https://orcid.org/0000-0002-2466-2442
Figures
Abstract
Background
Globally, over one-third of pulmonary tuberculosis (TB) disease diagnoses are made based on clinical criteria after a negative bacteriological test result. There is limited information on the factors that determine clinicians’ decisions to initiate TB treatment when initial bacteriological test results are negative.
Methods and findings
We performed a systematic review and individual patient data meta-analysis using studies conducted between January 2010 and December 2022 (PROSPERO: CRD42022287613). We included trials or cohort studies that enrolled individuals evaluated for TB in routine settings. In these studies, participants were evaluated based on clinical examination and routinely used diagnostics and were followed for ≥1 week after the initial test result. We used hierarchical Bayesian logistic regression to identify factors associated with treatment initiation following a negative result on an initial bacteriological test (e.g., sputum smear microscopy (SSM), Xpert MTB/RIF).
Multiple factors were positively associated with treatment initiation: male sex [adjusted odds ratio (aOR) 1.61 (1.31, 1.95)], history of prior TB [aOR 1.36 (1.06, 1.73)], reported cough [aOR 4.62 (3.42, 6.27)], reported night sweats [aOR 1.50 (1.21, 1.90)], and having HIV infection but not on ART [aOR 1.68 (1.23, 2.32)]. Treatment initiation was substantially less likely for individuals testing negative with Xpert [aOR 0.77 (0.62, 0.96)] compared to smear microscopy and declined in more recent years. We were not able assess why clinicians made treatment decisions, as these data were not available.
Author summary
Why was this study done?
- Tuberculosis (TB) remains one of the leading causes of infectious disease death worldwide.
- Despite advancements in TB diagnostics, many diagnoses are still based on clinical judgment rather than bacteriological evidence.
- Understanding why clinicians decide to initiate TB treatment despite negative bacteriological test results can improve diagnostic accuracy and treatment outcomes.
What did the researchers do and find?
- We conducted a systematic review and meta-analysis of individual patient data from studies conducted between January 2010 and December 2022, where individuals were evaluated for TB.
- Key factors associated with initiating TB treatment after a negative bacteriological test included male sex, history of prior TB, reported cough and night sweats, and having HIV infection but not on antiretroviral therapy. Clinicians were less likely to initiate treatment if the initial test was a PCR-based diagnostic like Xpert MTB-RIF (as compared to sputum smear microscopy (SSM)).
What do these findings mean?
- The study identifies several factors that influence clinicians’ decisions to treat for TB despite negative bacteriological test results.
- These findings can help refine TB diagnostic and treatment protocols, improving patient outcomes and enhancing public health strategies.
- More evidence is needed on clinicians’ decision-making processes, which we did not assess in this study.
Citation: Kim S, Can MH, Agizew TB, Auld AF, Balcells ME, Bjerrum S, et al. (2025) Factors associated with tuberculosis treatment initiation among bacteriologically negative individuals evaluated for tuberculosis: An individual patient data meta-analysis. PLoS Med 22(1): e1004502. https://doi.org/10.1371/journal.pmed.1004502
Academic Editor: Aaloke Mody, Washington University School of Medicine, UNITED STATES OF AMERICA
Received: April 17, 2024; Accepted: November 20, 2024; Published: January 13, 2025
This is an open access article, free of all copyright, and may be freely reproduced, distributed, transmitted, modified, built upon, or otherwise used by anyone for any lawful purpose. The work is made available under the Creative Commons CC0 public domain dedication.
Data Availability: The data included in this study are subject to restrictions imposed by the Institutional Review Board (IRB) and require data use agreements for access; therefore, they are not publicly available. For access to the dataset, please contact the IRB of the Harvard T.H. Chan School of Public Health at irb@hsph.harvard.edu and refer to the respective authors listed in Table K in S1 Supplement.
Funding: This work was supported by the National Institute of Allergy And Infectious Diseases of the National Institutes of Health (Award Number U01AI152084 to SD). Other authors have no funding to declare for this study. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: AFL has received research grant support to their institution unrelated to this work from Cepheid, Gilead, GSK, Merck and ViiV. She has received consulting fees from Vir. She has received pharmaceutical and laboratory donations as research support to her institution from Cepheid, Hologic and Mayne Pharma. NAM receives research funding from NIH, CDC, CSTE, and the European Commission, and is a consultant with the Global Fund to Fight AIDS, TB, and Malaria. Other authors have no competing interests to declare.
Abbreviations: aOR, adjusted odds ratio; ART, antiretroviral therapy; CI, credible interval; IPD, individual patient data; OR, odds ratio; PCR, polymerase chain reaction; RDT, rapid diagnostic test; SSM, sputum smear microscopy; TB, tuberculosis; WHO, World Health Organization
Introduction
Tuberculosis (TB) remains a leading cause of infectious disease death worldwide [1], and a key strategy for accelerating TB elimination is to improve capacity for rapid and accurate diagnosis in high-burden countries [2]. Traditional TB diagnostics have major limitations, with sputum smear microscopy (SSM) failing to identify a substantial fraction of TB cases, and sputum culture requiring up to 8 weeks to return results. However, since 2010 the World Health Organization (WHO) has endorsed several new PCR (polymerase chain reaction)-based diagnostics with the potential to improve TB case detection, including the Xpert MTB/RIF (Xpert), Xpert MTB/RIF Ultra (Xpert Ultra), Truenat MTB, Truenat MTB Plus, and Truenat MTB-RIF Dx assays [3]. These tests combine rapid turn-around time and high sensitivity, enabling timely and accurate TB diagnosis [3,4].
