were not changed when they were excluded from analyses. Third, all the data included in this review were derived from observational studies, most of which utilized individualized regimens; this may have introduced bias if certain drugs, or combinations and durations of drugs were preferentially used for patients with more extensive drug resistance, more severe disease, or worse co-morbidities. An example of this selection bias was the finding that use of any group 5 drug was associated with worse outcomes, particularly use of two or more group 5 drugs. As recommended [59], we controlled this potential confounding by restricting the analysis to patients who received only one of the group 5 agents. Fourth, the effects of individual drugs may have been difficult to detect because of the multidrug composition of all regimens used. Furthermore certain drugs such as amikacin were used less often—limiting power—while others, such as linezolid or high dose isoniazid were used so infrequently they could not be analyzed. The small size of the “no-injectable” group may have limited power to detect an impact of the injectables, perhaps explaining the observed modest effect.

Relapse was ascertained in only 14 studies, which could result in an over-estimate of treatment success. As seen in Table S7, success rates were non-significantly lower in centers that did measure relapse. Finally, since patients included in this analysis were treated in more than 32 different centers (some studies involved multiple centers), management differed considerably in terms of use of hospitalization, response to adverse events, use of adjunctive surgery, or directly observed therapy, and the resources and adherence support offered to patients. These unmeasured inter-center differences could have resulted in bias. For example in centers that could afford to use later generation fluoro-quinolones, which are more expensive, there may have been greater resources to enhance care in other ways. However, the random effects meta-analytic approach provided some control for these center-level effects.

This analysis suggests that it would be appropriate to use at least four likely effective drugs in the initial intensive phase and at least three likely effective drugs in the continuation phase. However, it is important to underline that this analysis was restricted to cohorts of patients in whom drug susceptibility testing was routinely performed. These results may not apply when standardized regimens are used without routine drug susceptibility testing. We had to base analysis of likely effective drugs on drug susceptibility testing only, because of limited information on the specific drug regimen for many of the previously treated patients. Hence caution is warranted given the well-known limitations of drug susceptibility testing for many of the drugs used, since prior use of these drugs may increase the likelihood of resistance, even if the laboratory result indicates susceptibility.

The highest odds of success were associated with duration of the initial intensive phase of 7–8.4 mo, and with a total duration of 18–20 mo. However, particular caution should be used for the interpretation of these results. First we did not have data on duration of therapy with individual drugs, only the different phases of treatment. Second, duration of therapy was individualized for most patients, based upon severity of disease, prior therapy, drug resistance patterns, response to therapy, and timing of sputum conversion. Hence duration of treatment may have been prolonged in patients with worse disease—as suggested in Tables S7a and S7b. This could have accounted for the finding that treatment success was less with very long treatment durations, although would not be expected to lead to the finding that the odds of success increased progressively with each interval of initial intensive phase therapy up to 7–8.5 mo (in fact would be expected to have the opposite effect). Third, we had limited information on microbiologic responses, and so could not analyze the effect of duration after microbiologic conversion—a cornerstone of current recommendations (although the published evidence for the relationship between sputum conversion and long term outcomes in MDR-TB is sparse). As a result conclusions must be cautious regarding the optimal duration of therapy, which must balance the burden of prolonged therapy on patients and health systems, with the possible benefits demonstrated in this analysis. A recent report from Bangladesh in which treatment success rates were high with much shorter treatment [54], underscores the need for appropriately framed randomized trials to address this issue [60].

Despite these limitations there were a number of important strengths. A large number of centers, from many different regions of the world, contributed clinical information on a large number of individual patients, allowing a detailed and comprehensive analysis. There was substantial variation in treatment given by different centers, only partially explained by patients' characteristics. In some centers this variation reflected availability of medications, but in other centers this likely reflected individual providers' preferences. This substantial variation in treatment approach would have been much less likely in patients treated at a single center, and enhanced our ability to assess the independent effect of treatment factors on patient outcomes.

Conclusions

This individual patient data meta-analysis of 9,153 patients suggests that treatment of MDR-TB should include a later generation quinolone, and ethionamide or prothionamide. In patients who have not received second-line drugs before, the optimal number of likely effective drugs appears to be at least four in the initial intensive phase, and at least three in the continuation phase. The duration of therapy associated with highest odds of success was 7–8.5 mo for the initial intensive phase, and 25–27 mo for total duration. In view of the serious limitations of this observational data, these findings should be considered to have highlighted several important questions for future clinical trials. These questions include the role and choice of injectables, the optimal duration of an injectable and total therapy, and the potential value of later generation quinolones as well as certain group 4 and group 5 drugs. Randomized trials are urgently needed to address these questions and determine the optimal treatment regimens for MDR-TB patients.

