Two interferon gamma release assays for predicting active tuberculosis: the UK PREDICT TB prognostic test study

Identification of latent tuberculosis infection

It is estimated that between one-quarter and one-third of the world’s population is latently infected with the tubercle bacterium4,5 and it is widely accepted that, among those with LTBI, there is approximately a 10% lifetime chance of progression to active TB. Until recently, the tuberculin skin test (TST) was the only tool for the diagnosis of LTBI. TST assesses the delayed-type hypersensitivity response to a purified protein derivative, which contains antigens shared by several mycobacteria and Mycobacterium bovis BCG, by measuring the size of the skin induration following the injection of the antigen. Limitations to the validity (sensitivity and specificity) of TST as a measure of latent infection have been recognised for many years. These are partly attributable to variability in the quality of the administration and interpretation of the test, as well as the state of host immunity.

National and international guidelines and policy context

Diagnostic performance of interferon gamma release assays for development of active tuberculosis: evidence summary

Because of the known limitations of the TST, there is no gold standard against which to compare IGRAs to estimate their sensitivity, specificity, positive predictive value (PPV) or negative predictive value (NPV) for the presence of LTBI. However, for clinical and public health purposes, arguably a more important issue is what a positive or negative IGRA result means for the risk of developing active TB. Several studies have estimated PPV and NPV, sensitivity or specificity of IGRAs for the development of active TB, and two recent systematic reviews15,16 have included meta-analyses of these parameters. A further, more up-to-date, review12 concluded that the two tests were not significantly different in terms of their PPV and NPV.

The two recently published reviews15,16 had slightly different objectives, inclusion criteria and analytical approaches and, for this reason, reached different conclusions. One of them15 estimated a pooled PPV of 2.7% [95% confidence interval (CI) 2.3% to 3.2%] from 17 studies and 6.8% (95% CI 5.6% to 8.3%) from 13 studies that were conducted in high-risk groups. The pooled NPV was considerably higher, at 99.7% (95% CI 99.5% to 99.8%). The included studies were heterogeneous for both PPV and NPV. The other review16 compared rates and risks of TB disease in IGRA-positive and IGRA-negative individuals, as well as estimating sensitivity and specificity of IGRAs for the development of active TB. The pooled rate ratio was 2.10 (95% CI 1.42 to 3.08) and the pooled risk ratio was 3.54 (95% CI 2.23 to 5.60). Sensitivity was estimated as 72% (95% CI 58% to 82%) and specificity as 50% (95% CI 41% to 58%) for the ELISPOT assay; for whole-blood ELISA, only two eligible studies estimated sensitivity and specificity, with values of 79% (95% CI 61% to 91%) and 43% (95% CI 18% to 72%) for sensitivity and 34% (95% CI 32% to 36%) and 59% (95% CI 55% to 62%) for specificity. In the subset of studies that assessed both IGRA and TST, Rangaka et al. 16 reported incidence rate ratios (IRRs) of 2.11 (95% CI 1.29 to 3.46) for IGRA, 1.60 (95% CI 0.94 to 2.72) for TST with a 10-mm cut-off point and 1.43 (95% CI 0.75 to 2.72) for TST with a 5-mm cut-off point. In the review by Diel et al. ,15 PPV appeared higher for IGRA (2.1%, 95% CI 1.7% to 2.5%) than for TST (1.4%, 95% CI 1.2% to 1.8%) in studies that assessed both tests.

The review by Rangaka et al. 16 was updated by Pai et al. ,11 who identified an additional five longitudinal studies that compared the incidence of TB between individuals with positive and negative IGRA results. Three of these studies were conducted in immunocompromised populations: human immunodeficiency virus (HIV)-positive patients,17 patients with various immunosuppressive conditions18 and renal transplant recipients19. The remaining two studies assessed the development of active TB among contacts of TB cases in the UK7 and France,20 comparing those with a positive IGRA result at baseline with those with a negative result. In the UK, this generated a rate ratio of 4.9 (95% CI 2.7 to 8.7) over 2 years,7 whereas the French study reported a PPV of 1.96% among contacts who were not given chemoprophylaxis (CPX), over a mean follow-up period of 34 months. 20