Despite the potential of these new diagnostics, several studies have found limited effects on TB diagnoses and mortality following their introduction [5–13]. Evidence from programmatic settings suggests that clinical diagnosis (diagnosis based on clinical criteria alone, made when a bacteriological test result is unavailable or is negative) may partially explain this finding [14–17]. In many countries, clinical diagnosis represents a substantial fraction of notified TB cases despite the widespread adoption of Xpert, and in 2022 clinical diagnoses represented 38% of total global notifications for pulmonary TB [1]. If some of the individuals testing false-negative on an initial bacteriological test are subsequently treated based on clinical criteria, this may reduce the incremental impact achieved by adopting a more sensitive diagnostic. However, the widespread use of clinical diagnosis may also increase the number of individuals incorrectly treated for TB and overlook cases of drug-resistant TB, as studies of the performance of clinical diagnosis suggest that the specificity of clinical algorithms can be low [18–20]. For certain types of tuberculosis, such as extrapulmonary and pediatric TB, clinical diagnosis may be the primary diagnostic approach.
As higher sensitivity diagnostics become more commonly used, it is useful to understand current practices around clinical diagnosis, and the factors that affect clinical decision-making when diagnostic test results are negative. These clinical decisions will affect the overall sensitivity and specificity of TB diagnostic algorithms, as well as determining the incremental health impact of new diagnostics. In this study, we conducted a systematic review of studies reporting diagnostic decisions and treatment initiation following a negative test result received as part of routine TB diagnosis. Using these data, we conducted an individual patient data (IPD) meta-analysis to identify the factors that affect clinicians’ decisions to treat for pulmonary TB despite a negative test result.
Methods
The target population for this study was individuals evaluated for pulmonary TB disease in routine clinical settings, who had received a negative result on an initial diagnostic test (e.g., smear microscopy, Xpert MTB/RIF). We conducted a systematic review to identify data sets describing the individual characteristics as well as the outcome of TB diagnosis (i.e., whether or not TB treatment was initiated) for individuals in this target population. The protocol was registered with PROSPERO: CRD42022287613 [21] and approved by the Institutional Review Board of the Harvard School of Public Health (IRB21-1488).
Search strategy and selection criteria
Studies were identified by searching Medline/PubMed (National Library of Medicine, NCBI) and Embase (Elsevier, embase.com). Controlled subject vocabulary terms (i.e., MeSH, Emtree) were included when available and appropriate. The search strategies were designed and carried out by a health sciences librarian (CM). The publication date was limited to 2010 to 2022 in order to restrict the analysis to the period over which new TB diagnostics were being introduced. The exact search strategies are provided in Text A in S1 Supplement. We also contacted subject matter experts to identify ongoing or recently completed studies not identified in the database search.
Studies eligible for the review included randomized controlled studies or cohort studies (a) that enrolled individuals evaluated for TB after presenting for care at routine healthcare settings; (b) where treatment decisions were based on diagnostic tests in routine use in that setting (i.e., additional tests conducted for research purposes were not used); and (c) where participants were followed for least 1 week following the initial diagnostic test to record whether or not treatment was initiated. We excluded systematic reviews and studies of nonhuman subjects, latent TB, hospitalized patients, multidrug-resistant TB, and active case finding. We also excluded study participants younger than 18 years of age.
Authors SK and MC independently reviewed the titles and abstracts of each identified article, assessing them for inclusion or exclusion using Covidence (Veritas Health Innovation, Melbourne, Australia, available at www.covidence.org). During the second screening stage, full-text articles were obtained for all articles considered relevant or possibly relevant (“yes” or “maybe”) by both reviewers based on the initial title and abstract review. The authors then independently evaluated each full-text article to determine its eligibility. SK and NM contacted the investigators of studies meeting the inclusion criteria to obtain de-identified patient-level data. This study is reported as per the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guideline (see Table A in S1 Supplement).
Variables of interest
For each study data set, we extracted data on individual-level variables describing the type of initial test received (e.g., Xpert, Ultra, Truenat, SSM), age (18 years or older), sex, presence of TB-related symptoms (cough, fever, night sweats, weight loss), results for any non-bacteriological tests performed (e.g., chest radiography), HIV status, morbidity score (e.g., Karnofsky score ranging from 0 (dead) to 100 indicating no evidence of disease) if available, TB diagnosis, whether TB treatment was initiated, date of treatment initiation, date of testing, date culture result was returned (if applicable), and duration of follow-up. We also extracted contextual variables including calendar year, country, and type of clinic at which the patient was evaluated (primary, secondary). We excluded individuals with inconclusive or missing results for the initial diagnostic test.