Supporting Information

Chinese translation of the abstract by C-YC and Yuhong Liu.

(DOCX)

Click here for additional data file.

French translation of the abstract by JR.

(DOCX)

Click here for additional data file.

German translation of the abstract by CL.

(DOCX)

Click here for additional data file.

Italian translation of the abstract by GS and GBM.

(DOCX)

Click here for additional data file.

Korean translation of the abstract by T-SS, W-JK, and J-JY.

(DOCX)

Click here for additional data file.

Spanish translation of the abstract by DP, MLGG, JS-O, AP-d-L, MHV, and CP-G.

(DOCX)

Click here for additional data file.

Japanese translation of the abstract by MN and YS.

(DOCX)

Click here for additional data file.

Russian translation of the abstract by DM.

(DOC)

Click here for additional data file.

(a) Overview of settings of 32 studies included in individual patient data meta-analysis of MDR-TB supplemental tables (for on-line supplement—references for included studies are found in main text. (b) Overview of 35 studies excluded from individual patient data meta-analysis of MDR-TB.

(DOC)

Click here for additional data file.

Dosages of drugs used for MDR-TB treatment at sites of studies included in individual patient data meta-analysis on MDR-TB supplemental tables (for on-line supplement—references for included studies are found in main text.

(DOC)

Click here for additional data file.

Summary of treatment outcome definitions used in studies included in individual patient data meta-analysis.

(DOC)

Click here for additional data file.

Characteristics of patients associated with history of prior TB therapy.

(DOC)

Click here for additional data file.

Association of Individual drugs with treatment success (compared to failure/relapse)—stratified by history of previous treatment.

(DOC)

Click here for additional data file.

Secondary analyses to assess impact of covariates on duration of therapy. (A) Duration analysis also adjusted for use of four drugs. (B) Analysis restricted: excluded patients who received two or more injectables, and/or had a serious adverse event to an injectable (204 patients in two studies treated with first-line drugs also excluded). (C) Restricted analyses: use of second-line drugs only. (D) Stratified analysis: by use of later generation quinolones: 204 patients receiving first-line drugs only excluded.

(DOC)

Click here for additional data file.

Pooled outcomes—studies stratified by study level factors (event rates pooled across studies—study level random effects meta-analysis).

(DOC)

Click here for additional data file.

(a) Assessment of potential confounding of clinical characteristics and drug resistance with initial duration (only patients analyzed for success versus fail/relapse). (B) Assessment of potential confounding of clinical characteristics with total duration of therapy (only patients analyzed for success versus fail/relapse).

(DOC)

Click here for additional data file.

Summary of variance of estimates—individual drugs with treatment success.

(DOC)

Click here for additional data file.

Summary of variance of estimates—likely effective drugs with treatment success—during different phases of treatment.

(DOC)

Click here for additional data file.

Association of number of drugs used and treatment success compared to default only.

(DOC)

Click here for additional data file.

Summary of variance of estimates—duration of treatment with success versus failure/relapse—patients grouped by treatment history.

(DOC)

Click here for additional data file.

The authors also thanks the following individuals for help in the following ways: data gathering in the Philippines by Ruffy Guilatco, Glenn Balane, and Maricar Galipot; data gathering in Toronto: Maja Haslah, and Jane McNamee, facilitation of the study at the US Centers for Disease Control by Philip Lobue; comments on initial analyses and drafts by Karen Shean, Charles Daley, and Fraser Wares; assistance in data management by D. Weissman, Sidney Atwood, Tran Buu, Ed Desmond, Midori Kato-Maeda, Joanne Kirsten, and Grace Lin; secretarial and administrative assistance by Ria Choe and Sandra Ramoutar; statistical and logistic help with South African data from Piet Becker.

