|Home | About | Journals | Submit | Contact Us | Français|
New diagnostic tools are urgently needed to interrupt the transmission of tuberculosis and multidrug-resistant tuberculosis. Rapid, sensitive detection of tuberculosis and multidrug-resistant tuberculosis in sputum has been demonstrated in proof-of-principle studies of the microscopic-observation drug-susceptibility (MODS) assay, in which broth cultures are examined microscopically to detect characteristic growth.
In an operational setting in Peru, we investigated the performance of the MODS assay for culture and drug-susceptibility testing in three target groups: unselected patients with suspected tuberculosis, prescreened patients at high risk for tuberculosis or multidrug-resistant tuberculosis, and unselected hospitalized patients infected with the human immunodeficiency virus. We compared the MODS assay head-to-head with two reference methods: automated mycobacterial culture and culture on Löwenstein–Jensen medium with the proportion method.
Of 3760 sputum samples, 401 (10.7%) yielded cultures positive for Mycobacterium tuberculosis. Sensitivity of detection was 97.8% for MODS culture, 89.0% for automated mycobacterial culture, and 84.0% for Löwenstein–Jensen culture (P<0.001); the median time to culture positivity was 7 days, 13 days, and 26 days, respectively (P<0.001), and the median time to the results of susceptibility tests was 7 days, 22 days, and 68 days, respectively. The incremental benefit of a second MODS culture was minimal, particularly in patients at high risk for tuberculosis or multidrug-resistant tuberculosis. Agreement between MODS and the reference standard for susceptibility was 100% for rifampin, 97% for isoniazid, 99% for rifampin and isoniazid (combined results for multidrug resistance), 95% for ethambutol, and 92% for streptomycin (kappa values, 1.0, 0.89, 0.93, 0.71, and 0.72, respectively).
A single MODS culture of a sputum sample offers more rapid and sensitive detection of tuberculosis and multidrug-resistant tuberculosis than the existing gold-standard methods used.
Every year, 1.7 million people die of tuberculosis, a curable disease.1 The poor are disproportionately affected, and tuberculosis further impoverishes individual people and societies. Goal 6 of the Millennium Development Goals of the United Nations includes the halting and reversal of the rising incidence of tuberculosis, and the Stop TB Partnership aims to halve the prevalence of tuberculosis and resulting deaths by 2015.2 Existing control strategies miss important opportunities to interrupt transmission. Improved tuberculosis detection and early identification of multidrug-resistant tuberculosis are key gaps.
Sputum-smear–based diagnosis under the Direct Observation of Therapy (Short Course) (DOTS) strategy of the World Health Organization for global tuberculosis control misses half of incident cases at first presentation. Transmission continues until cases are detected with more advanced (smear-positive) disease and are correctly treated. Multidrug-resistant tuberculosis increases morbidity and mortality and, through treatment failure,3 facilitates continuing transmission from patients who (like their health care providers) wrongly believe they are being cured. The use of treatment failure to prompt drug-susceptibility testing relies on the same illness-threshold effect as waiting for smears to become positive in patients with negative smears.4 Both scenarios could be addressed by appropriate new diagnostics.5
The microscopic-observation drug-susceptibility (MODS) assay for the detection of tuberculosis and multidrug-resistant tuberculosis, directly from sputum, relies on three principles: first, that Mycobacterium tuberculosis grows faster in liquid medium than in solid medium; second, that characteristic cord formation can be visualized microscopically in liquid medium at an early stage; and third, that the incorporation of drugs permits rapid and direct drug-susceptibility testing con-comitantly with the detection of bacterial growth.
In proof-of-principle studies, the MODS assay distinguished patients with tuberculosis from healthy controls.6,7 In accordance with recommendations,8 we undertook an operational evaluation of the MODS assay in order to answer two questions: How well does it distinguish between patients with and those without active tuberculosis among those with suspected tuberculosis? Among patients with active tuberculosis, how well does it distinguish drug-resistant disease from drug-sensitive disease?
