Articles Predictive performance of interferon-gamma release assays and the tuberculin skin test for incident tuberculosis: an individual participant data meta-analysis Yohhei Hamada,a,* Rishi K. Gupta,a Matteo Quartagno,b Abbie Izzard,a Carlos Acuna-Villaorduna,c Neus Altet,d,e Roland Diel,f Jose Dominguez,g Sian Floyd,h Amita Gupta,i Helena Huerga,j Edward C. Jones-López,k Aarti Kinikar,l Christoph Lange,m,n,o,p,q Frank van Leth,q,r,s Qiao Liu,t Wei Lu,t Peng Lu,t Irene Latorre Rueda,g Leonardo Martinez,u Stanley Kimbung Mbandi,v Laura Muñoz,w Elisabeth Sánchez Padilla,j Mandar Paradkar,x,y Thomas Scriba,v Martina Sester,q,z Kwame Shanaube,aa Surendra K. Sharma,ab,ac,ad Rosa Sloot,ae Giovanni Sotgiu,q,af Kannan Thiruvengadam,ag Richa Vashishtha,ab Ibrahim Abubakar,a and Molebogeng X. Rangakaa,ah a Institute for Global Health, University College London, London, United Kingdom b MRC Clinical Trials Unit, Institute of Clinical Trials and Methodology, University College London, London, United Kingdom c Section of Infectious Diseases, Boston University Medical Center, Boston, MA, USA d Unitat de Tuberculosis, Hospital Universitari Vall d’Hebron-Drassanes, Barcelona, Spain e Unitat de TDO de la Tuberculosis ‘Servicios Clínicos’, Barcelona, Spain f Institute for Epidemiology, University Hospital Schleswig-Holstein, Campus Kiel, Kiel, Germany g Institut d’Investigació Germans Trias i Pujol, CIBER Enfermedades Respiratorias, Instituto de Salud Carlos III, Universitat Autònoma de Barcelona, Barcelona, Spain h Department of Infectious Disease Epidemiology, London School of Hygiene & Tropical Medicine, London, United Kingdom i Johns Hopkins University School of Medicine, Baltimore, MD, USA j Epicentre, Paris, France k Division of Infectious Diseases, Department of Medicine, Keck School of Medicine of USC, University of Southern California, Los Angeles, CA, USA l Byramjee Jeejeebhoy Government Medical College and Sassoon General Hospital, Pune, Maharashtra, India m Division of Clinical Infectious Diseases, Research Center Borstel, Borstel, Germany n German Center for Infection Research (DZIF), Clinical Tuberculosis Unit, Borstel, Germany o Respiratory Medicine & International Health, University of Lübeck, Lübeck, Germany p Baylor College of Medicine and Texas Children’s Hospital, Houston, TX, USA q Tuberculosis Network European Trials Group (TBnet), Borstel, Germany r Department of Health Sciences, VU University, Amsterdam, the Netherlands s Amsterdam Public Health research institute, Amsterdam, the Netherlands t Department of Chronic Communicable Disease, Center for Disease Control and Prevention of Jiangsu Province, Nanjing, Jiangsu Province, PR China u Department of Epidemiology, School of Public Health, Boston University, Boston, MA, USA v South African Tuberculosis Vaccine Initiative, Institute of Infectious Disease and Molecular Medicine, and Division of Immunology, Department of Pathology, University of Cape Town, South Africa, Western Cape, South Africa w Department of Clinical Sciences, University of Barcelona, Barcelona, Spain x Byramjee Jeejeebhoy Government Medical College-Johns Hopkins University Clinical Research Site, Pune, Maharashtra, India y Johns Hopkins India, Pune, Maharashtra, India z Department of Transplant and Infection Immunology, Saarland University, Homburg, Germany aa Zambart, Lusaka, Zambia ab Department of Internal Medicine, All India Institute of Medical Sciences, New Delhi, India ac Department of Molecular Medicine, Jamia Hamdard Institute of Molecular Medicine, Hamdard University, Delhi, India ad Departments of General Medicine & Pulmonary Medicine, JNMC, Datta Meghe Institute of Medical Sciences, Maharashtra, India ae Desmond Tutu TB Centre, Department of Paediatrics and Child Health, Faculty of Medicine and Health Sciences, Stellenbosch University, Cape Town, South Africa af Clinical Epidemiology and Medical Statistics Unit, Department of Medicine, Surgery and Pharmacy, University of Sassari, Sassari, Italy ag National Institute for Research in Tuberculosis, Indian Council of Medical Research, Chennai, Tamil Nadu, India ah School of Public Health, and Clinical Infectious Disease Research Institute-Africa, University of Cape Town, Cape Town, South Africa *Corresponding author. E-mail address: y.hamada@ucl.ac.uk (Y. Hamada). www.thelancet.com Vol 56 February, 2023 1 Articles eClinicalMedicine Summary 2023;56: 101815 Background Evidence on the comparative performance of purified protein derivative tuberculin skin tests (TST) and Published Online xxx interferon-gamma release assays (IGRA) for predicting incident active tuberculosis (TB) remains conflicting. We https://doi.org/10. conducted an individual participant data meta-analysis to directly compare the predictive performance for incident 1016/j.eclinm.2022. TB disease between TST and IGRA to inform policy. 101815 Methods We searched Medline and Embase from 1 January 2002 to 4 September 2020, and studies that were included in previous systematic reviews. We included prospective longitudinal studies in which participants received both TST and IGRA and estimated performance as hazard ratios (HR) for the development of all diagnoses of TB in partici- pants with dichotomised positive test results compared to negative results, using different thresholds of positivity for TST. Secondary analyses included an evaluation of the impact of background TB incidence. We also estimated the sensitivity and specificity for predicting TB. We explored heterogeneity through pre-defined sub-group analyses (e.g. country-level TB incidence). Publication bias was assessed using funnel plots and Egger’s test. This review is registered with PROSPERO, CRD42020205667. Findings We obtained data from 13 studies out of 40 that were considered eligible (N = 32,034 participants: 36% from countries with TB incidence rate ≥100 per 100,000 population). All reported data on TST and QuantiFERON Gold in- Tube (QFT-GIT). The point estimate for the TST was highest with higher cut-offs for positivity and particularly when stratified by bacillus Calmette–Guérin vaccine (BCG) status (15 mm if BCG vaccinated and 5 mm if not [TST5/15 mm]) at 2.88 (95% CI 1.69–4.90). The pooled HR for QFT-GIT was higher than for TST at 4.15 (95% CI 1.97–8.75). The difference was large in countries with TB incidence rate <100 per 100,000 population (HR 10.38, 95% CI 4.17–25.87 for QFT-GIT VS. HR 5.36, 95% CI 3.82–7.51 for TST5/15 mm) but much of this difference was driven by a single study (HR 5.13, 95% CI 3.58–7.35 for TST5/15 mm VS. 7.18, 95% CI 4.48–11.51 for QFT-GIT, when excluding the study, in which all 19 TB cases had positive QFT-GIT results). The comparative performance was similar in the higher burden countries (HR 1.61, 95% CI 1.23–2.10 for QFT-GIT VS. HR 1.72, 95% CI 0.98–3.01 for TST5/15 mm). The predictive performance of both tests was higher in countries with TB incidence rate <100 per 100,000 population. In the lower TB incidence countries, the specificity of TST (76% for TST5/15 mm) and QFT-GIT (74%) for predicting active TB approached the minimum World Health Organization target (≥75%), but the sensitivity was below the target of ≥75% (63% for TST5/15 mm and 65% for QFT-GIT). The absolute differences in positive and negative predictive values between TST15 mm and QFT-GIT were small (positive predictive values 2.74% VS. 2.46%; negative predictive values 99.42% VS. 99.52% in low-incidence countries). Egger’s test did not show evidence of publication bias (0.74 for TST15 mm and p = 0.68 for QFT-GIT). Interpretation IGRA appears to have higher predictive performance than the TST in low TB incidence countries, but the difference was driven by a single study. Any advantage in clinical performance may be small, given the numerically similar positive and negative predictive values. Both IGRA and TST had lower performance in countries with high TB incidence. Test choice should be contextual and made considering operational and likely clinical impact of test results. Funding YH, IA, and MXR were supported by the National Institute for Health and Care Research (NIHR), United Kingdom (RP-PG-0217-20009). MQ was supported by the Medical Research Council [MC_UU_00004/07]. Copyright © 2022 The Author(s). Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/). Keywords: LTBI; IGRA; Tuberculin skin test; Prophylaxis; Prevention Introduction protein derivative (PPD) tuberculin skin test (TST) and The World Health Organization (WHO) recommends interferon-gamma release assays (IGRA).1 In both, a preventive treatment in populations who are at risk for simple dichotomization to positive and negative denotes the development of tuberculosis (TB).1 Individuals who probable infection or not, respectively. None are are infected with Mycobacterium tuberculosis without considered “gold standard tests for TB infection” since clinical manifestation, known as latent TB infection, are they measure immune sensitization to mycobacterial at risk for the development of TB.1 Two types of indirect antigens and, consequently, they cannot differentiate diagnostics are recommended to detect individuals who current TB infection from past infection, as well as might benefit from TB preventive treatment: purified infection from disease.2 TST also lacks specificity for TB 2 www.thelancet.com Vol 56 February, 2023 Articles Research in context Evidence before this study and IGRA. This enabled the exploration of systematic We searched Medline and Embase from 1 January 2002 to 4 differences between study settings, study designs, TST cut-off September 2020 with search terms related to ‘tuberculosis values, participants and follow-up intervals and address (TB)’, ‘ interferon-gamma release assays (IGRA)’, and limitations of previous systematic reviews. We found that ‘tuberculin skin test (TST)’ and studies that were included in QuantiFERON Gold in-Tube (QFT-GIT) had a higher point previous systematic reviews. Current global policy strongly estimate of predictive performance. The difference appears to recommends, albeit with very low certainty of evidence, that be driven by one study, in which all TB cases had positive either TST or IGRA can be used for TB infection screening in QFT-GIT results in a low TB incidence setting. An analysis all populations and settings regardless of the background TB without this study resulted in similar point estimates for TST incidence. This was based on a previous meta-analysis of and IGRA, suggesting comparable performance. When we head-to-head studies which showed both tests imprecisely stratified study countries into two groups based on their TB indicated an increased risk of incident TB, and although IGRA incidence rate, the pooled predictive performance for TST and had higher relative risk estimates than the TST, confidence IGRA was higher in countries with TB incidence rates <100 per intervals overlapped. The higher relative risk estimates of 100,000 population. In low TB incidence settings, the higher IGRA was driven by studies with a high risk of incorporation predictive performance of IGRA resulted in a minimal where these were conducted in low-incidence resource-rich difference in positive and negative predictive values for countries that had widely adopted the test and used positive incident TB. The performance between TST and IGRA was IGRA results to trigger investigations of active TB only in test similar in high TB incidence countries. positives. The conclusion was that, in head-to-head In low TB incidence countries, both IGRA and TST had around evaluations, IGRA showed no clear predictive advantage over 75% specificity (74% for QFT-GIT and 76% for TST using the TST and that the tests had similar predictive performance. 15 mm cut-off for bacillus Calmette−Guérin vaccinated and In contrast, a recent systematic review concluded that IGRA 5 mm for unvaccinated) over two years, which is at World had better predictive performance than TST. However, the Health Organization (WHO) targets in low TB incidence primary analysis was indirect and not restricted to head-to- countries (>75%) but the sensitivity was below the target. The head studies. Both these previous reviews conducted study- performance of both tests was worse in countries with higher level meta-analyses. This approach is prone to bias due to TB incidence and did not meet WHO targets. systematic differences between study settings, participants, Implications of all the available evidence and follow-up intervals. The comparative prognostic ability of IGRA appears to predict TB incidence better than TST in low TST and IGRA has therefore remained unclear, due to these TB incidence settings, but the difference was largely driven by conflicting analyses. We updated the search on 11 December a single study, and the absolute predictive advantage may be 2022 with the same search terms, which identified two small. By contrast, the tests perform similarly in high TB prospective cohort studies in low TB incidence countries that incidence countries. Both have limited prognostic ability, tested participants with both TST and IGRA; however, one particularly in settings with higher TB incidence. While we (Calzada-Hernández 2022) included a small sample size and await development, validation and roll-out of new tests with event rate (two TB cases in 283 participants), and the other improved prognostic ability, other strategies such as the use (Ho 2022) reported that data on TB incidence were not yet of quantitative TB infection results combined with finalised at the time of manuscript writing. demographic and clinical factors in multivariable algorithms Added value of this study could be explored in high TB incidence countries. To inform policy, we conducted an individual participant data meta-analysis to compare the predictive performance of TST infection owing to cross-reactivity among individuals In head-to-head comparisons,3,4 previous meta- vaccinated with Mycobacterium bovis bacillus Calm- analyses using aggregated data suggest IGRA’s ette−Guérin vaccines (BCG) or infected by non- increased specificity does not appear to confer greater tuberculous mycobacteria (NTM). IGRA, including prognostic ability to predict individuals who will prog- T-SPOT.TB and QuantiFERON-TB Gold in tube (QFT- ress from infection to disease than the TST. Despite GIT) or Plus (QFT-Plus), use M. tuberculosis-specific IGRA’s higher relative risk estimates for incident TB, antigens, ESAT-6 and CFP-10. Thus, they are not confidence intervals were imprecise and overlapped. In affected by BCG or most NTM (other than Mycobacte- addition, the higher relative risk estimates of IGRA was rium kansasii, marinum and szulgai) and are expected to driven by studies with a high risk of incorporation increase the diagnostic specificity. where these were conducted in low-incidence www.thelancet.com Vol 56 February, 2023 3 Articles resource-rich countries that had widely adopted the test were stored in a data safe haven held by the University and used positive IGRA results to trigger investigations College London. of active TB only in test positives.4 These findings underpinned the initial policy on the use of IGRA for prediction of TB in low TB incidence countries, and Eligibility criteria later updates were used to inform global policy that We adapted the eligibility criteria from our previous recommends the use of either test interchangeably in all systematic review that developed a multivariable model settings regardless of the background TB incidence. to predict the development of active TB, limited to low Most recently, Zhou et al. conducted an updated TB incidence countries (annual incidence ≤20/100,000 meta-analysis, which reported superior predictive per- persons; PROSPERO CRD42018115357). The previous formance of IGRA over TST.5 However, the primary review included TST and IGRA results as model pa- analysis was not restricted to head-to-head analysis, rameters but did not compare their predictive perfor- where participants received both tests. Thus, the results mance. For the present review, we included studies could be affected by the differences in the set of studies from low to high TB incidence settings which system- included for each test. In addition, without individual atically tested participants with both TST and at least participant data, systematic differences between study one commercially available IGRA (paired) and followed settings, study designs, TST cut-off values, participants them for incident TB for the median follow-up duration and follow-up intervals cannot be fully considered in of at least one year in any world region (Table S1 in the meta-analyses using aggregate data. The definitive appendix). evaluation of whether IGRA improves the predictive performance over TST using a robust methodology is Definitions essential for policymakers to decide which test to roll For the primary analysis, incident TB disease was out. This is especially important for low and middle- defined as bacteriologically confirmed or clinically