PubMed was searched on 19 July 2016 to update the evidence base, using the search strategy of Rangaka et al. 16 (which was itself based on strategies used in previous reviews21) and restricting to publication dates in 2013 or later. We sought to identify primary studies in which individuals underwent IGRA testing at baseline and were followed up for the development of active TB, and which either presented estimates of PPV and NPV or provided sufficient information for these to be estimated. Studies in which no cases of active TB occurred were considered not to be informative for this purpose and were excluded.

The search of PubMed generated 2399 results (Figure 1), of which 597,9,10,22–77 included the required information on PPV and/or NPV (see Appendix 2, Figure 14). In addition, we are aware of a further relevant study78 that was published after we conducted our review. Many of these 59 studies did not have the primary aim of assessing diagnostic accuracy. Many recorded few cases of active TB (often < 10)22,25–27,29–33,35–38,40,42–45,47,48,50,51,56–59,62–65,67–75,77,78 and/or were restricted to groups such as patients infected with HIV,9,28,29,36,49,51,56,60,65,68,75 recipients of haemodialysis or kidney transplants,35,43,44,62–64 patients taking or initiating treatment with biological agents such as tumour necrosis factor alpha (TNF-α) inhibitors,37–40,42,45,47,48,58,71,74 recipients of haematopoietic stem cell transplants,50,57 lung cancer patients30 or patients with combinations of these conditions. 61 Four further studies were conducted among health-care workers. 59,69,73,77 The duration of follow-up ranged from 8 to 10 weeks to 10 years. Consistent with the previous systematic review,15 estimates of PPV were consistently low, ranging from 0% to 24%; NPV was considerably higher (always ≥ 90%).

FIGURE 1.

Flow diagram summarising study selection.

Eighteen studies (including the additional study78) did specifically assess the diagnostic accuracy of IGRAs for future development of active TB (Table 1) or otherwise quantify the association between a positive IGRA and subsequent TB. Of these, five assessed both IGRAs and TST. 23,52,56,62,72 Except when noted, these studies did not include treatment for LTBI. In this subset of studies, estimates of the PPVs ranged from < 1% to 17%, whereas the NPV was ≥ 97%. Sensitivity was reported as 37.5–100.0% and specificity as 27.0–86.0%. Even among these studies, case numbers were sometimes low (see Table 1). In addition, several of the studies29,31,32,35,36,52,56,62,72 did not explicitly state that prevalent cases were excluded, although the extent to which this may have affected their estimates was not always clear.

Within the subset of 18 studies, several compared the frequency of TB disease between IGRA-positive and IGRA-negative participants, consistently finding that rates, hazards or odds of TB were higher among the former (although CIs sometimes crossed the null value). Unadjusted estimates included a hazard ratio of 4.0 (95% CI 0.8 to 19.3),36 odds ratio of 2.10 (95% CI 0.13 to 34.4)35 and rate ratios of 8.5 (no CI given),34 2.9 (95% CI 2.8 to 21.2)78 and 7.7 (95% CI 2.8 to 21.2). 52 Doyle et al. 29 estimated the hazard ratio as 42.4 (95% CI 2.2 to 827) after adjustment for age, gender, cluster of differentiation 4 (CD4) count, viral load and country of birth. Of the studies that did not specifically aim to assess diagnostic performance, such estimates included unadjusted rate ratios of 2.12,24 3.4650 and 73.9 (95% CI 3.9 to 1397.7);75 odds ratios of 9.26 (95 CI 1.02 to 84.0)30 and 7.9 (95% CI 3.46 to 18.06);66 and hazard ratios were 16.82 (95% CI 5.84 to 48.46), excluding treated individuals, and 2.38 (95% CI 1.15 to 4.91), including treated individuals. 41 Adjusted values included a hazard ratio of 2.0 (95% CI 1.2 to 3.3), after adjustment for BCG vaccination scar, maternal education and current/prior household TB contact,54 and an odds ratio of 14.9 (95% CI 2.77 to 80.3), adjusted for the number of IGRA-positive contacts of the same index case, receipt of treatment and place of contact (household vs. other). 76 These effect estimates varied considerably and were often imprecise. In addition, analyses of data from contacts often did not account for clustering by index case, and logistic regression did not account for differences in time at risk of developing active TB.