After data extraction, we created a master list of variables available from each study. Relevant variables that could influence diagnostic decision-making were selected based on TB diagnostic algorithms and guidelines consolidated by WHO [3,22]. Given that each study has different variables and units, we selected common variables across studies for meta-analysis and converted variable types for consistency across studies (e.g., conversion of continuous variables to categorical variables for symptom durations (unit in weeks)). We collated the harmonized IPD into a single data set.
Our primary outcome was whether or not an individual initiated TB treatment following a negative SSM, Xpert, or Xpert Ultra result (i.e., the standard of care for initial TB testing in each setting at the time of the study). While some studies undertook additional investigational tests that were not part of the routine care, clinicians were blinded to these results (recorded in each trial report). Although most studies collected samples for sputum culture, we restricted our analysis to the period before culture results became available.
For studies that recorded a variable indicating whether or not treatment was provided on clinical grounds, we used this variable as our outcome measure. For all other studies, we defined clinical diagnosis as instances where treatment was initiated following negative initial test results but before culture results became available.
Data analysis
IPD meta-analysis was performed via logistic regression, specified for the binary outcome of whether or not an individual initiated treatment as defined above. To do so, we employed a hierarchical Bayesian model with country random effects (see Text B in S1 Supplement) to account for country-specific differences in diagnostic practices not reflected in other variables [23,24]. For the primary analysis, we fit univariable and multivariable regression models considering age (18 to 30 years, 31 to 40 years, >40 years), sex (female, male), history of prior TB (no, yes, unknown), reported cough (yes, no), reported night sweats (yes, no), reported fever (yes, no), HIV status (negative, positive (not on ART), positive (on ART), unknown), test type (SSM, Xpert, Xpert Ultra), and calendar year. These variables were included in the primary analysis based on their availability in the majority of data sets.
We conducted 2 secondary analyses using variables not available for a subset of data sets. First, using the data sets that provided information on symptom duration, we fit a modified version of the regression model for the primary analysis, in which the binary variables for cough, fever, and night sweats were replaced by versions of these variables that each stratified the observations into one of 3 levels (none, less than 2 weeks, 2 weeks and above). Second, for the data sets containing chest X-ray results, we reran the regression model for the primary analysis with this additional variable (normal, abnormal, unknown).
As a robustness check, we re-estimated the results of the main analysis with 2 alternative regression specifications. First, we adopted an alternative outcome definition, in which clinical diagnosis was defined as treatment initiation within 7 days of the initial diagnostic test. While potentially excluding some clinical diagnoses, this stricter definition may reduce the risk of bias due to variation in the definition of clinical diagnosis adopted by each study. Second, we re-estimated results using a regression model in which the country random effects were replaced by study random effects.
Additionally, we conducted a sensitivity analysis using the probit model, complemented by cross-validation. To address concerns about heterogeneity, we performed stratified analyses based on diagnostic tests and HIV status. We also reran the analysis with a model that combined Xpert and Xpert Ultra into a single category of rapid diagnostic tests (RDTs) for comparison against SSM. All statistical analyses were performed in R (v.4.2.3) using the “brms” package (v.2.19.0) [25–27].
Results
Our database search identified 4,286 potentially eligible studies. After removal of duplicates, this resulted in 3,428 unique references for screening. After review of title and abstract of those references, full-text screening was performed on 161 studies, with 51 eligible studies identified (Fig 1). Following communication with investigators for each study, we obtained data from 18 eligible studies. Six of these studies were excluded after initial data cleaning due to missing key variables or considering a different target population. The final data set included observations collected between 2011 and 2020, covering 13 countries across 12 studies. Most of these countries are classified as high-burden for TB by the WHO. Table 1 reports demographic and clinical characteristics for the full analytic sample, and Table C and Text C in S1 Supplement provide details of each included study.
Primary analysis
The main analysis included data for 15,121 adults evaluated for pulmonary TB for whom the initial TB test was negative. Of these individuals, 477 were initiated on TB treatment following clinical diagnosis. Table 2 summarizes the meta-analysis results as odds ratios (ORs) and adjusted odds ratios (aORs) produced by univariable and multivariable regression models, respectively, representing the odd ratio of TB treatment initiation among individuals with a given factor compared to the reference category.
Based on the multivariable analysis, we identified statistically significant increases in the odds of TB treatment initiation associated with male sex (aOR 1.61 compared to female sex, 95% credible interval (CI): 1.31, 1.95), having a history of prior TB (aOR 1.36 compared to individuals without prior TB, 95% CI: 1.06, 1.73), having reported cough (aOR 4.62 compared to no cough, 95% CI: 3.42, 6.27), having reported night sweats (aOR 1.50 compared to no night sweats, 95% CI: 1.21, 1.90), and having HIV infection but not on ART (aOR 1.68 compared to HIV–negative, 95% CI: 1.23, 2.32).