income countries, which are now expanding TB pre- diagnosed TB as per study definitions. In the sensitivity ventive treatment to all household contacts, for whom analysis, we restricted to bacteriologically confirmed TB. testing for TB infection are desirable in light of the latest We excluded cases of TB disease diagnosed within 42 guidelines.1 However, the previous review by Zhou et al. days of enrolment, which were deemed cases of preva- included limited data from those countries.6 lent TB in line with the definition used previously.10 We We conducted a systematic review and individual explored alternative definitions of prevalent TB in participant data (IPD) meta-analysis to directly compare sensitivity analyses (see below). the predictive ability of the TST and IGRAs in both high For TST, we explored multiple cut-off values: ≥5 mm and low TB incidence countries. First, as recommended (TST5 mm), ≥10 mm (TST10 mm), ≥15 mm (TST15 mm), by the WHO framework for evaluation of tests for TB and BCG-stratified (TST5/15 mm), which is ≥15 mm for infection, we estimated the risk for the development of BCG-vaccinees and ≥5 mm for unvaccinated. The re- active TB in participants with positive test results sults of QFT-GIT were defined as per its package compared to negative results for both TST and IGRA.7 insert.11 For T-SPOT.TB results, we classified “border- Second, we compared their predictive ability when us- line positive” and “borderline negative” results as posi- ing quantitative test results. Lastly, we investigated how tive and negative, respectively, as previously.8 No studies the performance differed by setting and population to reported data on QFT-plus. Indeterminate results were explore the best test for specific risk groups. We further excluded for both QFT-GIT and TSPOT.TB. evaluated the predictive performance against the mini- mum targets set by WHO (>75% sensitivity and speci- ficity over a two-year interval).8 Search strategy We identified relevant studies by updating the search conducted by Gupta et al., which included studies Methods published in low TB incidence settings since 1 January We performed a systematic review and IPD meta- 2002 until December 2018.10 We ran the updated search analysis following Preferred Reporting Items for a using Medline and Embase to identify papers published Systematic Review and Meta-analysis of Individual since 1 January 2019. The search strategy included Participant Data (PRISMA-IPD) guidance.9 The protocol related terms for ‘TB’, ‘IGRA’, and ‘TST’ for each search for this review is registered on PROSPERO (https:// engine. The full search strategy for Medline is shown in www.crd.york.ac.uk/prospero/display_record.php?Record Appendix 2. The search was not restricted by language. ID=205667) This study involved analyses of anonymised We supplemented the search by consulting experts in data from previously published cohort studies, with data the field. Additionally, we screened all studies that were pooling via a secure system. Ethical approvals for retained for full-text review by Gupta et al., which sharing of data were sought and obtained by contribu- searched for longitudinal studies assessing the risk of tors of individual participant data, where required. Data progression to TB in individuals tested for TB 4 www.thelancet.com Vol 56 February, 2023 Articles infection.10 While the review and analysis by Gupta et al. Predictive performance of binary test results focused on studies conducted in low TB incidence set- We intended to synthesise all the available evidence, so tings, all relevant studies, regardless of settings, were we collected all the available data rather than performing retained for the full-text review. We also reviewed formal sample size calculations. studies included in previous reviews, including those Our primary effect measures were hazard ratios that informed WHO guidelines.3,5,12 (HR) for the development of TB disease in participants with positive test results compared with those with negative results, for each index test. The use of HR, Study selection, data collection, and quality rather than diagnostic accuracy measures of sensitivity assessment and specificity, allowed accounting for person-time. We Two investigators (YH and RG) independently reviewed conducted a one-stage meta-analysis to estimate pooled titles and abstracts identified through the electronic hazard ratios in our primary analysis using mixed ef- search. Two investigators (YH and AI) screened full-text fects Cox regression model. The model included study- of relevant articles in duplicate. The two reviewers dis- specific intercepts and random slopes for test results by cussed disagreements, and if consensus was not study. The meta-analysis was limited to participants who reached, they were resolved through arbitration by a did not receive TB preventive treatment. We compared third reviewer (RG or MXR). pooled estimates across head-to-head studies by exam- We contacted authors of eligible studies by email to ining point estimates and 95% confidence intervals invite them to contribute to anonymised IPD, including (CIs). We investigated heterogeneity by visualizing for- a set of pre-specified variables (Appendix 3). Where est plots, and we estimated the proportion of variation there is no response, we followed up with them multiple across studies attributable to between-study heteroge- times. We checked IPD for consistency with data re- neity (I2 statistics, by fitting a two-stage model). Addi- ported in study publications and potentially invalid and tionally, to explore the source of heterogeneity, we implausible values. Issues raised were resolved by con- assessed changes in predictive performance by setting tacting the study authors. We treated height, weight, and and population characteristics. body mass index (BMI) which are biologically implau- The analysis produces HR for each test and does not sible as missing, following criteria used in previous give a single value indicating the relative performance of studies.12,13 IPD were mapped to a master variables list, one test VS. the other. Thus, we adopted a method used and variables were standardised across IPD for synthe- in Abubakar et al.13 to make pair-wise comparisons of sis. Two investigators (YH and AI) independently con- the predictive performance of TST with different cut-off ducted a quality assessment of included studies using a values VS. QFT-GIT or T-SPOT.TB. Accordingly, we modified version of the Newcastle–Ottawa Scale (NOS) fitted a mixed effect logistic regression model with a (Appendix 4). Comparability was considered irrelevant positive result as an outcome and an interaction term and thus was not assessed since participants received between test type and the development of TB disease to both tests, and TB risk was compared between test- assess the strength of the association between positive positive and-negative individuals; hence two groups results and TB disease development and how it differs were inherently different. by test type. The model accounted for clustering within individuals and studies. The results from this model were expressed in the ratios of two odds ratios for TST Statistical analysis and IGRA, respectively, for positive test results in par- Handling of missing data ticipants who developed TB disease compared with We excluded participants with missing follow-up dura- those who did not. tions. Most (95%) of them lacked outcome data and We planned to conduct the same analysis in those contributed little information. We assumed Missing at who received TB preventive treatment; however, the Random (MAR) given all the available information was small number of participants given treatment (n = 1838) the most likely missingness mechanism for the and subsequent disease events precluded this analysis. remaining observations; hence, we conducted multiple imputations using multilevel fully conditional specifi- Predictive performance of quantitative test results cation accounting for clustering by study. The models We also compared the predictive ability of TST and included the outcome (TB disease), test results (TST, IGRA using their quantitative results to maximise the QFT-GIT, and TSPOT), variables used for sub-group use of all the available information rather than dichot- analysis (age, TB disease incidence in study countries, omizing them, as previously.10 We transformed the contact history, human immunodeficiency virus infec- quantitative values to a percentile