Given the challenges of the evaluation of techniques for the diagnosis of tuberculosis,9–13 our study was conducted in accordance with the emerging consensus about the design and reporting of diagnostic-test evaluations14–18 and the minimum standards required for diagnostic trials of tuberculosis.8,19,20
The study was conducted in Lima, Peru, from April 2003 through July 2004 in three target groups, with consecutive recruitment. The first group consisted of otherwise unselected patients who presented with suspected tuberculosis to the National TB Programme at 10 government clinics in north Lima. The second consisted of patients who presented with suspected tuberculosis to the National TB Programme at five government clinics in east Lima and who were at high risk for tuberculosis or multidrug-resistant tuberculosis. Inclusion in this group required the presence of one or more constitutional symptoms (fever, weight loss, night sweats, hemoptysis) or one risk factor for tuberculosis or multidrug-resistant tuberculosis (prior treatment for tuberculosis, known contact with a patient with tuberculosis, infection with human immunodeficiency virus [HIV], employment as a health care or prison worker, hospitalization during the previous year, or any previous incarceration). The third group consisted of otherwise unselected hospitalized patients with HIV infection at two Lima hospitals, regardless of the diagnosis on admission. Exclusion criteria for all groups were an age under 18 years or an inability or unwillingness to give written informed consent. Study protocol and consent forms were approved by the institutional review boards of Universidad Peruana Cayetano Heredia, Asociación Benéfica PRISMA, Dirección de Salud–III Lima Norte and Dirección de Salud–IV Lima Este (regional Ministry of Health), Hospital Nacional Hipolito Unanue, Hospital Nacional General Arzobispo Loayza, Johns Hopkins Bloomberg School of Public Health, and Imperial College London.
Patients with suspected tuberculosis and prescreened patients at high risk for tuberculosis or multidrug-resistant tuberculosis submitted two samples of sputum to the National TB Programme for routine Ziehl–Neelsen staining and consented to their subsequent use in the study (at Universidad Peruana Cayetano Heredia). Among the hospitalized patients with HIV infection, two samples per patient were collected exclusively for the study, of which 20 were from gastric washing in 15 patients who were unable to provide adequate sputum samples.
Patients with suspected tuberculosis and prescreened patients at high risk for tuberculosis or multidrug-resistant tuberculosis who were eligible for the study were not recruited if they presented at the clinic outside of normal working hours (>90% of excluded patients), declined to participate (<5%), or if the clinic staff were unavailable (<5%). Numbers of eligible patients were derived from data extracted from National TB Programme log entries during the recruitment period; the numbers may have been overestimated, since follow-up patients (who were ineligible) were not always clearly identified. Of the 25 hospitalized patients with HIV infection who were eligible but not recruited, more than 90% were incapable of providing informed consent, owing to their clinical condition, or were receiving ongoing treatment for tuberculosis. Most, but not all, patients submitted two sputum samples.
Sputum samples were decontaminated according to the sodium hydroxide–N-acetyl-L-cysteine method.21 An aliquot was used for microscopical examination of auramine-stained sputum smears, and the remainder was used for parallel Löwenstein–Jensen culture, automated mycobacterial culture, and MODS culture (see Fig. I in the Supplementary Appendix, available with the full text of this article at www.nejm.org). Löwenstein–Jensen culture and automated mycobacterial culture with the use of the MBBacT system (bioMérieux) were selected because they are reference methods commonly used in developing and industrialized countries, respectively. After inoculation of 250 μl of decontaminant, Löwenstein–Jensen slants were incubated at 37°C and examined twice weekly from day 7 through day 60.21 MBBacT bottles were inoculated with 500 μl of decontaminant, and cultures were monitored continuously for 42 days according to the recommendations of the manufacturer.