Table 2 summarises the studies that compared IGRAs with TST in the same participants and assessed predictive values, combining the results of our literature search with earlier studies identified in previous reviews. 12 Only four of these studies23,79–81 were conducted in low-incidence countries with a PPV range of 3% to 17% for IGRA compared with 2–5% for TST. The estimates of NPV for both IGRA (range 98–100%) and TST (99–100%) were very high. Studies from intermediate- and high-burden settings56,62,72,75,82–84 had a wider range of PPV for both IGRA and TST and, similarly, a high but wider range of NPVs.

Our evidence review, and the previously published systematic reviews and meta-analyses, highlight several important research gaps. There has been relatively little study of the diagnostic performance of IGRAs, compared with TST, for the development of active TB in low-incidence countries and particularly in the UK. The one study in Table 1 that was UK based7 did not include a head-to-head comparison with TST and was focused on contacts of TB cases. Although contacts are an important risk group, current guidance in many low-incidence countries includes testing of new entrants from high-burden countries, in whom there is limited evidence of diagnostic performance. Predictive values, both positive and negative, depend on the prevalence of the condition within the population of interest; consequently, results from high-incidence settings may not be generalisable to the UK. Owing to the low risk of progression from LTBI to active TB disease, the PPV of IGRAs for this outcome will, in general, be low. 11 It is therefore important to identify which individuals, among those with a positive IGRA (or TST) result, are most likely to develop disease and, therefore, benefit from treatment of LTBI. The PREDICT (Prognostic Evaluation of Diagnostic IGRAs ConsorTium) study aimed to address these gaps in the evidence base in order to inform UK and international policy for testing and treating LTBI in contacts of TB cases and in migrants.

Review of cost-effectiveness studies

The NICE guideline on TB, which was published in 2006,85 recommended a two-step testing strategy to diagnose LTBI: TST followed by IGRA for patients with a positive TST result. This recommendation was informed by a cost-effectiveness analysis (CEA) based on a simple decision tree model, a systematic review of diagnostic accuracy of TST compared with IGRA, and a series of assumptions about the predictive performance of IGRA in populations at different levels of risk. The impact of transmission was estimated by assuming a fixed number of incident cases in contacts of the modelled cohort. The Guideline Development Group noted the high degree of uncertainty over the estimates of cost-effectiveness of different testing strategies for LTBI.

In 2011, NICE published an updated review86 of diagnostic accuracy of tests for LTBI and economic analysis, reaching rather different conclusions. The revised model suggested that the IGRA-alone strategy (without TST) was optimal, although the estimated differences in costs and quality-adjusted life-years (QALYs) were small. Given the high level of uncertainty, the 2011 NICE guideline recommended IGRA-alone or TST-IGRA strategies as options. Owing to limitations in the evidence base, the 200685 and 201186 NICE guidelines did not distinguish between the two types of commercially available IGRAs [QuantiFERON TB Gold (QFT-G) and T-SPOT.TB], and compared only three screening strategies (TST alone, IGRA alone and TST followed by IGRA if TST result was positive). Both the 2006 and 2011 guidelines followed policy at the time, recommending two cut-off points for defining a positive TST result: induration of ≥ 15 mm for patients with a prior BCG vaccination or ≥ 6 mm for patients without a BCG vaccination. More recent NICE guidelines,14 published in 2016, based on a new decision model, led to revised recommendations, described further in the following paragraphs.

A series of cost-effectiveness analyses have also been published in the literature. In 2011, Nienhaus et al. 87 published a review of economic studies evaluating TST and/or IGRA as components of screening strategies for groups at high risk of TB, such as migrants, close contacts and health-care workers, mostly in low-incidence countries. 87 Of the 13 papers that met their inclusion criteria, eight88–95 involved CEA, whereas five96–100 were cost analyses.