scale using the study tion (HIV) status, and body mass index (BMI)) and data set (see Appendix 5 for a look-up table). The auxiliary variables, including previous BCG vaccination, transformation allowed the comparison of the test re- previous TB, smoking, preventive TB treatment, and sults reported using different scales. Then, we examined interaction terms (See Appendix 1 for details). the association between normalised results for each test www.thelancet.com Vol 56 February, 2023 5 Articles and the risk of incident TB disease. We performed In order to assess the impact of studies that did not mixed-effects Cox proportional hazards models with provide IPD, we combined their aggregated data with those restricted cubic splines with three knots for quantitative from studies with IPD and conducted meta-analysis. We test results at recommended intervals.14 We fitted a one- used data among participants not given TB preventive stage model with study-specific intercepts and random treatment and estimated pooled odds ratio using mixed slopes for test results by study. effects logistic regression models (See Appendix 1). We assessed publication bias using funnel plots and Predictive performance by population and setting and Egger’s test. sensitivity analysis Where possible, we presented sub-group analysis and assessed within-study interactions (see Appendix 1). Role of the funding source We estimated the sensitivity and specificity of tests to The funder of the study had no role in study design, data predict TB disease incidence over two years by setting collection, data analysis, data interpretation, or writing and compared them against the minimum performance of the report. targets of >75% sensitivity and specificity set by WHO (Appendix 1).8 To explore the clinical impact, we pre- sented positive and negative predictive values (PPV and Results NPV, respectively) in hypothetical populations with Characteristics of included studies different pre-test probabilities of developing TB disease. From 1018 titles and abstracts screened, we identified We repeated the analyses: (1) using a two-stage meta- 38 eligible studies, for which IPD were sought. IPD analysis fitting Poisson regression models to estimate were obtained from 11 studies13,15–24 while two additional incident rate ratios for TB; (2) using shorter (<14 days) datasets were obtained through other sources and and longer (<180 days) temporal definitions of prevalent included (Fig. S1).25,26 IPD could not be obtained from TB disease; and (3) including only incident TB disease 27 studies27–53 (n = 20,333) most commonly because of (<42 days) events with microbiological confirmation. non-response (n = 20) (Fig. 1). Characteristics of those Furthermore, we repeated sub-group analysis by TB studies are presented in Table S2. Most of them (24/27) incidence by excluding HIV-positive individuals to were conducted in countries with TB incidence rate assess the impact of confounding by HIV status. <100 per 100,000 population, mainly including adults. Fig. 1: Study selection. IPD: individual participant data; TPT: TB preventive treatment; LTBI; latent tuberculosis infection. 6 www.thelancet.com Vol 56 February, 2023 Articles The 13 included study datasets comprised 33,093 for TST5/15 mm; however, its confidence interval was participants so that while only 33% of studies were wide with a substantial overlap with that of TST5/15 mm included, these included 62% of the potential partici- (Table 2). The two-stage meta-analysis showed similar pants. Among these, we excluded 141 participants who point estimates in the predictive performance between had prevalent TB disease, 917 participants with missing TST15 mm and QFT-GIT (HR 2.76, 95% CI 1.67–4.58 follow-up durations, and one with negative follow-up VS. HR 3.03 95% CI 1.71–5.35) (Fig. S3). This differ- duration, leaving 32,034 participants for subsequent ence in the point estimates was driven by one study analysis. Among 917 participants with missing follow- (n = 1414) in Germany,19 in which all 19 TB cases had up duration, 871 (95.0%) had missing data on the inci- positive QFT-GIT results; thus, HR could not be directly dence of TB disease, which provides little information estimated. Without this study, the one-stage meta-anal- about the predictive performance of TST VS. IGRA. ysis estimate for the HR reduced slightly to 3.12 (95% The median (IQR) follow-up duration was 3.6 CI 1.75–5.57) for QFT-GIT, which was similar to HR (2.0–5.6) years; 518 (1.6%) participants developed inci- 2.92 for TST15 mm (95% CI 1.73–4.92). Similarly to TST, dent TB disease (Table 1). The median (IQR) time to the the HR of QFT-GIT varied widely from 1.31 to 46.90 development of TB disease was 1.1 years (0.5–1.7). The (tau2 = 0.71, I2 = 76.1, 95% CI 58.1–86.3) (Fig. S2). In median time to the development of TB was slightly the study of Lu et al. conducted in China, there was the longer in studies where TB incidence rate was ≥100 per largest difference in the point estimate between QFT- 100,000 than that in the other studies (1.20 VS. 0.90 years). GIT and TST15 mm (HR 7.82; 95% CI 2.36–25.98 for Among participants, 39%, 8.9%, and 14.8% were screened QFT-GIT VS. 3.99; 95% CI 1.27–12.56 for TST15 mm).17 for TB infection because of contact with TB, immuno- In the pair-wise comparisons of the test perfor- suppression, and migration from high TB incidence mance, the ratio of odds ratios for QFT-GIT VS. TST15 mm countries, respectively. Besides, 19% were general adoles- was 1.48 (95% CI 1.03–2.13), suggesting a slightly higher cents in South Africa15 and 17% were general adults in predictive performance of QFT-GIT than TST15 mm China.17 Among all participants, 7.8% were HIV-positive. (Table S6). This was again driven by the same one study Only 1% (n = 396) of the participants were children less as above,19 and when it was excluded, the ratio declined, than five years old. Participants from European countries and its confidence interval crossed one (1.23; 95% accounted for 44.3% of the population, followed by African CI 0.84–1.80). Likewise, the ratio for QFT-GIT VS. countries (South Africa and Zambia, 28.2%) (Tables S3 TST5 mm/15 mm based on all studies was 1.14 (95% CI and S4). Data on TST and QFT-GIT were available in 0.72–1.81). 27,400 participants (85.5%, 13 studies)13,15–26 and TST and We compared the predictive performance of TST VS. TSPOT.TB in 8612 participants (26.9%, three studies).13,16,18 T-SPOT.TB, using individual data from three studies None of the studies blinded results of TB infection tests for with data on T-SPOT.TB.13,16,18 The predictive perfor- diagnosing incident TB. mance was similar between TST15 mm, TST5/15 mm, and Eleven studies achieved a NOS quality score of 7/8, T-SPOT.TB (Table 2 and Fig. S4). The pooled HRs for and the remaining two achieved 6/8 due to unclear TST15 mm and TST5/15 mm were 7.87 (95% CI methods for excluding prevalent TB disease at the 3.42–18.12) and 5.98 (95% CI 3.51–10.21), respectively, baseline (Table S5). compared to that of 6.45 (95% CI 3.32–12.55) for T- SPOT.TB, and 8.98 (95% CI 3.17–25.41) for QFT-GIT. Because of the small number of studies and events, Predictive performance of binary test results the pooled estimates from the two-stage meta-analysis Figure S1 presents cumulative TB incidence curves by were very wide (Fig. S5). test results for each test, which showed a higher risk of The sensitivity analyses that used only bacteriologi- TB in participants with positive test results. The point cally confirmed TB and those using a shorter (within 14 estimates for TST predictive performance tended to be days from enrolment) or longer (within 180 days) tem- higher when a higher cut-off value was used for posi- poral definition of prevalent TB disease did not change tivity (Table 2). The pooled HR was 2.71 (95% CI the overall trends significantly (Tables S7 and S8). 