The MODS assay was performed as described previously.6,7 Briefly, broth cultures were prepared in 24-well tissue-culture plates (Becton Dickinson), each containing 720 μl of decontaminant, Middlebrook 7H9 broth (Becton Dickinson), oxalic acid, albumin, dextrose, and catalase (OADC) (Becton Dickinson), and polymyxin, amphotericin B, nalidixic acid, trimethoprim, and azlocillin (PANTA) (Becton Dickinson). For each sample, 12 wells were used: in 4 control wells, no drug was added, and each of the remaining 8 wells contained one of four drugs at one of two concentrations tested. The cultures were examined under an inverted light microscope at a magnification of 40× every day (except Saturday and Sunday) from day 4 to day 15, on alternate days from day 16 to day 25, and twice weekly from day 26 to day 40. To minimize cross-contamination and occupational exposure, plates were permanently sealed inside plastic ziplock bags after inoculation and were subsequently examined within the bag. Positive cultures were identified by cord formation, characteristic of M. tuberculosis growth, in liquid medium in drug-free control wells, as described previously.6,7,22 Nontuberculous mycobacteria were recognized by their lack of cording or, for M. chelonae (which is the only nontuberculous mycobacteria that does form cords), by rapid overgrowth by day 5. Fungal or bacterial contamination was recognized by rapid overgrowth and clouding in wells.
If contamination was detected, the original sample was cultured again after being decontaminated once more. Spacer oligonucleotide typing (spoligotyping), polymerase chain reaction with multiple primers,23 or both were applied to all isolates from each of the three types of cultures in order to confirm the presence of M. tuberculosis.
Direct drug-susceptibility testing was performed with the use of the MODS assay, as previously described.6,7 Growth in drug-free control wells but not in drug-containing wells indicated susceptibility. The drug concentrations used were as follows: isoniazid, 0.1 and 0.4 μg per milliliter; rifampin, 1 and 2 μg per milliliter; ethambutol, 2.5 and 5.0 μg per milliliter; and streptomycin, 2 and 6 μg per milliliter. Drug-sensitive control strains were tested daily. Indirect drug-susceptibility testing was performed with the use of the proportion method21 for isolates from Löwenstein–Jensen culture (by an external laboratory) and with the automated MBBacT system24–29 for isolates from the automated mycobacterial culture. For purposes of discrepant analysis, the microplate alamar blue assay30–35 was performed in parallel, both for all Löwenstein–Jensen isolates and for isolates from drug-free control wells in the MODS assay. All procedures were performed by six staff members of the mycobacteriology laboratory who were unaware of the results of the other tests.
Data were analyzed with the use of Stata 7 software, with the sample as the unit of analysis, to reflect the operational performance of a routine service laboratory. The Wilcoxon signed-rank test was used to compare the times to each end point among the three methods. A P value of less than 0.05 was used to indicate statistical significance. The concordance of susceptibility results was determined with the use of the sensitivity, specificity, and positive and negative predictive values for the detection of resistance (with 95% confidence intervals [CIs]), as well as with kappa values.
To address the inherent difficulty of evaluating a test that is more sensitive than the reference tests, and to minimize incorporation bias (the use of results from the investigational test as part of the reference result), we previously undertook a comprehensive microbiologic, molecular, and epidemiologic investigation of the discordant cultures.36 We identified 17 cross-contaminated cultures (12 MODS cultures, 4 automated mycobacterial cultures, and 1 Löwenstein–Jensen culture) from 14 samples. For sensitivity and specificity of detection and predictive-value calculations for each of the three methods, a positive reference result was defined as a positive culture according to at least one method for which cross-contamination had been conclusively ruled out.36 A negative reference result was defined as any sample in which all three culture methods yielded negative results or in which two were negative and the third indeterminate, owing to repeated bacterial or fungal overgrowth, or a sample for which cross-contamination was demonstrated to be the only cause of the positive culture.36
Thus, the 17 false positive cultures were defined as positive in calculations of performance characteristics for the relevant methods, but the reference result for the 14 samples was determined to be negative. McNemar’s χ2 test was used to compare the sensitivities of detection of the three methods.
Concordant results from automated mycobacterial culture and the proportion method were regarded as the reference result for drug susceptibility. Discordant results from automated mycobacterial culture and the proportion method were resolved by means of discrepant analysis, with the use of the two parallel results from the microplate alamar blue assay. If both results from this assay agreed, that result was used as the reference result; if not, the strain was designated indeterminate (see Table I in the Supplementary Appendix).