ReferencesWorld Health Organization (2008) Guidelines for the programmatic management of drug resistant tuberculosis - emergency update. World Health Organization247 pCamineroJA, SotgiuG, ZumlaA, MiglioriGB (2010) Best drug treatment for multidrug-resistant and extensively drug-resistant tuberculosis. Lancet10: 621629Akcakir Y (2010) Correlates of treatment outcomes of multidrug-resistant tuberculosis (MDR-TB): a systematic review and meta-analysis [PhD dissertation]. Montreal: Department of Epidemiology & Biostatistics, McGill University. 110 p.OrensteinEW, BasuS, ShahNS, AndrewsJR, FriedlandGH, et al (2009) Treatment outcomes among patients with multidrug-resistant tuberculosis:systematic review and meta-analysis. Lancet Infect Dis9: 15316119246019JohnstonJC, ShahidiNC, SadatsafaviM, FitzGeraldJM (2009) Treatment outcomes of multidrug-resistant tuberculosis: a systematic review and meta-analysis. PLoS One4: e6914 doi:10.1371/journal.pone.000691419742330StewartLA, TierneyJF (2008) Clarke M on behalf of the Cochrane Individual Patient Data Meta-analysis Methods Group (2008) Reviews of individual patient data. Higgins JPT, Green S, editors. Cochrane Handbook for Systematic reviews of Intervention. Wiley-Blackwell548558World Health Organization (2011) Guidelines for the programmatic management of drug-resistant tuberculosis, 2011 update. Geneva: World Health Organization. Available: http://whqlibdoc.who.int/publications/2011/9789241501583_eng.pdfWorld Health Organization (2006) Extensively drug-resistant tuberculosis (XDR-TB): recommendations for prevention and control. Wkly Epidemiol Rec81: 43043217096498LasersonKF, ThorpeLE, LeimaneV, WeyerK, MitnickCD, et al (2005) Speaking the same language: treatment outcome definitions for multidrug-resistant tuberculosis. Int J Tuberc Lung Dis9: 64064515971391Schabenberger O (2011) Introducing the GLIMMIX Procedure for Generalized Linear Mixed Models. Available: http://www2.sas.com/proceedings/sugi30/196-30.pdf Accessed 10 September 2010.ThompsonSG, TurnerRM, WarnDE (2001) Multilevel models for meta-analysis, and their application to absolute risk difference. Stat Methods10: 375392TurnerRM, OmarRZ, YangM, GoldsteinH, ThompsonSG (2000) A Multilevel model framework for meta-analysis of clinical trials with binary outcomes. Stat Med19: 3417343211122505HigginsJP, ThompsonSG (2002) Quantifying heterogeneity in meta-analysis. Stat Med21: 1539155812111919PinheiroJ, BatesDM (1995) Approximations to the log-likelihood function in the nonlinear mixed-effects model. J Computational Graphical Satistics4: 1235Yarandi HN (2011) Handling missing data with multiple imputation using PROC MI in SAS. Available: http://analytics.ncsu.edu/sesug/2002/ST14.pdf Accessed 1 February 2011.AvendanoM, GoldsteinRS (2000) Multidrug-resistant tuberculosis: long term follow-up of 40 non-HIV-infected patients. Can Respir J7: 38338911058206BurgosM, GonzalezLC, PazEA, GournisE, PazA, et al (2005) Treatment of multidrug-resistant tuberculosis in San Francisco: an outpatient-based approach. Clin Infect Dis40: 96897515824988IsemanMD, MadsenL, GobleM, PomerantzM (1990) Surgical intervention in the treatment of pulmonary disease caused by drug-resistant mycobacterium tuberculosis. Am Rev Respir Dis141: 6236252106811ChanED, LaurelV, StrandMJ, ChanJF, HuynhMLN, et al (2004) Treatment and outcome analysis of 205 patients with multidrug-resistant tuberculosis. Am J Respir Crit Care Med169: 1103110914742301ChiangCY, EnarsonDA, YuMC, BaiKJ, HuangRM, et al (2006) Outcome of pulmonary multidrug-resistant tuberculosis: a 6-yr follow-up study. Eur Respir J28: 98098516837502CoxHS, StobdanK, AllamuratovaS, SizaireV, TigayZN, et al (2007) Multidrug-resistant tuberculosis treatment outcomes in Karakalpakstan, Uzbekistan: treatment complexity and XDR-TB among treatment failures. PLoS One2: e1126 