1.63–4.50) at TST15 mm, 2.48 (95% CI 1.57–3.92) for Egger’s test did not show the evidence of publication TST10 mm, and 2.30 (95% CI 1.45–3.64) at TST5 mm. bias (Fig. S6). When TST cut-off was stratified by BCG status (i.e., TST5/15 mm), the HR was 2.88 (95% CI 1.69–4.90). There was great variability in effect estimates by study. For Predictive performance of quantitative test results example, for TST15 mm, the HR ranged from 0.92 in a Fig. 2 presents changes in HR by quantitative test re- study in India among household contacts to 30.29 in sults relative to participants with test results at the 0th household contacts in Spain (tau2 = 0.52, I2 = 73.1%; percentile (i.e., no induration for TST and 0 quantitative 95% CI 52.1–84.9%) (Fig. S2). values for IGRA). Across all three tests, HR appeared to The pooled HR of QFT-GIT was 4.15 (95% CI increase as the quantitative test results increase. For 1.97–8.75) for incident TB, numerically higher than that example, in participants with TST induration size at the www.thelancet.com Vol 56 February, 2023 7 Articles Variables Missing data n (%) Follow-up duration (median years (IQR)) 3.6 years (2.0–5.6) Age (mean (SD)) 33.05 (18.27) 68 (0.1) Female 16,936 (52.9) 73 (0.3) Reason for screening (%) 102 (0.3) Contact 12,564 (39.2) Immunosuppression 2862 (8.9) Migrant 4739 (14.8) Othersa 11,767 (36.7) Previous BCG vaccination (%) 19,372 (63.3) 5166 (16.1) Previous TB diagnosis (%) 1855 (5.8) 9909 (30.9) HIV positive (%) 2497 (7.8) 7559 (23.6) BMI (mean (SD)) 24⋅00 (5.52) 14,114 (44.1) TB preventive treatment (%) 1883 (5.9) 859 (2.7) Study country (%) 0 (0) Brazil 890 (2.8) China 5404 (16.9) Europe 14,184 (44.3) India 2519 (7.9) South Africa or Zambia 9037 (28.2) QFT-GIT (%) 1994 (6.2) Indeterminate 462 (1.4) Negative 19,210 (60.0) Positive 10,368 (32.4) TSPOT.TB (%) 17,006 (63.9) Borderline 433 (1.4) Indeterminate 232 (0.7) Negative 7083 (22.1) Positive 1876 (5.9) TST 3205 (12.0) ≥5 mm 14,589 (45.5) ≥10 mm 10,858 (33.9) ≥15 mm 6124 (19.1) SD: standard deviation; IQR: interquartile range; BCG: bacillus Calmette−Guérin; HIV: human immunodeficiency virus; QFT-GIT: QuantiFERON Gold in Tube; BMI: body mass index; TST: tuberculin skin test. aOthers include HIV-negative adolescents (Mahomed 2011) and the general adults (Lu 2021). Table 1: Characteristics of study participants (N = 32,034). 50th, 66th, and 83rd percentile, which corresponds to Similarly, in participants with QFT-GIT results at the the induration sizes of 5 mm, 10 mm, and 15 mm, same percentile scales, which corresponds to the values respectively, the HRs were 1.64 (95% CI 0.48–5.68), 2.54 of 0.09–0.099 IU/ml, 0.62–0.699 IU/ml, and 3.77–4.269 (95% CI 0.63–10.29), and 4.52 (95% CI 0.95–21.36). IU/ml, HRs were 2.52 (95% CI 0.72–8.78), 4.00 (95% CI Test All studies (n = 13) Studies with TSPOT.TB (n = 3) HRa (95% CI) I2 (95% CI) HRa (95% CI) I2 (95% CI) TST5 mm 2.3 (1.45–3.64) 56.1 (16.2–77) 5.11 (3.16–8.25) 0 (0–89.6) TST10 mm 2.48 (1.57–3.92) 62.2 (29.3–79.8) 5.24 (3.4–8.06) 0 (0–89.6) TST15 mm 2.71 (1.63–4.5) 73.1 (52.1–84.9) 7.87 (3.42–18.12) 71.1 (1.8–91.5) TST5/15 mmb 2.88 (1.69–4.9) 64.8 (34.8–81) 5.98 (3.51–10.21) 63 (0–89.4) QFT-GIT 4.15 (1.97–8.75) 76.1 (58.1–86.3) 8.98 (3⋅17–25⋅41) 75.2 (18–92.5) T-SPOT.TB – – 6.45 (3⋅32–12⋅55) 0 (0–89.6) HR: hazard ratio; CI: confidence interval; QFT-GIT: QuantiFERON Gold in Tube; TST: tuberculin skin test. When one study (Diel 2011) where HR could not be estimated directly was excluded since all TB cases had positive QFT-GIT, HR was 2.19 (95% CI 1.36–3.53) for TST5 mm, 2.49 (95% CI 1.50–4.15) for TST10 mm, 2.90 (95% CI 1.72–4.9) for TST15 mm, 2.64 (95%CI 1.52–4.59) for TST5/15 mm, and 3.09 (95% CI 1.74–5.50) for QFT-GIT. aThe risk of TB relative to participants with negative test results. bTST5/ 15 mm indicates BCG-stratified cut-off, which is ≥15 mm for BCG-vaccinees and ≥5 mm for unvaccinated. Table 2: Pooled estimates of the predictive performance of TST VS. IGRA for all TB. 8 www.thelancet.com Vol 56 February, 2023 Articles Fig. 2: The association between quantitative results and the risk of TB. (a) TST and QFT-GIT. (b) TST and TSPOT-TB. TB: tuberculosis; HR: Hazard ratio; QFT-GIT: QuantiFERON TB Gold In-Tube; TST: tuberculin skin test. Quantitative results of TST, QFT-GIT, and TSPOT-TB are normalized to a percentile scale. HR indicates the risk of TB incidence with participants who have results at zero percentile scale as a reference group (i.e. HR = 1). The areas around the lines indicate 95% confidence intervals. 0.99–16.16), and 6.95 (95% CI 1.49–32.4). For TST VS. WHO minimum targets (>75% sensitivity and speci- T-SPOT.TB, since only three studies had data, the con- ficity). In low TB incidence countries, the performance fidence intervals were wide; no significant difference was closer to the targets than in high TB incidence was identified, although HR for TST remained higher countries. In high TB incidence countries, the specificity than T-SPOT.TB throughout the percentile scales. for TST5 mm was lower than in low TB incidence countries (44% VS. 51%), and so was the sensitivity Sub-group analysis (60% VS. 83%). When we stratified data into two groups based on the TB Based on the sensitivity and specificity estimates incidence rate of the study country, the pooled predictive stratified by TB incidence, we estimated PPVs and NPVs performance was higher among individuals in studies in a hypothetical population of 10,000 individuals conducted in countries with TB incidence rates <100 per (Table S11). The median cumulative TB incidences were 100,000 population (Brazil, China and European coun- 0.95% and 5.2% for countries with TB incidence rate tries). For example, the HR for TST15 mm was 4.43 (95% <100 and ≥100 per 100,000 population, respectively; CI 2.24–8.77) in countries with TB incidence rate <100 thus, pre-test probability of TB incidence was assumed per 100,000 population compared to 1.78 (95% CI to be 1% and 5% over two years for each setting. In 1.18–2.66) in countries with TB incidence rate ≥100 per countries with TB incidence rate <100 per 100,000, PPV 100,000 population (India, South Africa and Zambia), were numerically similar between TST15 mm (2.75%) respectively (Fig. 3). In countries with TB incidence rate and QFT-GIT (2.46%) and so were NPV (99.42% VS. <100 per 100,000 population, the point estimate for 99.52%), even though the overall performance (i.e. PPV/ QFT-GIT was substantially higher (HR 10.38; 95% CI [100−NPV]) was slightly higher for QFT-GIT. In coun- 4.17–25.87) with a wide confidence interval. The dif- tries with TB incidence rate ≥100 per 100,000, the PPV ferences in the performance by setting were statistically for TST15 mm was higher than for QFT-GIT (7.54% VS. significant (Table S9). 6.04%). NPV was lower for TST15 mm than for QFT-GIT The exclusion of HIV-positive participants replicated (95.66% VS. 96.26%). the higher predictive performance in countries with low All three studies using T-SPOT.TB were from Eu- TB incidence than those with high TB incidence ropean countries with TB incidence rate <100 per observed in the primary analysis (Fig. S7). 100,000 populations; thus, sub-group analysis was not Figs. S8, S9, and Table S10 present the sensitivities possible. and specificities for predicting the development of active The predictive performance for TST tended to be TB over two years. None of the performances reached lower in participants of older age (Table S9, Fig. S10). www.thelancet.com Vol 56 February, 2023 9 Articles Fig. 3: Predictive performance of TST VS. QFT-GIT for all TB, by TB incidence in study countries. TB: tuberculosis; HR: hazard ratio; CI: confidence interval; QFT-GIT: QuantiFERON Gold in Tube; TST: tuberculin skin test. HR indicates the risk of TB is participants with positive test results relative to those with negative results. In the original dataset before multiple imputation, studies in countries with TB incidence rate <100/100,000 and ≥100/100,000 population included 19,242 and 10,050 participants who did not receive TB preventive treatment, respectively. The analysis was performed using multiply imputed datasets in which missing data on the receipt of TB preventive treatment (n = 859) were imputed. When one study (Diel 2011) where HR could not be estimated directly was excluded from studies in TB incidence rate <100 per 100,000 populations, HR was 4.83 (95% CI 3.16–7.38) for TST 5 mm, 5.05 (95% CI 3.4–7.41) for TST10 mm, 5.86 (95% CI 4.17–8.22) for TST15 mm, 5.13 (95% CI 3.58–7.35) for TST5/15 mm, and 7.18 (95% CI 4.48–11.51) for QFT-GIT. Hazard ratios were 2–3% lower for each 10-year in- overall population. When stratified by national TB inci- crease in age. The same trend was not observed for dence, the difference was pronounced in low TB inci- QFT-GIT. This was in part driven by a study among dence countries, yet with wide confidence intervals. This household contacts in India (n = 1510),24 in which the is consistent with observations in previous head-to-head HR for TB in children with positive QFT-GIT results meta-analyses which showed higher point estimates for was lower (HR 1.25, 95% CI 