Of the 3760 sputum samples collected, 401 (10.7%) were positive for M. tuberculosis cultures, 3356 were negative, and 3 were indeterminate (and were removed from analysis), because all three types of cultures were repeatedly contaminated by bacterial overgrowth. MODS culture had a greater overall sensitivity of detection than either automated mycobacterial culture or Löwenstein–Jensen culture (97.8%, 89.0%, and 84.0%, respectively; P<0.001); this difference was maintained in all groups. The overall specificity of detection was 99.6% for MODS culture, 99.9% for automated mycobacterial culture, and 100.0% for Löwenstein–Jensen culture. Predictive values and data according to group are shown in Table 1, and in Figure II of the Supplementary Appendix.
The incremental benefit of a second smear for acid-fast bacilli and a second sputum culture is shown in Figure 2. A second MODS culture detected an additional 8.2% of cases among patients with suspected tuberculosis but offered no added value among prescreened patients at high risk for tuberculosis or multidrug-resistant tuberculosis.
Of the 401 sputum samples positive for M. tuberculosis, 325 were culture-positive according to all three methods and were thus included in the head-to-head analysis of time to culture positivity (Fig. 3). The median time to culture positivity was significantly shorter for MODS than for the automated mycobacterial or Löwenstein–Jensen cultures (7 days [interquartile range, 6 to 8] vs. 13 days [interquartile range, 10 to 16] and 26 days [interquartile range, 21 to 33], respectively; P<0.001). Smear status had a clinically unimportant, though significant, effect on time to culture positivity in MODS culture (median, 6 days for a smear-positive sample vs. 7 days for a smear-negative sample; P<0.001). Though samples that were culture-negative (69) or contaminated (7) according to at least one method were excluded from this analysis, results were unchanged by their inclusion (data not shown).
The median time from sample collection to sample processing was 3 days, and most but not all samples were refrigerated en route. At least one culture per sample was contaminated in 739 of 3760 samples (20.0%), though in only 63 samples (1.7%) were all cultures by the three methods contaminated. The proportion of initially contaminated samples was 8.1% (95% CI, 7.2 to 9.0) for MODS culture, 4.4% (95% CI, 3.8 to 5.3) for automated mycobacterial culture, and 14.2% (95% CI, 13.1 to 15.3) for Löwenstein–Jensen culture. However, ultimately contaminated (indeterminate) cultures were less frequent in MODS culture (6 cultures, 0.2%) than in either automated mycobacterial culture (11 cultures, 0.3%; P=0.01) or Löwenstein–Jensen culture (55 cultures, 1.5%; P<0.001). The median time from initial processing of samples to the results of culture testing for initially contaminated cultures was shorter for MODS culture (24 days; 95% CI, 19 to 28) than for automated mycobacterial culture (32 days; 95% CI, 25 to 39; P=0.03) or Löwenstein–Jensen culture (50 days; 95% CI, 44 to 56; P<0.001). Positive MODS cultures that were contaminated but able to be evaluated accounted for less than 2.5% of all positive cultures for the assay.
Valid drug-susceptibility testing in the MODS assay depends on the observed growth of M. tuberculosis in all four control wells, as was the case for 349 of 392 positive MODS cultures (89.0%). Of the 43 samples for which drug-susceptibility testing in the MODS assay was not possible, 28 (65.1%) were culture-negative according to automated mycobacterial culture and Löwenstein–Jensen culture. Resistance to rifampin was detected in 10.7% of all samples; to isoniazid, in 19.5%; to rifampin and isoniazid (combined to test for multidrug resistance), in 10.4%; to ethambutol, in 10.1%; and to streptomycin, in 21.4% (Table 2). The proportion of samples for which susceptibility results agreed between MODS culture and the gold-standard methods was 100% for rifampin, 96.7% for isoniazid, 98.8% for rifampin and isoniazid, 95.4% for ethambutol, and 91.7% for streptomycin. Over-all median times from initial sample processing to the results of drug-susceptibility testing were 7 days for MODS culture, 22 days for automated mycobacterial culture, and 68 days for Löwenstein–Jensen culture.