The published cost-effectiveness studies reviewed by Nienhaus et al. 87 all used Markov modelling to estimate transition of a cohort through different health states, assuming a fixed rate of infection in the cohort. None modelled the dynamics of transmission within a wider population. One CEA study95 presented results in terms of cost per avoided TB case, whereas the remaining studies88–94 used QALYs or life-years gained as treatment outcomes. The time horizon for analysis after initial testing for LTBI ranged from 2 years to lifetime in the different CEA models. Some populations were stratified, for example health-care workers with/without BCG vaccination. Although the CEA studies generally supported the IGRA-alone or IGRA-based strategies, Nienhaus et al. 87 noted that comparison of the different models was hindered by a wide range of assumptions. The models were most sensitive to assumptions about specificity of TST, which varied among non-BCG-vaccinated individuals from as low as 34% in one study to 98% in another. There was also significant uncertainty surrounding the rates of progression to active TB following a positive test, as well as overassumptions for IGRA and TST sensitivities.

In 2011, after the Nienhaus et al. 87 review, Pareek et al. 101 published a UK-based prospective assessment and an economic analysis. They sought to address the limitations of the existing models and to examine how LTBI prevalence varied by TB-incidence thresholds in migrants’ regions of origin. They found that single-step QFT-GIT without port-of-arrival chest radiography could be cost-effective at certain incidence thresholds. The results are, however, limited by the small number of participants and the fact that not all of the participants received all three tests.

The NICE TB clinical guideline was updated again in 2016,14 making the following recommendations for LTBI testing in contacts and new entrants to the UK.

• Offer TST to diagnose LTBI in adults aged 18–65 years who are close contacts of a person with pulmonary or laryngeal TB:

  • If the TST is inconclusive, refer the person to a TB specialist.

  • If the TST is positive (an induration of ≥ 5 mm, regardless of BCG vaccination history), assess for active TB.

  • If the TST is positive but a diagnosis of active TB is excluded, consider an IGRA if more evidence of infection is needed to decide treatment.

  • If the TST is positive, and if an IGRA was carried out and that is also positive, offer treatment for LTBI.

• Offer TST as the initial diagnostic test for LTBI in people who have recently arrived from high-incidence countries and who present to health-care services. If the TST is positive (an induration of ≥ 5 mm, regardless of BCG vaccination history):

  • assess for active TB

  • and if this assessment is negative, offer treatment for LTBI

  • if TST is unavailable, offer IGRA.

These recommendations differ from the previous NICE guidelines85,86 in two key respects. First, they downplay the role of IGRA, recommending it as an option following TST when more evidence is needed before treating contacts, or when TST is not available for new entrants to the UK. Second, they recommend a single cut-off point for a positive TST (an induration of ≥ 5 mm) regardless of BCG vaccination status, rather than two cut-off points, as recommended in previous guidelines (an induration of ≥ 6 mm without prior BCG vaccination or an induration of ≥ 15 mm with prior BCG vaccination).

The 2016 NICE recommendations14 were informed by an evidence review and new economic analysis by Auguste et al. 102 (henceforth referred to as the Warwick Evidence report). Auguste et al.'s model consisted of two stages. In the first stage (illustrated in Figures 2 and 3), a decision tree was used to represent LTBI screening, active TB diagnosis and treatment adherence. They examined four diagnostic strategies:

1. IGRA alone

2. TST alone

3. combined strategies – TST followed by IGRA if TST is positive

4. simultaneous testing – both TST and IGRA are carried out.

FIGURE 2.

Starting decision tree of the Warwick Evidence analysis for new entrants from high-incidence countries. Reproduced from Auguste et al. 102 Contains information licensed under the Non-Commercial Government Licence v2.0.

Reproduced from Auguste et al. 102 Contains information licensed under the www.nationalarchives.gov.uk/doc/non-commercial-government-licence/version/2/.