0.59–2.66) than that in QFT-GIT but with overlapping wide CIs.5,20 The higher adults (HR 2.46, 95% CI 1.18–5.15). In the same study, estimates observed in this IPD analysis were driven by a more children who did not develop TB disease had single study in a low TB incidence setting. In low TB positive results for QFT-GIT (62.4%) than for TST10 mm incidence settings, test differences between PPV and (45.6%), suggesting a lower specificity of QFT-GIT in NPV were minimal. This can be explained by two fac- this population. tors. First, studies reporting high predictive perfor- We did not find a difference in predictive perfor- mance (e.g. Diel et al.) had less impact on the pooled mance by BMI (Table S9). The estimates for the dif- estimates when sensitivity and specificity estimates are ference in predictive performance were imprecise with pooled because they are bounded between 0 and 100%. very wide confidence intervals for contact history Second, pre-test probability for developing TB is low in (Table S8 and Fig. S11), or HIV status (Table S9). Six low TB incidence settings. The minimal difference in studies included both HIV-positive and negative PPV and NPV suggests that the higher performance individuals,13,15,16,18,25,26 but in four of them,13,15,18,26 HIV- does not appear to translate into a significant difference positive individuals accounted for only <2%. There in clinical impact. On the other hand, the performance were no TB incident cases among HIV-positive in- of TST and QFT-GIT was similar in high TB incidence dividuals in one of the two remaining studies.25 Full countries. These findings support the current WHO results of the one-stage meta-analysis estimating inter- guidelines that recommend using TST and IGRA action effects are presented in Table S9 and Table S12. interchangeably in instances where TB infection testing Aggregated data were most commonly available for is recommended. Our IPD also found that the predictive TST10 mm and QFT-GIT 10 studies, performance of both tests was higher in low TB inci- Fig. S12. 34–36,42,44–46,48–50 When they were pooled with dence countries—and approaching WHO targets of seven studies from low TB incidence countries that ≥75% for sensitivity (77% for TST5/15 mm and 73% for provided IPD, the pooled odds ratios were 4.20 (95% CI QFT-GIT), but not specificity (62% for TST5/15 mm and 2.71–6.49) for TST10 mm and 12.00 (95% CI 4.46–32.30) 65% for QFT-GIT)—and was lower in countries with for QFT-GIT (Fig. S13). high TB incidence. The most recent review by Zhou et al. reported that the predictive performance of IGRA was almost twice as Discussion high as TST (pooled risk ratio of 9.35 VS. 4.25).5 All but Our IPD meta-analysis found that QFT-GIT appeared to one study included in the review were conducted in have higher predictive performance than the TST in the countries with TB incidence rate 100 < 100,000 10 www.thelancet.com Vol 56 February, 2023 Articles populations. The higher predictive performance for Our review intended to conduct a priori defined IGRA in Zhou et al. was consistent with our estimates subgroup analyses. We found that the predictive per- in low TB incidence countries despite the limitations in formance for TST tended to be higher in younger pop- the previous review, such as the inclusion of non-head- ulations. This might reflect that in younger populations, to-head studies in the primary analysis, heterogeneity in positive results can be better explained by recent infec- TST cut-off values, and the use of effect measures not tion. Interestingly, the same trend was not observed for accounting for the follow-up duration.6 However, the QFT-GIT. The difference in the trend might be IPD approach shows that the effect estimates remain explained by the declining sensitivity of TST in older imprecise, and the added prognostic advantage of IGRA adults reported in previous studies.57 On the other hand, over the TST may still be small. the lack of the trend for QFT-GIT was affected by a In contrast to previous aggregated-data meta-ana- single study in which the test had a lower predictive lyses, we were able to assess the difference in the pre- performance in children than adults.24 However, given dictive performance by background TB incidence, using the small number of events and multiple analyses, this the standardised TST cut-off values and the definition of observation may be a chance finding. Furthermore, TB, which showed a lower predictive performance in since we could not fit multivariable models due to high TB incidence countries. Differences in TB epide- insufficient sample size and event rate, and we did not miology could explain this finding. First, in countries derive a specific causal model, the presence of con- with high TB incidence, pre-test probability of devel- founding might also explain these findings. Other sub- oping TB is higher than in countries with low TB inci- group analyses had low statistical power, and some risk dence, which results in lower NPV even if the sensitivity groups—such as individuals with co-morbidities (e.g. and specificity of the tests do not differ. Moreover, more diabetes, autoimmune diseases, or HIV), of advanced people with negative test results at baseline may get age, or receiving immunosuppressants or chemotherapy infection and develop TB, which lowers the sensitivity. —were not considered due to limited data, currently In that case, the actual performance to predict progres- limiting our understanding of the impact of those sion to active TB from infection might be higher than covariates. Likewise, a low sample size of individuals estimated. Second, in high TB incidence countries, <18 years old precluded disaggregation of individuals people would have more chances to be infected early in into children and adolescents in the sub-groups anal- their life before they are identified as contacts. As a ysis. About a third of the study proportion were in- result, a large subset of positive results at baseline may dividuals who do not belong to specific risk groups reflect the evidence of remote exposure rather than (adolescents in South Africa and general adults in recent infection that is likely to progress. This lowers the China), who are at lower risk for disease progression. specificity and predictive performance of tests for TB The performance is likely to differ if restricted to those infection. The poorer performance in high TB incidence at high risk. In addition, we did not consider the impact countries calls for an urgent need for a better test or of socioeconomic status (e.g. living environment and multivariable clinical algorithm for predicting the education level) due to limited availability and difficulty development of TB in settings which have the greatest in standardizing those variables. burden of disease.54 While mRNA signatures have None of the included studies blinded results of TB shown promising results in early studies,55,56 they have infection tests for diagnosing incident TB. Thus, the suboptimal performance for long-term prediction over knowledge of TB infection test results might have two years and are not yet available commercially. In the affected the ascertainment of incident TB and then meantime, the role of TB infection tests needs to be overestimated the predictive performance (i.e. incorpo- carefully considered in light of the trade-off between ration bias). This impact might be stronger for IGRA if false positives and negatives and their clinical and public clinicians believed that IGRA was more accurate or health impact. For instance, preventive treatment could there was differential work-up for positive results than be given without tests in populations who are at very negative ones. high risk for TB, especially those at high risk for severe The median (IQR) follow-up duration in our study diseases such as young children and people living with was 3.6 (2.0–5.6) years, which may have not been long HIV, as already recommended by WHO.1 On the con- enough for assessing incident TB. Nevertheless, WHO’s trary, tests can reduce the number of people who are targets for tests for predicting TB development are set started on preventive treatment and reduce opportunity for two-year time horizon, given that the risk is highest costs. This would be particularly important in pop- shortly after infection.58 We were able to compare the ulations who are at risk for adverse events. Furthermore, performance of TST and IGRA with these targets. consideration of quantitative results