In terms of the tests used for initial TB diagnosis, we found lower odds of treatment initiation for individuals who had received a negative result on Xpert (aOR 0.77 compared to diagnosis via SSM, 95% CI: 0.62, 0.96) and who had received a negative result on Xpert Ultra (aOR 0.57 compared to diagnosis via SSM, 95% CI: 0.30, 1.07), although the results for Xpert Ultra were not statistically significant. We also estimated declining rates of treatment initiation over time, controlling for other factors (aOR 0.81 for each additional calendar year, 95% CI: 0.74, 0.90).
Secondary analyses
In the first secondary analysis, we estimated ORs for cough, fever, and night sweats categorized by duration of symptoms, using data from the 5 studies for which this variable was available (7,468 observations). These findings indicated strong positive associations between TB treatment initiation and a reported cough of 0 to 2 weeks duration (aOR 3.29 compared to no reported cough, 95% CI: 1.64, 7.34) and >2 weeks duration (aOR 5.34 compared to no reported cough, 95% CI: 2.72, 11.82) (Fig 2). Reported night sweats of 0 to 2 weeks duration also demonstrated elevated odds of treatment initiation (aOR 1.45 compared to no reported night sweats, 95% CI: 1.06, 2.00).
.* Reference group: Age 18–30 years old, female sex, no history of prior TB, no reported cough, no reported fever, no reported night sweats, HIV–negative, tested negative with SSM. Blue symbols signify ORs >1.0, red symbols signify ORs <1.0. ART, antiretroviral therapy; OR, odds ratio; SSM, sputum smear microscopy; TB, tuberculosis.
The second secondary analysis estimated differences in treatment initiation based on chest X-ray result, using data from the 3 studies in which X-ray was conducted as part of TB evaluation (2,449 observations). In these data, 1,677 individuals had a normal X-ray result (1.1% (18/1,677) initiated on treatment) and 456 had an abnormal X-ray result (6.1% (28/456) initiated on treatment). The results of this analysis showed that having an abnormal X-ray result had a strong positive association with treatment initiation, with an aOR of 6.89 (95% CI: 3.29, 14.42) compared to individuals with normal X-ray results (Table D in S1 Supplement).
Alternative model specifications
Table 3 presents results for 2 alternative model specifications. In the first alternative specification, we analyzed an alternative outcome defined as treatment initiation within 7 days of the initial diagnostic test, representing 1.4% (205/15,121) of all observations. These results were generally consistent with the results of the primary analysis, although the odds ratio estimated for receiving a negative result on Xpert Ultra was lower than in the primary analysis and statistically significant (aOR 0.35 compared to diagnosis via SSM, 95% CI: 0.17, 0.75). Additionally, the estimated time trend in treatment initiation was no longer significant (aOR 0.97 for each additional calendar year, 95% CI: 0.85, 1.09).
The results for the second alternative specification (results estimated with study random effects instead of country random effects) were generally consistent with the results of the primary analysis, although the odds ratio estimated for receiving a negative result on Xpert Ultra was lower than in the primary analysis and statistically significant (aOR 0.37 compared to diagnosis via SSM, 95% CI: 0.21, 0.64). The estimated time trend in treatment initiation was no longer significant (aOR 0.87 for each additional calendar year, 95% CI: 0.74, 1.04).
Our sensitivity analysis results showed that our findings remain consistent across different models (logit versus probit) with cross-validation confirming that our main model exhibits a better fit (Table F in S1 Supplement). In addition, stratified analyses across different diagnostic tests and HIV status demonstrated the robustness and consistency of our results across various subgroups (Tables G and H in S1 Supplement). Further, the results combining Xpert and Xpert Ultra into a single category of RDT were also consistent with the findings of the primary analysis, as shown in Table I in S1 Supplement. Lastly, Table J in S1 Supplement provides additional evidence of the influence of each predictor on treatment decisions, reporting the absolute risks and risk differences of treatment initiation associated with a change in each predictor, holding others constant.
Discussion
This study examined the factors associated with treatment initiation among adults evaluated for TB in routine healthcare settings, who had received a negative result on an initial bacteriological test for TB. Our analyses showed that male sex, a history of prior TB, reported cough, and having HIV infection but not receiving ART were positively associated with clinicians’ decisions to initiate TB treatment. Among the 3 tests used for initial diagnosis, individuals receiving a negative result on Xpert were substantially less likely to be initiated on treatment compared to individuals who had received a negative result with SSM. Though not statistically significant in the main analysis, a negative result on Xpert Ultra was also associated with lower treatment initiation rates compared to SSM. In addition, the secondary analyses demonstrated increasing odds of treatment initiation with longer duration of cough (specifically, cough persisting for over 2 weeks). Similarly, the presence of an abnormal chest X-ray result was found to have a strong positive association with treatment initiation. We also observed a lower likelihood of treatment initiation in more recent years, controlling for other factors.