FIGURE 3.

Example decision tree extension from Figure 2, showing only strategy A (i.e. IGRA alone). +ve, positive; –ve, negative. Reproduced from Auguste et al. 102 Contains information licensed under the Non-Commercial Government Licence v2.0.

Reproduced from Auguste et al. 102 Contains information licensed under the www.nationalarchives.gov.uk/doc/non-commercial-government-licence/version/2/.

Only the pathway for the IGRA-alone strategy (A) is presented here (see Figure 3). A do-nothing (no test) strategy was not explored. The IGRA strategies A, C and D were evaluated for three types of commercially available IGRAs: QTF-G, QFT-GIT and T-SPOT.TB.

The decision tree was used to estimate the proportion of patients emerging from the initial assessment, diagnosis and treatment stages with different risks of disease progression in the second stage of the model. Three populations were considered: (1) children, (2) immunocompromised individuals and (3) recently arrived individuals from high-incidence countries.

The second stage of the Warwick Evidence model102 (Figure 4) was a transmission model, operationalised using discrete event simulation (DES), programmed using R (The R Foundation for Statistical Computing, Vienna, Austria). Individuals arriving from the decision tree were grouped as (1) having active TB, (2) already successfully treated for LTBI and, hence, free of LTBI/TB or (3) untreated for LTBI. The model then tracked disease progression, deaths and cases of new secondary infection. Secondary infection in other individuals was modelled as a proportion of the initial cohort (sampled from a Poisson distribution) who progressed or relapsed to active TB. The analysis assumed that individuals who entered the transmission model with LTBI would ultimately develop active TB if they did not receive treatment, although they might die of other causes before developing TB. In the initial population, as well as in secondary cases, people with LTBI who do not progress to active TB were not simulated in the second part of the model. The risk of an event in the transmission model, such as death or onset of active TB, was dependent on an individual’s age, TB status and current treatment, and was updated when any of these factors changed. Time to activation of TB for individuals entering stage two of the model with LTBI (untreated or unsuccessfully treated LTBI cases) was simulated assuming a constant activation rate over time.

FIGURE 4.

Warwick Evidence analysis (Tuberculosis Clinical Guidance) transition model. Reproduced from Auguste et al. 102 Contains information licensed under the Non-Commercial Government Licence v2.0.

Reproduced from Auguste et al. 102 Contains information licensed under the www.nationalarchives.gov.uk/doc/non-commercial-government-licence/version/2/.

The Warwick Evidence analysis102 presented per-patient costs as well as a breakdown of total costs (including costs of diagnosis, LTBI treatment, active TB and treatment-related hepatitis) for all of the strategies, based on the mean values of all parameters after 250,000 individual patient simulations. The incidence rates (IRs) of active TB in the initial cohort, the number of secondary infections and the mean life-years and QALYs per strategy were also presented. In addition, based on 2000 Monte Carlo simulations, probabilistic estimates of the total cost, incremental costs, diagnostic errors (false-positive results plus false-negative results), incremental diagnostic error, mean QALYs, incremental QALYs and incremental cost-effectiveness ratios (ICERs) were also presented.

The analysis suggested that, for new entrants from high-incidence countries, simultaneous testing strategies were dominated by the equivalent sequential strategies. The QFT-GIT-alone strategy was least costly, but the TST-positive (an induration of ≥ 5 mm) followed by QFT-GIT strategy had the fewest errors. At a willingness-to-pay threshold of £20,000 per QALY gained, the TST-alone (≥ 5 mm) strategy was optimal, with an ICER of £1524 per QALY gained compared with the QFT-GIT strategy. The TST-negative (an induration of < 5 mm) followed by QFT-GIT strategy was ranked as the next most cost-effective strategy, with an ICER of £58,720 per QALY gained compared with TST (an induration of ≥ 5 mm). The other strategies were dominated. Univariate sensitivity analysis suggested that the TST-alone strategy remained cost-effective in most scenarios. However, results were sensitive to LTBI prevalence, sensitivity of QFT-GIT and sensitivity of TST.