This operational study extends and provides support for the findings of earlier proof-of-principle studies6,7,22–37 and demonstrates that the MODS assay outperforms the gold-standard reference methods of developing and industrialized countries. For all three study groups, the MODS assay detected M. tuberculosis in sputum with greater sensitivity and speed and reliably identified multidrug-resistant tuberculosis strains in less time than did Löwenstein–Jensen or automated mycobacterial cultures. These data indicate that the MODS assay could be considered for use in appropriate settings.
Our study was designed to address conventional pitfalls.8,15,18,20,38 Specifically, it was performed in an appropriately broad group of patients with or without disease and in pertinent patient groups (without selection bias); all tests were performed in all patients (preventing verification bias); and all results were interpreted by staff members who were unaware of the other test results, using appropriate gold-standard reference methods for comparison.
The robustness of our study derives from its operational, real-world design. Meticulous resolution of discordant results is essential when an investigational diagnostic method is more sensitive than existing reference standards. Use of two established reference methods for comparison and two samples per patient facilitated the rigorous definition of true positive results, addressing the problem of incorporation bias. The high specificity and infrequent cross-contamination in the MODS assay36 relate to the containment of the plates in ziplock bags and the absence of manipulation after inoculation, which also improve biologic security.
The greater sensitivity and speed of detection in MODS culture than in the gold standards were predicted on the basis of previous studies.6,7,37 The increased sensitivity of liquid medium has long been known, and a discerning human eye can scrutinize cultures better than can automated systems with their use of necessarily rigid cutoff values. It is simpler to recognize the characteristic cord formation than to read a malarial smear; within 1 week, students training in our laboratory can comfortably and accurately read one well per minute, considerably faster than the time it takes to read a smear for acid-fast bacilli. Training in the MODS assay can be completed in less than 2 weeks (similar to training in Löwenstein–Jensen and automated mycobacterial cultures; training in drug-susceptibility testing with the proportion method takes several months). Beyond standard laboratory equipment, automated mycobacterial culture requires computer-linked automated culture incubators, whereas MODS culture requires only an inverted light microscope. As purchased by us, the cost of $2 per sample for MODS culture compares favorably with the $6 cost per sample for Löwenstein–Jensen culture and the proportion method and the cost of $52 per sample for automated mycobacterial culture; however, labor costs may be higher for MODS culture.
Increased sensitivity carries the risk of increased bacterial overgrowth (for MODS culture and automated mycobacterial culture), though even after repeated decontamination, the sensitivity and specificity of MODS culture were unaffected. High speed, sensitivity, and specificity, and the requirement for only one culture, all enhance tuberculosis rule-out procedures and potentially simplify tuberculosis-screening algorithms for use in patients with HIV infection before the initiation of prophylactic treatment with isoniazid. If a MODS culture is negative on day 15, there is a 99.7% chance that the sample is truly culture-negative. Thus, we believe that a negative MODS culture can be discarded after 3 weeks.
In settings with a high tuberculosis burden, the only susceptibility data that are likely to effect a change in therapy at the programmatic level are those for the detection of multidrug-resistant tuberculosis, for which the performance of the MODS assay is highly reliable and rapid (median time to the results of susceptibility testing, 7 days), providing clinically important information in a meaningful time frame. Although direct drug-susceptibility testing is conventionally viewed with suspicion — indirect testing of cultured strains is preferred — our data refute that view for rifampin and isoniazid in the MODS assay. However, susceptibility testing for M. tuberculosis is complex, and concordance among even regional laboratories performing gold-standard testing is particularly variable for ethambutol and streptomycin.39 Our findings for these two drugs agreed with previous data for the MODS assay,7 demonstrating insufficient concordance of the assay (at least in its current format) with gold standards to recommend usage.
Our study defines strengths and redundancies in the first-generation MODS assay and should enable the development of a streamlined, clinically useful method. The use of fewer wells per sample than we used — two wells with no drug (to ensure high specificity), one with rifampin (1 μg per milliliter), and one with isoniazid (0.4 μg per milliliter) — reduces costs by 40% but does not affect performance. The MODS assay is “laboratory freeware,” not a commercial product or a kit. Any laboratory that is adequately biologically secured, has an incubator and a centrifuge, and is capable of microscopy can safely perform MODS culture. All ingredients are available from major laboratory suppliers.