Most results from the alternative model specifications were consistent with those of the primary analysis. For the first alternative specification (outcome defined as treatment initiation within 7 days of the initial TB test), the fraction diagnosed clinically was lower than in the primary analysis (3.2% versus 1.4%), and this outcome definition may have excluded some individuals who were treated clinically but with a greater delay. However, this outcome definition reduced potential inter-study variation in the definition of clinical diagnosis, and the risk of bias due to clinicians accessing culture results before making treatment decisions. The second alternative specification assumed that residual variation in clinical decision-making was primarily attributable to study-specific factors (versus country-specific factors in the main analysis). That the estimated odds ratios were mostly consistent across different model specifications provides some assurance that these results are robust. One small difference was for Xpert Ultra, for which in both alternative specifications individuals testing negative on Xpert Ultra were estimated to be significantly less likely to begin treatment compared to those who received a negative result from SSM, with these odds ratios lower than those estimated in the primary analysis, and statistically significant. In addition, the results describing the time trend were no longer statistically significant in both alternative specifications.
The findings for individual covariates can be interpreted in light of factors that clinicians may consider during TB diagnosis. These considerations include the pre-test probability of disease (prevalence of TB disease among individuals being tested), the expected magnitude of harms resulting from an incorrect negative diagnosis relative to the harms of an incorrect positive diagnosis, and the expected sensitivity and specificity of the tests being used. Several of the covariates examined in this study are relevant to these considerations.
First, several of the covariates we examined may influence clinician’s beliefs about the pre-test probability of disease. Based on WHO guidelines for TB diagnosis and treatment in HIV-prevalent and resource constrained settings, a history of prior TB and symptoms suggestive of TB imply a higher pre-test probability of disease, and therefore may increase clinical suspicion for TB [28]. Similarly, in many settings persons living with HIV have higher TB incidence compared to HIV–negative individuals, and men have elevated incidence rates compared to women, such that clinicians may expect these characteristics to imply a higher disease prevalence among those evaluated for TB. In light of these relationships (each of which was linked to elevated treatment initiation rates), it is somewhat surprising that reported fever had a modest association with treatment initiation. While the presence of fever has been associated with TB, it is also associated with many other conditions, and therefore may be of limited value in distinguishing TB from other alternative diagnoses (as has been found with antibiotic trial as a diagnostic modality [29]).
For the second consideration (harms resulting from an incorrect negative diagnosis relative to the harms from an incorrect positive diagnosis), it is possible that this contributes to the elevated treatment initiation odds estimated for persons living with HIV, as compared to HIV–negative individuals. Individuals with both HIV and TB experience rapid disease progression and are less likely to survive the TB episode compared to HIV–negative individuals with TB [30–32]. As a consequence, the urgency of initiating TB treatment (if TB is suspected) will be much greater for individuals found to have HIV compared to those living without HIV. In contrast, the harms produced by a false-positive diagnosis, while not trivial, may not differ substantially between individuals with and without HIV.
For the third consideration (test sensitivity and specificity), this may explain the results estimated for the different test types (smear microscopy, Xpert, Xpert Ultra). The poor sensitivity of smear microscopy for pulmonary TB is well known, as is the improved performance of Xpert and Xpert Ultra compared to smear microscopy [33,34]. Because of the higher sensitivity of these new PCR-based tests, an individual testing negative on one of these tests is less likely to have TB than if the individual had instead tested negative with smear, all other things being equal. Clinicians aware of these relationships may be more hesitant to recommend treatment for patients that have tested negative with a high-sensitivity test. It is also true that each of the tests examined is known to have lower sensitivity among individuals with HIV infection [35], and this may be an additional factor contributing to the higher odds of treatment initiation for HIV–positive individuals following a negative test.
If it is true that clinicians are less likely to make a clinical diagnosis following a negative Xpert or Xpert Ultra result (versus a negative result on SSM), this could have implications for the impact of these new diagnostic tests on overall algorithm performance. Earlier modeling studies have demonstrated that clinical diagnosis could reduce the incremental effects of Xpert introduction on algorithm sensitivity, assuming that negative Xpert and negative SSM results are treated the same way by clinicians [36,37]. These effects could be magnified if clinicians are less likely to make clinical diagnoses following a negative Xpert, further reducing the overall impact of Xpert introduction on algorithm sensitivity, while at the same time increasing algorithm specificity. For individuals testing false-negative with Xpert, greater hesitance to initiate treatment based on clinical criteria could increase diagnostic delays, prolonging TB-related morbidity and mortality risks. Given the urgency of increasing TB case detection, further research on these potential mechanisms—and how to optimally balance the trade-offs involved in TB diagnosis—is needed.
There are several limitations to this study. First, we were not able to analyze all factors that potentially inform clinician decision-making, due to differences in the covariates recorded in the study data sets. It is possible that additional individual characteristics—such as recent weight loss or reporting a known TB contact, or medical comorbidities such as diabetes or other immunosuppressive conditions—may impact clinical decision-making but were not consistently captured in the study data. When X-ray results were available, they were found to have a major impact on clinical decision-making, but X-ray was only performed in a minority of studies. Similarly, it is possible that factors related to the healthcare setting, differences in national guidelines or protocols, or the capabilities of clinicians performing diagnosis may influence rates of clinical diagnosis. However, these setting-specific data were not available for analysis, contributing to differences in treatment initiation across countries. These setting-specific differences were substantial (quantified by the country random effects included in the main analysis), pointing to the existence of additional determinants of treatment initiation not captured by our analysis. Additional research to identify these factors is needed.