A major common limitation of all current CEA of screening strategies for LTBI relates to assumptions surrounding the predictive value of IGRAs and TST, because there is no gold standard against which to establish the sensitivity or specificity of the different tests. Previous CEA studies have based their estimates of LTBI predictive values on various, and often conflicting, sources and assumptions. This leaves considerable uncertainty over the cost-effectiveness of LTBI screening strategies.

We adopted a different approach, basing our economic analysis on the ability of the three tests (TST, QFT-GIT and T-SPOT.TB), individually and in combination, to predict progression to active TB in the PREDICT cohort. This represents a much stronger evidence base for modelling the predictive performance of LTBI test strategies than was available for previous economic evaluations, with three main advantages. First, the cohort provides a coherent source of baseline data, including risk factors and TST, QFT-GIT and T-SPOT.TB test results for large samples of new entrants and contacts being tested for LTBI in the UK. Second, the long-term follow-up and predictive modelling of TB incidence bypasses the problems of conventional diagnostic accuracy in the absence of a reliable reference standard test. Third, this evidence source enables the comparison of two commercially available IGRAs (T-SPOT.TB and QFT-GIT) with TST strategies with a single cut-off point (an induration of ≥ 5 mm) or with different cut-off points for BCG-naive and BCG-experienced patients (indurations of ≥ 5 mm and ≥ 15 mm, respectively).

Objectives

Structure of the report

In this chapter of the report, we have included a description of the background of the project as well as the primary and secondary objectives. Chapter 2 describes the methods used in the research. In Chapter 3, we present the results of the epidemiological study by describing the cohort and the development of active TB in the cohort. We present the results of the economic analysis examining multiple screening scenarios in Chapter 4 and, in Chapter 5, we discuss the findings, implications for TB testing and treatment guidelines and recommended future research areas.

Setting, population and disease burden

This prospective cohort study recruited individuals aged ≥ 16 years who were (1) close contacts of patients with active TB or (2) new entrants to the UK arriving in the last 5 years from high-incidence countries, defined as exceeding 40 cases per 100,000 people, and operationalised by focusing on those from sub-Saharan Africa or Asia.

Participants were recruited in TB clinics from a network of hospitals in London, Leicester and Birmingham, general practices (GPs) in London and several community settings in London (see Appendix 1). All study sites were co-ordinated from the tuberculosis section of PHE.

Recruitment and inclusion criteria

Participants were recruited between May 2010 and July 2015. Contacts of active TB (pulmonary and extrapulmonary) patients and new entrants arriving in the UK in the last 5 years from high-incidence countries (as defined in Setting, population and disease burden) and attending participating TB clinics or primary care centres for screening were invited to take part. Those born in high-incidence countries who entered the UK > 5 years ago, but who had spent > 1 year (cumulative) in the past 5 years in a high-incidence country, as per the study’s defined list, were also invited to participate. Contacts included all individuals with a cumulative duration of exposure of > 8 hours to the relevant index case in a confined space during the period of infectiousness (prior to initiation of treatment).

Eligible persons were identified by study TB specialists or practice nurses, and written informed consent was obtained following the provision of information sheets (translated as appropriate). Baseline assessment questionnaires, including demographic and clinical information, were completed by the research nurse.

In addition, recruitment was undertaken en masse using two methods.

1. New entrants to the UK were recruited from non-NHS community settings, such as places of worship and community centres.

2. Contacts of active cases were recruited through mass screening events organised as part of the public health response to a case of active TB in some situations. For example, clinical teams may attend workplaces or colleges where an exposure had taken place, to facilitate screening of large numbers of contacts.

For recruitment through primary care, study flyers and the contact details of the co-ordinating centre were displayed in general practices so that interested people could contact the study team to book an appiontment. At the appointment, as with all recruitment meetings, a study nurse went through the full patient information leaflet before taking written informed consent to undertake study procedures. We also utilised the primary care trust (PCT)-held Flag 4 data (records held by the local primary care group about international migrants who register with a NHS GP) to invite newly registered patients, recently arrived from the countries of interest, to take part in the study.