as well as socio- We could not obtain data from 27 of 40 studies that demographic and clinical factors might enable a more were considered eligible. Nonetheless, we were able to accurate estimate of the risk of TB and individualised include data from 32,034 participants, more than the management, currently being done in high-income, 20,333 from the remaining studies, corresponding to low-incidence contexts.10,54 >60% of the total participants. The sample size is the www.thelancet.com Vol 56 February, 2023 11 Articles largest for a head-to-head analysis of TST VS. IGRA, consultation service to INSMED and received speakers honoraria from representing both high and low TB incident settings, INSMED, GILEAD, JANSSEN and is a member of the Data Safety Board of trials from Medicines sans Frontiers, all outside of the submitted compared to 4875, mostly from low TB incidence work. ILR has a patent (PCT/EP2019/064885), in vitro method for the countries, included in the most recent review.5 diagnosis or detection of non-tuberculous mycobacteria. MS reports Furthermore, our sensitivity analysis that included receipt of test kits free of charge from Qiagen and Oxford Immunotec. aggregated data from studies without IPD consistently MS also reports receipt of research grants from Biotest and Astellas, showed a higher predictive performance for IGRA than consulting fees and honoraria from Biotest, Moderna, Qiagen, and Takeda, support for travel from Biotest, and participation on advisory TST in low TB incidence countries. board for Biotest and Moderna, all outside of this work. GS reports Lastly, we are aware of other IGRA kits than QFT- receipt of consulting fees from Pfizer, Diasorin, and INSMED. All other GIT and TSPOT.TB.59 We did not include them due to authors declare no competing interests. lack of data on their predictive performance. Some of those tests have been suggested to give results that are highly concordant with the existing tests.59 For those Acknowledgments tests that have a high level of agreement with QFT-GIT YH, MXR, and IA were supported by the National Institute for Health and/or TSPOT.TB, the results of our analysis are likely and Care Research (NIHR), UK (RP-PG-0217-20009). RKG was sup- ported by NIHR (DRF-2018-11-ST2-004). MQ was supported by the to be applicable. Medical Research Council [MC_UU_00004/07], UK. AG was supported Our IPD meta-analysis found that IGRA appeared to by the US NIH. CL was supported by the German Center of Infection have higher performance than TST for predicting the Research. SF was supported by Bill and Melinda Gates Foundation and development of TB disease in countries with low TB Medical Research Council, UK. incidence. However, the difference was driven by a single study and does not appear to make a large dif- Appendix A. Supplementary data ference in clinical impact. Both tests had performance Supplementary data related to this article can be found at https://doi. close to minimum WHO targets in low TB incidence org/10.1016/j.eclinm.2022.101815. countries, whereas performance was poorer in high TB incidence countries. This reinforces the urgent need for References a better test to predict the development of TB disease. 1 World Health Organization. WHO consolidated guidelines on tuber- While we wait for a new test, test choice needs to be culosis: module 1: prevention: tuberculosis preventive treatment. made considering operational and clinical impact, and Geneva, Switzerland: WHO; 2020. 2 Mack U, Migliori GB, Sester M, et al. LTBI: latent tuberculosis alternative strategies such as the use of quantitative re- infection or lasting immune responses to M. tuberculosis? A sults combined with demographic and clinical factors TBNET consensus statement. Eur Respir J. 2009;33(5):956–973. 3 Latent tuberculosis infection: updated and consolidated guidelines for should be explored. programmatic management. Geneva, Switzerland: WHO; 2018. 4 Rangaka MX, Wilkinson KA, Glynn JR, et al. Predictive value of interferon-gamma release assays for incident active tuberculosis: a Contributors systematic review and meta-analysis. Lancet Infect Dis. YH, RKG, MQ, IA, and MXR designed the study and protocol and 2012;12(1):45–55. interpreted the results. YH, RKG, and AI did the systematic review. 5 Zhou G, Luo Q, Luo S, et al. Interferon-gamma release assays or CAV, NA, RD, JD, SF, AG, HH, ECJ, AK. CL, FvL, QL, WL, PL, ILR, tuberculin skin test for detection and management of latent LMa, SKM, LMu, ESP, MP, TS, MS, KS, SKS, RS, GS, KT, RV, and IA tuberculosis infection: a systematic review and meta-analysis. contributed data to the meta-analysis. YH analysed the data with assis- Lancet Infect Dis. 2020;20:1457. tance from RKG and QM. YH wrote the first draft of the manuscript, 6 Hamada Y, Gupta RK, Matteelli A, Abubakar I, Rangaka MX. which was revised based on comments from co-authors. YH and RKG Predictive performance of interferon-gamma release assays and tuberculin skin tests. Lancet Infect Dis. 2020;20(12):1371–1372. accessed and verified the data. All authors approved the final version of 7 Hamada Y, den Boon S, Cirillo DM, et al. Framework for the the manuscript. evaluation of new tests for tuberculosis infection. Eur Respir J. 2021;58(2):2004078. 8 World Health Organization. Consensus Meeting Report: development Data sharing statement of a Target Product Profile (TPP) and a framework for evaluation for a The IPD database will be stored within the UCL Data Repository and can test for predicting progression from tuberculosis infection to active disease. be shared subject to the approval of the corresponding authors of the Geneva, Switzerland: WHO; 2017. original studies. 9 Stewart LA, Clarke M, Rovers M, et al. Preferred reporting items for systematic review and meta-analyses of individual participant data: the PRISMA-IPD statement. JAMA. 2015;313(16):1657–1665. 10 Gupta RK, Calderwood CJ, Yavlinsky A, et al. Discovery and vali- Declaration of interests dation of a personalized risk predictor for incident tuberculosis in YH and MXR report donation of QIAreach, an IGRA, by QIAGEN for a low transmission settings. Nat Med. 2020;26(12):1941–1949. LTBI infection survey. YH, MXR, and IA report donations of Cy-TB, a 11 QuantiFERON-TB gold ELISA package insert. https://www. TB-specific skin test for detection of LTBI, by the Serum Institute of quantiferon.com/wp-content/uploads/2019/03/L1075115-QuantiFERON- India for a study on the feasibility and patient-important outcomes. They TB-Gold-ELISA-IFU-CE-rev07.pdf. 12 Getahun H, Matteelli A, Abubakar I, et al. Management of latent had no role in the submitted work. RD declared the receipt of payment Mycobacterium tuberculosis infection: WHO guidelines for for lectures from Qiagen and Oxford Immunotec. JD declared honoraria low tuberculosis burden countries. Eur Respir J. 2015;46(6): for lectures from Oxford Immunotec. AG declared receipt of research 1563–1576. grants from the US National Institute of Health (NIH) and membership 13 Abubakar I, Drobniewski F, Southern J, et al. Prognostic value of in NIH Council and IndoUS Science Technology Forum. CL provided interferon-γ release assays and tuberculin skin test in predicting the 12 www.thelancet.com Vol 56 February, 2023 Articles development of active tuberculosis (UK PREDICT TB): a prospec- anti-TNF-alfa treatment. Int J Tuberc Lung Dis. 2012;16(10):1300– tive cohort study. Lancet Infect Dis. 2018;18(10):1077–1087. 1306. 14 Harrell FE. Biostatistical modeling. http://biostat.mc.vanderbilt. 34 Kim C, Jung H, Kim K, et al. Comparison of quantiferon-TB gold edu/wiki/pub/Main/BioMod/notes.pdf; 2004. with tuberculin skin test for detecting latent tuberculosis infection 15 Mahomed H, Hawkridge T, Verver S, et al. The tuberculin skin test before kidney transplantation. Am J Transplant. 2013;13:341. versus QuantiFERON TB Gold® in predicting tuberculosis disease 35 Kruczak K, Duplaga M, Sanak M, et al. Comparison of IGRA tests in an adolescent cohort study in South Africa. PLoS One. 2011;6(3): and TST in the diagnosis of latent tuberculosis infection and pre- e17984. dicting tuberculosis in risk groups in Krakow, Poland. Scand J Infect 16 Sester M, van Leth F, Bruchfeld J, et al. Risk assessment of Dis. 2014;46(9):649–655. tuberculosis in immunocompromised patients. A TBNET study. 36 Lee SS, Chou KJ, Su IJ, et al. High prevalence of latent tuberculosis Am J Respir Crit Care Med. 2014;190(10):1168–1176. infection in patients in end-stage renal disease on hemodialysis: 17 Lu P, Liu Q, Zhou Y, et al. Predictors of discordant tuberculin skin comparison of QuantiFERON-TB GOLD, ELISPOT, and tuberculin test and QuantiFERON-TB gold in-tube results in eastern China: a skin test. Infection. 