doi:10.1371/journal.pone.000112617987113DeRiemerK, Garcia-GarciaL, Bodadilla-del-ValleM, Palacinos-MartinezM, Martinez-GamboaA, et al (2005) Does DOTS work in populations with drug-resistant tuberculosis?Lancet365: 1239134515811457EscuderoE, PenaJM, Alvarez-SalaR, VazquezJJ, OrtegaA (2006) Multidrug-resistant tuberculosis without HIV infection: success with individualised therapy. Int J Tuberc Lung Dis10: 40941416602405GeerligsWA, van AltenaR, delangeWCM, van SoolingenD, van der WerfTS (2000) Multidrug-resistant tuberculosis: long-term treatment outcome in the Netherlands. Int J Tuberc Lung Dis4: 75876410949328GranichRM, OhP, LewisB, PorcoTC, FloodJ (2005) Multidrug resistance among persons with tuberculosis in California, 1994–2003. JAMA293: 2732273915941802BanerjeeR, AllenJ, WestenhouseJ, OhP, ElmsW, et al (2008) Extensively drug-resistant tuberculosis in California, 1993–2006. Clin Infect Dis47: 45045718616396HoltzTH, LancasterJ, LasersonKF, WellsCD, ThorpeL, et al (2006) Risk factors associated with default from multidrug-resistant tuberculosis treatment, South Africa, 1999–2001. Int J Tuberc Lung Dis10: 64965516776452KimDH, KimHJ, ParkSK, KongS, KimYS, et al (2008) Treatment outcomes and long-term survival in patients with extensively drug-resistant tuberculosis. Am J Respir Crit Care Med178: 1075108218703792KimHR, HwangSS, KimHJ, LeeSM, YooCG, et al (2007) Impact of extensive drug resistance on treatment outcomes in non-HIV-infected patients with multidrug-resistant tuberculosis. Clin Infect Dis45: 1290129517968823KwonYS, KimYH, SuhGY, ChungM, KimH, et al (2008) Treatment outcomes for HIV-uninfected patients with multidrug-resistant and extensively drug-resistant tuberculosis. Clin Infect Dis47: 49650218611154RiekstinaV, LeimaneV, HoltzTH, LeimansJ, WellsCD (2007) Treatment outcome cohort analysis in an integrated DOTS and DOTS-Plus TB program in Latvia. Int J Tuberc Lung Dis11: 58558717439686LeimaneV, RiekstinaV, HoltzTH, ZarovskaE, SkripconokaV, et al (2005) Clinical outcome of individualised treatment of multi-drugresistant tuberculosis in Latvia: a retrospective cohort study. Lancet386: 31832615664227HoltzTH, StenbergM, KammererS, LasersonKF, RiekstinaV, et al (2006) Time to sputum culture conversion in multidrug-resistant tuberculosis: predictors and relationship to treatment outcome. Ann Intern Med144: 65065916670134LockmanS, KruunerA, BinkinN, LevinaK, WangYC, et al (2001) Clinical outcomes of Estonian patients with primary multidrug-resistant versus drug-susceptible tuberculosis. Clin Infect Dis32: 3738011170944MasjediMR, TabarsiP, ChitsazP, BaghaeiP, MiraeidiM, et al (2008) Outcome of treatment of MDR-TB patients with standardised regimens, Iran, 2002–2006. Int J Tuberc Lung Dis12: 75075518544199MiglioriGB, EspinalM, DanilovaID, PungaVV, GrzemskaM, et al (2002) Frequency of recurrence among MDR-TB cases ‘successfully’ treated with standardised short-course chemotherapy. Int J Tuberc Lung Dis6: 85886412365571MiglioriGB, BesozziG, GirardiE, KliimanK, LangeC, et al (2007) Clinical and operational value of the extensively drug-resistant tuberculosis definition. Eur Respir J30: 62362617690121MitnickC, BayonaJ, PalaciosE, ShinSS, FurinJ, et al (2003) Community-based therapy for multidrug-resistant tuberculosis in Lima, Peru. N Engl J Med348: 11912812519922MitnickC, ShinSS, SeungKJ, RichML, AtwoodSS, et al (2010) Comprehensive treatment of extensively drug-resistant tuberculosis. N Engl J Med359: 56357418687637MunsiffSS, AhujaSD, LiJ, DriverCR (2006) Public-private collaboration for multidrug-resistant tuberculosis control in New York city. Int J Tuberc Lung Dis10: 63964816776451LiJ, BurzynskiJN, LeeYA, BergD, DriverCR, et al (2004) Use of therapeutic drug monitoring for multidrug-resistant tuberculosis patients. Chest126: 1770177615596672NaritaM, AlonsoP, LazardoM, HollenderE, PitchenikA, et al (2001) Treatment