At the time of the study, treatment of LTBI was recommended only for individuals aged < 35 years. We therefore prioritised the recruitment of patients aged > 35 years (who were not eligible for CPX) in order to estimate and compare the ability of the TST and IGRAs to predict natural progression to active disease in the absence of treatment. Individuals aged 16–35 years were also eligible, as they may not be offered, or may not accept, CPX.

The general practioners of all participants were informed of their patients’ participation by letter.

Health technologies being assessed

Participants were tested by the Mantoux test and two IGRAs: ELISA (QFT-GIT) and ELISPOT assay (T-SPOT.TB). All tests were conducted using standardised protocols. 17–19 Tests for LTBI were undertaken among contacts at about 6 weeks after last known exposure (consistent with NICE guidelines14), and at ≥ 6 weeks for new entrants after arrival in the UK.

The QFT-GIT and T-SPOT.TB assay samples were transported to a study testing centre (see Statistical analysis).

After IGRA testing, samples were used for repetition of indeterminate assays. Paxgene tubes, plasma and peripheral blood mononuclear cells were stored for future research.

Test results and further action

All test results were obtained by clinicians who were blind to other results. Clinicians were informed of all test results by the testing laboratory for tests conducted at the National Mycobacterial Reference Laboratory (NMRL), whereas those results from the Imperial College London laboratory were not reported. Subsequent action after testing followed the version of NICE guidance that was applicable at the time of screening and took place after discussion with participants. Further action followed one of three options:86

1. If the participant had a negative result in the TST and IGRA, follow-up only was conducted.

2. If the participant had a positive result in either the TST or IGRA, and active TB was excluded, follow-up was conducted irrespective of age.

3. If the participant had a positive result in the TST and either one or both IGRA(s), follow-up only was conducted for those aged ≥ 35 years, and, for those aged 16–35 years, CPX was offered within the routine care setting after excluding active TB, with balanced advice about potential benefits and risks.

Follow-up

Participants were followed prospectively after IGRA/TST testing. Follow-up utilised multiple data sources to enable the comprehensive national identification of participants and to minimise loss attributable to transfer of care to other centres by physicians. Data were sourced from (1) telephone calls to the participants at 12 months and at 24 months or at the end of follow-up, (2) a search of national mandatory enhanced TB surveillance (ETS) reports, (3) a search of the national database of culture-proven TB and (4) a search of clinical records. Individuals found to have active TB were excluded from the analysis if they had evidence of active disease at the time of recruitment (see Primary outcome). Follow-up continued until the development of active TB, death or May 2016.

Data collection

Baseline data were collected by trained research nurses using paper questionnaires. Outcome data were collected using the four approaches outlined in Follow-up. Questionnaire data were entered into a purpose-built database [using Microsoft Access^® (Microsoft Corporation, Redmond, WA, USA)] that was maintained at the TB section at PHE and linked, as necessary, to data from the other sources. IGRA results were received from laboratories in Microsoft Excel^® (Microsoft Corporation, Redmond, WA, USA) spreadsheets and imported into the database.

Data collected from participants included age; gender; country of birth and date of entry to the UK for non-UK-born persons; ethnicity; current employment; the nature and duration (in hours) of contact with the index case (for contacts); the time interval between the most recent suspected exposure date and the date of diagnosis in the index case; details of any previous contact tracing; history of previous TB, including treatment, results of previous TST and chest radiographic findings; BCG vaccination status (assessed based on criteria in the Green Book:103 a scar or reliable recall or vaccination record); associated medical diagnoses; and use of immunosuppressive agents. Self-reported HIV infection status was collected, and further work through record linkage with the national HIV surveillance system will be used to improve the completeness of HIV data. Quality-of-life data were collected using the EuroQol-5 Dimensions (EQ-5D) questionnaire. During follow-up, information on the drugs used for the treatment of LTBI and adverse effects of CPX was collected and validated through a review of clinical records.