Downstream effects on patient care are the litmus test of the utility of a new method, and the added value will therefore depend on context and strategy for implementation. In countries where smear-negative tuberculosis is frequently diagnosed and treated empirically, the incremental benefit of MODS culture on case detection, as compared with the smear alone, would be less than that in Peru, where only 21% of treated cases are smear-negative and where MODS culture has recently been incorporated into Ministry of Health guidelines (www.minsa.gob.pe/normaslegales/2006/RM383-2006.pdf). However, the high specificity rate would save patients and society money by minimizing overtreatment, and the early detection and treatment of multidrug-resistant tuberculosis would interrupt transmission. Equity in the access to high-performance techniques to diagnose tuberculosis thus benefits both individual health40,41 and public health. Despite the low cost per sample, in many resource-limited settings with a high tuberculosis burden, testing by the MODS assay of all patients with suspected tuberculosis (<5% of whom have culture-positive disease in Peru) would be a challenge financially and operationally. In our targeted, high-risk groups, only one MODS culture (collected in one visit) is required to achieve culture-positive rates of 20%, a good return on the investment. Programmatic studies are now needed to determine the optimal implementation strategy to maximize the effect and cost-effectiveness of this tool.
The MODS assay addresses two key gaps in resource-limited settings with a high tuberculosis burden: rapid, accurate detection of M. tuberculosis and simultaneous identification of multidrug-resistant tuberculosis. The use of culture-based diagnostic techniques for case detection may not be the future as envisaged by the International Union against Tuberculosis and Lung Disease42; promotion of such a strategy may conflict with the view that the scale-up of coverage and improvement of smear microscopy is currently a more important priority. However, we believe the MODS assay could now be implemented in settings in which smear microscopy is being optimally used and the augmentation of case detection is feasible and desirable.
Presented in part at the 35th UNION World Conference on Lung Health, Paris, October 28–November 1, 2004 (abstract nos. PS-456-493 and PC-456-496).
Supported by a grant from the Wellcome Trust (064672). Dr. Gilman and Ms. Caviedes were supported by a National Institutes of Health–Fogarty–Global Research Training Grant (TW006581).
Drs. Moore and Evans report having received grant support from Sequella for clinical trials of another tuberculosis diagnostic test. Dr. Gilman reports having received consulting fees from Sequella. Dr. Vivar reports having received grant support from Merck, Chiron, and Cerexa. No other potential conflict of interest relevant to this article was reported.
We are indebted to all participants in this study, for almost all of whom this was their first experience with clinical research; to the numerous medical and laboratory staff members at the community-clinic and hospital study sites who ensured that the protocol ran smoothly, particularly Yuri García, Adolfo OrellanaMarin, Raul Miranda Arrostigue (Centro de Salud [CS] Carlos Cueto Fernandini), Guillermo Vera Mallqui (Centro Base de Salud Los Olivos), Luis Rivera Pérez (CS Infantas), Walter Ramos Maguiña (CS Villa Norte), Luz Vásquez Chávez (CS Primavera), Félix Pari Loayza (CS Juan Pablo II), Ruth Flores Escobar (Puesto de Salud [PS] Los Olivos de Pro), Jesús Castillo Diaz (CS-Comites Locales de Administracíon Compartida [CS-CLAS] Laura Caller Iberico), Milciades Reátegui Sanchez (CS-CLAS San Martín de Porres), Alicia Vigo Alegria (PS Enrique Milla Ochoa); and to laboratory, support, and field staff members at Universidad Peruana Cayetano Heredia and Asociación Benéfica PRISMA, particularly Paula Maguiña, Fanny Garcia, Eleana Sanchez, Yrma Chuquiruna, Rosmery Gutierrez, Sonia Lopez, Christian Solis, Indira Villaverde, Pilar Navarro, and Natalie Bowman.