2009;37(2):96–102. population-based, cohort study. Clin Infect Dis. 2021;72(11):2006– 37 Lee YM, Lee SO, Choi SH, et al. A prospective longitudinal 2015. study evaluating the usefulness of the interferon-gamma 18 Altet N, Dominguez J, Souza-Galvao M-LD, et al. Predicting the releasing assay for predicting active tuberculosis in allogeneic development of tuberculosis with the tuberculin skin test and hematopoietic stem cell transplant recipients. J Infect. QuantiFERON testing. Ann Am Thorac Soc. 2015;12(5):680–688. 2014;69(2):165–173. 19 Diel R, Loddenkemper R, Niemann S, Meywald-Walter K, 38 Leung CC, Yam WC, Ho PL, et al. T-Spot.TB outperforms tuber- Nienhaus A. Negative and positive predictive value of a whole-blood culin skin test in predicting development of active tuberculosis interferon-γ release assay for developing active tuberculosis: an among household contacts. Respirology. 2015;20(3):496–503. update. Am J Respir Crit Care Med. 2011;183(1):88–95. 39 Leung CC, Yam WC, Yew WW, et al. T-Spot.TB outperforms tu- 20 Rangaka MX, Wilkinson RJ, Boulle A, et al. Isoniazid plus antire- berculin skin test in predicting tuberculosis disease. Am J Respir troviral therapy to prevent tuberculosis: a randomised double-blind, Crit Care Med. 2010;182(6):834–840. placebo-controlled trial. Lancet. 2014;384(9944):682–690. 40 Mathad JS, Bhosale R, Balasubramanian U, et al. Quantitative IFN- 21 Acuna-Villaorduna C, Jones-Lopez EC, Fregona G, et al. Intensity of gamma and IL-2 response associated with latent tuberculosis test exposure to pulmonary tuberculosis determines risk of tuberculosis discordance in HIV-infected pregnant women. Am J Respir Crit infection and disease. Eur Respir J. 2018;51(1). Care Med. 2016;193(12):1421–1428. 22 Muñoz L, Gomila A, Casas S, et al. Immunodiagnostic tests’ pre- 41 McCarthy KM, Scott LE, Gous N, et al. High incidence of dictive values for progression to tuberculosis in transplant re- latent tuberculous infection among South African health cipients: a prospective cohort study. Transplant Direct. workers: an urgent call for action. Int J Tuberc Lung Dis. 2015;1(3):e12–e. 2015;19(6):647–653. 23 Huerga H, Sanchez-Padilla E, Melikyan N, et al. High prevalence of 42 Noorbakhsh S, Mousavi J, Barati M, et al. Evaluation of an infection and low incidence of disease in child contacts of patients interferon-gamma release assay in young contacts of active tuber- with drug-resistant tuberculosis: a prospective cohort study. Arch culosis cases. East Mediterr Health J. 2011;17(9):714–718. Dis Child. 2019;104(7):622–628. 43 Pullar ND, Steinum H, Bruun JN, Dyrhol-Riise AM. HIV patients 24 Sharma SK, Vashishtha R, Chauhan LS, Sreenivas V, Seth D. with latent tuberculosis living in a low-endemic country do not Comparison of TST and IGRA in diagnosis of latent tuberculosis develop active disease during a 2 year follow-up; a Norwegian infection in a high TB-burden setting. PLoS One. 2017;12(1). prospective multicenter study. BMC Infect Dis. 2014;14:667. e0169539-e. 44 Ringshausen FC, Nienhaus A, Schablon A, Schlosser S, Schultze- 25 Shanaube K. The association between the magnitude of T-cell Werninghaus G, Rohde G. Predictors of persistently positive interferon-gamma responses to Mycobacterium tuberculosis spe- Mycobacterium-tuberculosis-specific interferon-gamma responses cific antigens and risk of progression to tuberculosis in household in the serial testing of health care workers. BMC Infect Dis. contacts tested with QuantiFERON-TB Gold In-Tube Assay. PhD 2010;10:220. thesis, London School of Hygiene & Tropical Medicine https://doi. 45 Debulpaep S, Corbière V, Levy J, et al. Contribution of org/10.17037/PUBS.01856014; 2014. QuantiFERON-TB gold-in-tube to the diagnosis of Mycobacterium 26 Paradkar M, Padmapriyadarsini C, Jain D, et al. Tuberculosis pre- tuberculosis infection in young children in a low TB prevalence ventive treatment should be considered for all household contacts country. Front Pediatr. 2019;7:291. of pulmonary tuberculosis patients in India. PLoS One. 2020;15(7): 46 Seyhan EC, Gunluoglu G, Gunluoglu MZ, Tural S, Sökücü S. e0236743. Predictive value of the tuberculin skin test and QuantiFERON- 27 Ahmed A, Feng P-JI, Gaensbauer JT, et al. Interferon-gamma tuberculosis Gold In-Tube test for development of active tubercu- release assays in children <15 years of age. Pediatrics. losis in hemodialysis patients. Ann Thorac Med. 2016;11(2):114– 2020;145(1):e20191930. 120. 28 Bergot E, Haustraete E, Malbruny B, Magnier R, Salaun M-A, 47 Song SE, Yang J, Lee KS, et al. Comparison of the tuberculin skin Zalcman G. Observational study of QuantiFERON-TB gold in-tube test and interferon gamma release assay for the screening of assay in tuberculosis contacts in a low incidence area. PLoS One. tuberculosis in adolescents in close contact with tuberculosis TB 2012;7(8):e43520. patients. PLoS One. 2014;9(7):e100267. 29 Bourgarit A, Baron G, Breton G, et al. Latent tuberculosis infection 48 Torres Costa J, Silva R, Ringshausen FC, Nienhaus A. Screening screening and 2-year outcome in antiretroviral-naive HIV-infected for tuberculosis and prediction of disease in Portuguese healthcare patients in a low-prevalence country. Ann Am Thorac Soc. workers. J Occup Med Toxicol. 2011;6(1):19. 2015;12(8):1138–1145. 49 Tsou P-H, Huang W-C, Huang C-C, et al. Quantiferon TB-Gold 30 Dorman SE, Belknap R, Graviss EA, et al. Interferon-gamma conversion can predict active tuberculosis development in elderly release assays and tuberculin skin testing for diagnosis of latent nursing home residents. Geriatr Gerontol Int. 2015;15(10):1179– tuberculosis infection in healthcare workers in the United States. 1184. Am J Respir Crit Care Med. 2014;189(1):77–87. 50 Verhagen LM, Maes M, Villalba JA, et al. Agreement between 31 Harstad I, Winje BA, Heldal E, Oftung F, Jacobsen GW. Predictive QuantiFERON-TB Gold In-Tube and the tuberculin skin test and values of QuantiFERON-TB gold testing in screening for tubercu- predictors of positive test results in Warao Amerindian pediatric losis disease in asylum seekers. Int J Tuberc Lung Dis. tuberculosis contacts. BMC Infect Dis. 2014;14:383. 2010;14(9):1209–1211. 51 Yang CH, Chan PC, Liao ST, et al. Strategy to better select HIV- 32 Joshi R, Narang U, Zwerling A, et al. Predictive value of latent infected individuals for latent TB treatment in BCG-vaccinated tuberculosis tests in Indian healthcare workers: a cohort study. Eur population. PLoS One. 2013;8(8):e73069. Respir J. 2011;38(6):1475–1477. 52 Zhang HC, Ruan QL, Wu J, et al. Serial T-SPOT.TB in household 33 Jung YJ, Lyu J, Yoo B, et al. Combined use of a TST and the T- contacts of tuberculosis patients: a 6-year observational study in SPOT®.TB assay for latent tuberculosis infection diagnosis before China. Int J Tuberc Lung Dis. 2019;23(9):989–995. www.thelancet.com Vol 56 February, 2023 13 Articles 53 Zhang LF, Liu XQ, Zhang Y, Deng GH, Pareek M, Lalvani A. 56 Zak DE, Penn-Nicholson A, Scriba TJ, et al. A blood RNA signature A prospective longitudinal study evaluating a T-cell-based assay for tuberculosis disease risk: a prospective cohort study. Lancet. for latent tuberculosis infection in health-care workers in a 2016;387(10035):2312–2322. general hospital in Beijing. Chin Med J. 2013;126(11): 57 Ho CS, Feng P-JI, Narita M, et al. Comparison of three tests for 2039–2044. latent tuberculosis infection in high-risk people in the USA: an 54 Saunders MJ, Wingfield T, Tovar MA, et al. A score to predict and observational cohort study. Lancet Infect Dis. 2022;22(1):85–96. stratify risk of tuberculosis in adult contacts of tuberculosis index 58 Kik SV, Schumacher S, Cirillo DM, et al. An evaluation framework cases: a prospective derivation and external validation cohort study. for new tests that predict progression from tuberculosis infection to Lancet Infect Dis. 2017;17(11):1190–1199. clinical disease. Eur Respir J. 2018;52(4):1800946. 55 Suliman S, Thompson EG, Sutherland J, et al. Four-gene pan- 59 Hamada Y, Cirillo DM, Matteelli A, Penn-Nicholson A, African blood signature predicts progression to tuberculosis. Am Rangaka MX, Ruhwald M. Tests for tuberculosis infection: land- J Respir Crit Care Med. 2018;197(9):1198–1208. scape analysis. Eur Respir J. 2021;58(5):2100167. 14 www.thelancet.com Vol 56 February, 2023