experience of multidrug-resistant tuberculosis in Florida, 1994–1997. Chest120: 34334811502627O'RiordanP, SchwabU, LoganS, CookeG, WilkinsonRJ, et al (2008) Rapid molecular detection of rifampicin resistance facilitates early diagnosis and treatment of multidrug-resistant tuberculosis: case control study. PLoS One3: e3173 doi:10.1371/journal.pone.000317318779863PalmeroDJ, AmbroggiM, BreaA, De LucasM, FulgenziA, et al (2004) Treatment and follow-up of HIV-negative multidrug-resistant tuberculosis patients in an infectious diseases reference hospital, Buenos Aires, Argentina. Int J Tuberc Lung Dis8: 77878415182150ParkSK, LeeWC, LeeDH, MitnickCD, HanL, et al (2004) Self-administered, standardized regimens for multidrug-resistant tuberculosis in South Korea. Int J Tuberc Lung Dis8: 36136815139476Pérez-GuzmánC, VargasMH, Martinez-RossierLA, Torres-CruzA, Villarreal-VelardeH (2002) Results of a 12-month regimen for drug-resistant pulmonary tuberculosis. Int J Tuberc Lung Dis6: 1102110912546119QuyHT, CobelensFGJ, LanNTN, BuuTN, LambregtsCSB, et al (2006) Treatment outcomes by drug resistance and HIV status among tuberculosis patients in Ho Chi Minh City, Vietnam. Int J Tuberc Lung Dis10: 455116466036SchaafHS, SheanK, DonaldPR (2003) Culture confirmed multidrug resistant tuberculosis: diagnostic delay, clinical features, and outcome. Arch Dis Child88: 1106111114670781ShinSS, PasechnikovAD, GelmanovaIY, PeremitinGG, StrelisAK, et al (2006) Treatment outcomes in an integrated civilian and prison MDR-TB treatment program in Russia. Int J Tuberc Lung Dis10: 40240816602404ShiraishiY, NakalimaY, KatsuragiN, KuraiM, TakahashiN (2004) Resectional surgery combined with chemotherapy remains the treatment of choice for multidrug-resistant tuberculosis. J Thorac Cardiovasc Surg128: 52352815457152TupasiTE, QuelapioMID, OrillazaRB, AlcantaraC, MiraNRC, et al (2003) DOTS-Plus for multidrug-resistant tuberculosis in the Philippines: global assistance urgently needed. Tuberculosis83: 525812758189TupasiTE, GuptaR, QuelapioMID, OrillazaRB, MiraNR, et al (2006) Feasibility and cost-effectiveness of treating multidrug resistant tuberculosis: a cohort study in the Philippines. PLoS Med3: e352 doi:10.1371/journal.pmed.003035216968123UffrediML, Truffot-PernotC, DautzenbergB, RenardM, JarlierV, et al (2007) An intervention programme for the management of multidrug-resistant tuberculosis in France. Intl J Antimicrob Agents29: 434439Van DeunA, SalimMAH, DasAPK, BastianI, PortaelsF (2004) Results of a standardised regimen for multidrug-resistant tuberculosis in Bangladesh. Int J Tuberc Lung Dis8: 56056715137531Van DeunA, MaugAKJ, SalimMAH, DasPK, SarkerMR, et al (2010) Short, highly effective and inexpensive standardized treatment of multidrug-resistant tuberculosis. Am J Respir Crit Care Med182: 68469220442432YewWW, ChanCK, LeungCC, ChauCH, TamCM, et al (2003) Comparative roles of levofloxacin and ofloxacin in the treatment of multidrug-resistant tuberculosis. Chest124: 1476148114555582YewWW, ChanCK, ChauCH, TamCM, LeungCC, et al (2000) Outcomes of patients with multidrug-resistant pulmonary tuberculosis treated with ofloxacin/levofloxacin-containing regimens. Chest117: 74475110713001EspinalMA, KimSJ, SuarezPG, KamKM, KhomenkoAG, et al (2000) Standard short-course chemotherapy for drug-resistance tuberculosis- treatment outcomes in 6 countries. JAMA283: 2537254510815117PsatyBM, SiscovickDS (2010) Minimizing bias due to confounding by indication in comparative effectiveness research the importance of restriction. JAMA304: 89789820736474MitnickCD, CastroKG, HarringtonM, SacksLV, BurmanW (2007) Randomized trials to optimize treatment of multidrug-resistant tuberculosis. PLoS Med4: e292 doi:10.1371/journal.pmed.004029217988168AbbreviationsAFB

acid fast bacilli

aOR

adjusted odds ratio

MDR-TB

multidrug resistant tuberculosis