Second, our analytic population excluded patients aged under 18. While diagnosis for older children and adolescents may be similar to adults, clinicians will have different decision criteria for diagnosis of infants, due to both the different presentation of TB and the poor performance of available TB diagnostics in young children.
Third, while we selected studies to only include those performed under routine clinical conditions, it is possible that the behavior of clinicians performing TB diagnosis could have been influenced by their participation in clinical research. For example, it is possible that rates of clinical diagnosis will be lower in trial settings, if clinicians believe that missed diagnoses can be resolved through additional diagnostic testing undertaken as part of the trial (such as via sputum culture, performed in the majority of studies included in our review). If no additional testing is expected, clinicians may be more willing to make clinical diagnoses. Conversely, for routine settings where follow-up testing is common (or where multiple tests are conducted concurrently) rates of clinical diagnosis could be similar to those observed in trial settings. It is also possible that trial protocols may have influenced the factors considered during clinical diagnosis. Moreover, the clinics in which these studies were conducted may have been selected based on their capacity to participate in research, which may limit their representativeness of the general context of TB care.
Fourth, while many of the findings of the analysis are consistent with general principles of good patient care (as discussed above), we did not have access to additional evidence describing why clinicians made the decisions they did. Fifth, we did not compare clinical diagnosis decisions with culture results that subsequently became available. While such a comparison was outside the scope of the current study—which focused on clinical decisions made before any additional test results became available—this comparison would be useful for judging the diagnostic accuracy of clinical diagnosis and could be addressed in a subsequent study.
In conclusion, in this multi-country IPD meta-analysis of clinical diagnosis for TB, we found multiple clinical factors to be associated with the decision to initiate TB among individuals who receive a negative result on an initial bacteriological test for TB. Understanding these factors will allow for a more nuanced interpretation of the data describing the impact of introducing new TB diagnostics [37–39] and can inform efforts to refine clinical diagnostic algorithms, determine the appropriate balance between sensitivity and specificity when revising diagnostic approaches [40], and improve the overall performance of TB case detection.
Supporting information
S1 Supplement.
Table A. PRISMA Checklist. Table B. Demographic and clinical characteristics of participants, by study. Table C. Odds ratios of TB treatment initiation following negative diagnostic test result: country random effects from primary analysis. Table D. Odds ratios of TB treatment initiation following negative diagnostic test result: secondary analysis for data sets including chest X-ray results. Table E. Odds of TB treatment initiation following negative diagnostic test result: study random effects from alternative model specification. Table F. Sensitivity analysis comparing Logit vs. Probit model. Table G. Stratified analysis by diagnostic tests. Table H. Stratified analysis by HIV results. Table I. Odds ratios of TB treatment initiation following a negative diagnostic test result (Xpert/Xpert Ultra combined). Table J. Absolute risk and risk difference of treatment initiation. Table K. Contact information for accessing each data set included in the study. Text A. Search terms for Embase (Elsevier, embase.com). Text B. Hierarchical Bayesian logistic regression model. Text C. Description of individual studies included in analysis.
https://doi.org/10.1371/journal.pmed.1004502.s001
(PDF)
S1 PROSPERO Protocol. PROSPERO: CRD42022287613.
https://doi.org/10.1371/journal.pmed.1004502.s002
(PDF)
References
- 1.
World Health Organization. Global Tuberculosis Report 2023 [Internet]. 2023 [cited 2024 Jan 15]. https://www.who.int/teams/global-tuberculosis-programme/tb-reports/global-tuberculosis-report-2023.
- 2.
World Health Organization. The end TB strategy [Internet]. 2015. https://www.who.int/publications-detail-redirect/WHO-HTM-TB-2015.19.
- 3.
World Health Organization. WHO consolidated guidelines on tuberculosis. Module 3: Diagnosis—Rapid diagnostics for tuberculosis detection 2021 update [Internet]. 2021. https://www.who.int/publications-detail-redirect/9789240029415.
- 4. Hong JM, Lee H, Menon NV, Lim CT, Lee LP, Ong CWM. Point-of-care diagnostic tests for tuberculosis disease. Sci Transl Med. 2022 Apr 6;14(639):eabj4124. pmid:35385338
- 5. Calligaro GL, Theron G, Khalfey H, Peter J, Meldau R, Matinyenya B, et al. Burden of tuberculosis in intensive care units in Cape Town, South Africa, and assessment of the accuracy and effect on patient outcomes of the Xpert MTB/RIF test on tracheal aspirate samples for diagnosis of pulmonary tuberculosis: a prospective burden of disease study with a nested randomised controlled trial. Lancet Respir Med. 2015 Aug;3(8):621–30. pmid:26208996
- 6. Churchyard GJ, Stevens WS, Mametja LD, McCarthy KM, Chihota V, Nicol MP, et al. Xpert MTB/RIF versus sputum microscopy as the initial diagnostic test for tuberculosis: a cluster-randomised trial embedded in South African roll-out of Xpert MTB/RIF. Lancet Glob Health. 2015 Aug;3(8):e450–7. pmid:26187490
- 7. Cox HS, Mbhele S, Mohess N, Whitelaw A, Muller O, Zemanay W, et al. Impact of Xpert MTB/RIF for TB diagnosis in a primary care clinic with high TB and HIV prevalence in South Africa: a pragmatic randomised trial. PLoS Med. 2014 Nov;11(11):e1001760. pmid:25423041
- 8. Durovni B, Saraceni V, van den Hof S, Trajman A, Cordeiro-Santos M, Cavalcante S, et al. Impact of Replacing Smear Microscopy with Xpert MTB/RIF for Diagnosing Tuberculosis in Brazil: A Stepped-Wedge Cluster-Randomized Trial. PLoS Med. 2014 Dec 9;11(12):e1001766. pmid:25490549
- 9. Hanrahan CF, Selibas K, Deery CB, Dansey H, Clouse K, Bassett J, et al. Time to treatment and patient outcomes among TB suspects screened by a single point-of-care xpert MTB/RIF at a primary care clinic in Johannesburg, South Africa. PLoS ONE. 2013;8(6):e65421. pmid:23762367
- 10. Mupfumi L, Makamure B, Chirehwa M, Sagonda T, Zinyowera S, Mason P, et al. Impact of Xpert MTB/RIF on Antiretroviral Therapy-Associated Tuberculosis and Mortality: A Pragmatic Randomized Controlled Trial. Open Forum Infect Dis. 2014 Mar;1(1):ofu038. pmid:25734106
- 11. Theron G, Zijenah L, Chanda D, Clowes P, Rachow A, Lesosky M, et al. Feasibility, accuracy, and clinical effect of point-of-care Xpert MTB/RIF testing for tuberculosis in primary-care settings in Africa: a multicentre, randomised, controlled trial. Lancet. 2014 Feb 1;383(9915):424–35. pmid:24176144
- 12. Trajman A, Durovni B, Saraceni V, Menezes A, Cordeiro-Santos M, Cobelens F, et al. Impact on Patients’ Treatment Outcomes of XpertMTB/RIF Implementation for the Diagnosis of Tuberculosis: Follow-Up of a Stepped-Wedge Randomized Clinical Trial. PLoS ONE. 2015 Apr 27;10(4):e0123252. pmid:25915745
- 13. Yoon C, Cattamanchi A, Davis JL, Worodria W, den Boon S, Kalema N, et al. Impact of Xpert MTB/RIF testing on tuberculosis management and outcomes in hospitalized patients in Uganda. PLoS ONE. 2012;7(11):e48599. pmid:23139799
- 14. Auld AF, Fielding KL, Gupta-Wright A, Lawn SD. Xpert MTB/RIF—why the lack of morbidity and mortality impact in intervention trials? Trans R Soc Trop Med Hyg. 2016 Aug;110(8):432–44. pmid:27638038
- 15. Lawn SD, Nicol MP, Corbett EL. Effect of empirical treatment on outcomes of clinical trials of diagnostic assays for tuberculosis. Lancet Infect Dis. 2015 Jan 1;15(1):17–8. pmid:25541165
- 16. Rie AV. Should countries implement Xpert MTB/RIF when empirical treatment precludes a clinical effect? Lancet Respir Med. 2015 Aug 1;3(8):591–3. pmid:26208997
- 17. Theron G, Peter J, Dowdy D, Langley I, Squire SB, Dheda K. Do high rates of empirical treatment undermine the potential effect of new diagnostic tests for tuberculosis in high-burden settings? Lancet Infect Dis. 2014 Jun 1;14(6):527–32. pmid:24438820
- 18. Swai HF, Mugusi FM, Mbwambo JK. Sputum smear negative pulmonary tuberculosis: sensitivity and specificity of diagnostic algorithm. BMC Res Notes. 2011 Nov 1;4:475. pmid:22044882
- 19. Huerga H, Varaine F, Okwaro E, Bastard M, Ardizzoni E, Sitienei J, et al. Performance of the 2007 WHO Algorithm to Diagnose Smear-Negative Pulmonary Tuberculosis in a HIV Prevalent Setting. PLoS ONE. 2012 Dec 19;7(12):e51336. pmid:23284681
- 20. Abebe G, Deribew A, Apers L, Abdissa A, Kiflie Y, Koole O, et al. Evaluation of the 2007 WHO guideline to diagnose smear negative tuberculosis in an urban hospital in Ethiopia. BMC Infect Dis. 2013 Sep 11;13(1):427.
- 21.
Kim S, Can M, Dorman S, Sweeney S, Cohen T, Menzies N. Factors associated with TB treatment initiation among bacteriologically-negative TB suspects: an individual patient data meta-analysis. PROSPERO 2022 CRD42022287613. 2022; https://www.crd.york.ac.uk/prospero/display_record.php?ID=CRD42022287613.
- 22.
World Health Organization. Algorithm for laboratory diagnosis and treatment-monitoring of pulmonary tuberculosis and drug-resistant tuberculosis using state-of-the-art rapid molecular diagnostic technologies. 2017.
- 23. Abo-Zaid G, Guo B, Deeks JJ, Debray TPA, Steyerberg EW, Moons KGM, et al. Individual participant data meta-analyses should not ignore clustering. J Clin Epidemiol. 2013 Aug;66(8):865. pmid:23651765
- 24. Burke DL, Ensor J, Riley RD. Meta-analysis using individual participant data: one-stage and two-stage approaches, and why they may differ. Stat Med. 2017;36(5):855–75. pmid:27747915
- 25. Bürkner PC. brms: An R Package for Bayesian Multilevel Models Using Stan. J Stat Softw. 2017 Aug 29;80:1–28.
- 26. Bürkner PC. Advanced Bayesian Multilevel Modeling with the R Package brms. R J. 2018;10(1):395–411.
- 27. Bürkner PC. Bayesian Item Response Modeling in R with brms and Stan. J Stat Softw. 2021 Nov 30;100:1–54.
- 28.
World Health Organization. Improving the diagnosis and treatment of smear-negative pulmonary and extrapulmonary tuberculosis among adults and adolescents. Recommendations for HIV-prevalent and resource-constrained settings [Internet]. 2006. chrome-extension://efaidnbmnnnibpcajpcglclefindmkaj/viewer.html?pdfurl=https%3A%2F%2Fwww.who.int%2Ftb%2Fpublications%2F2006%2Ftbhiv_recommendations.pdf&clen=393462&chunk=true
- 29. Divala TH, Fielding KL, Kandulu C, Nliwasa M, Sloan DJ, Gupta-Wright A, et al. Utility of broad-spectrum antibiotics for diagnosing pulmonary tuberculosis in adults: a systematic review and meta-analysis. Lancet Infect Dis. 2020 Sep;20(9):1089–98. pmid:32437700
- 30. Corbett EL, Watt CJ, Walker N, Maher D, Williams BG, Raviglione MC, et al. The growing burden of tuberculosis: global trends and interactions with the HIV epidemic. Arch Intern Med. 2003 May 12;163(9):1009–21. pmid:12742798
- 31. Straetemans M, Glaziou P, Bierrenbach AL, Sismanidis C, van der Werf MJ. Assessing Tuberculosis Case Fatality Ratio: A Meta-Analysis. PLoS ONE. 2011 Jun 27;6(6):e20755. pmid:21738585
- 32. MacPherson P, Dimairo M, Bandason T, Zezai A, Munyati SS, Butterworth AE, et al. Risk factors for mortality in smear-negative tuberculosis suspects: a cohort study in Harare, Zimbabwe. Int J Tuberc Lung Dis. 2011 Oct 1;15(10):1390–6. pmid:22283900
- 33. Dorman SE, Schumacher SG, Alland D, Nabeta P, Armstrong DT, King B, et al. Xpert MTB/RIF Ultra for detection of Mycobacterium tuberculosis and rifampicin resistance: a prospective multicentre diagnostic accuracy study. Lancet Infect Dis. 2018 Jan;18(1):76–84. pmid:29198911
- 34. Steingart KR, Schiller I, Horne DJ, Pai M, Boehme CC, Dendukuri N. Xpert MTB/RIF assay for pulmonary tuberculosis and rifampicin resistance in adults. Cochrane Database Syst Rev. 2014 Jan 21;(1):CD009593.
- 35. Horne DJ, Kohli M, Zifodya JS, Schiller I, Dendukuri N, Tollefson D, et al. Xpert MTB/RIF and Xpert MTB/RIF Ultra for pulmonary tuberculosis and rifampicin resistance in adults. Cochrane Database Syst Rev. 2019 Jun 7;2019(6):CD009593. pmid:31173647
- 36. Menzies NA, Cohen T, Lin HH, Murray M, Salomon JA. Population Health Impact and Cost-Effectiveness of Tuberculosis Diagnosis with Xpert MTB/RIF: A Dynamic Simulation and Economic Evaluation. PLoS Med. 2012 Nov 20;9(11):e1001347. pmid:23185139
- 37. Menzies NA, Cohen T, Murray M, Salomon JA. Effect of empirical treatment on outcomes of clinical trials of diagnostic assays for tuberculosis. Lancet Infect Dis. 2015 Jan;15(1):16–7. pmid:25541164
- 38. Sun AY, Denkinger CM, Dowdy DW. The impact of novel tests for tuberculosis depends on the diagnostic cascade. Eur Respir J. 2014 Nov 1;44(5):1366–9. pmid:25186263
- 39. Arinaminpathy N, Dowdy D. Understanding the incremental value of novel diagnostic tests for tuberculosis. Nature. 2015 Dec;528(7580):S60–7. pmid:26633767
- 40. Mwaura M, Kao K, Wambugu J, Trollip A, Sikhondze W, Omesa E, et al. Situating trade-offs: Stakeholder perspectives on overtreatment versus missed diagnosis in transition to Xpert MTB/RIF Ultra in Kenya and Swaziland. PLoS ONE. 2020 Feb 19;15(2):e0228669. pmid:32074142