Index cases (identified through clinical records) of contacts were identified through a web-based ETS system using data provided by the contact and/or clinicians. The characteristics of index cases were summarised descriptively. For contacts who progressed to active TB disease during follow-up, deoxyribonucleic acid (DNA) fingerprinting data [24-locus mycobacterial interspersed repetitive unit – variable number tandem repeats (MIRU-VNTR)] from the UK national TB strain typing database, hosted by PHE, were utilised to compare strain types between index cases and subsequent diagnoses among contacts. This enabled the assessment of whether any TB disease in secondary cases was likely to have arisen from the initial infection (i.e. that the test result related to the outcome) and not as a result of subsequent exposure to another case.

Primary analysis

The analysis estimated the ability of the tests (TST, T-SPOT.TB and QFT-GIT), individually and in combination, to predict progression to active TB in latently infected individuals, and made comparisons between tests. Estimated rates of progression to active TB were calculated, accounting for the follow-up period for each individual.

The predictive performance of each test (TST, T-SPOT.TB and QFT-GIT) and of each two-step test strategy [using combinations of TST + T-SPOT.TB, TST + QFT-GIT and IGRA (T-SPOT.TB or QFT-GIT) + TST] was summarised by the calculation of IRs of developing active TB in the ‘test-positive’ and ‘test-negative’ groups, expressed with 95% CIs computed using Poisson exact methods. In each combination strategy, the second test result was used if the first was positive, meaning that both tests must be positive for a participant to be positive for the strategy. IRs were calculated using all available follow-up data and also at 1 year and 2 years after study entry. The discriminatory predictive value of the test was based on the relative comparison of these rates (the relative IRs comparing those with positive test results and negative test results), together with the prevalence of positive test results (indicating the number of patients who would be treated should the test be used for recommending CPX).

The primary analyses of the performance of tests used only participants with test result data for all three tests. Participants who were treated for LTBI were excluded.

We made pairwise comparisons of the ability of tests and strategies to identify (1) those participants who progress and (2) those participants who do not progress using generalised estimating equation (GEE) marginal regression models. 105 These models estimate (1) ratios of positive test results and (2) ratios of negative test results in those participants who progress compared with those participants who do not, which are comparable to positive and negative likelihood ratios for diagnostic studies. We compute ratios of these ratios to make pairwise comparisons between tests, accounting for the paired data. A marginal regression GEE model was fitted, with the two test results as outcomes (with a positive test result coded as the event) and with test type and progression status as predictors, and an interaction term between test type and progression. The interaction term assessed whether or not the relationship between progression and test positivity was stronger for one test than the other. A similar second model was fitted with test negative as the outcome, such that the interaction term assessed whether or not the relationship between non-progression and test negativity was stronger for one test than the other. The model accounted for the intra-individual correlation between tests using an unstructured correlation matrix and was configured to give population average estimates. The primary analysis fitted models with a binomial error structure and a log link (as described by Pepe105). In a sensitivity analysis, we adapted the Poisson model to include follow-up using an offset and obtained identical point estimates to two decimal places and confidence limits to one decimal place. We report the binomial model results.

Regression models were fitted in Stata^® version 15.0 (StataCorp LP, College Station, TX, USA).

Positive predictive values and NPVs were also calculated as the percentage of people with a positive result developing TB across all follow-up (PPV) and as the percentage of people with a negative result who did not develop TB (NPV).

Owing to testing processes, a small number of participants had two T-SPOT.TB and/or QFT-GIT tests performed (T-SPOT.TB, n = 21; QFT-GIT, n = 33). For participants with multiple results for a single test, results from the primary research laboratories were used (Imperial College London and NMRL), with results from Imperial College London being used if the test was carried out at both Imperial College London and NMRL. For some participants, neither T-SPOT.TB result was obtained from Imperial College London or NMRL and, therefore, the result from Oxford Immunotec or SYNLAB (London, UK) was used.

The T-SPOT.TB test

The T-SPOT.TB test can have positive, borderline positive, negative, borderline negative or indeterminate results. All categories of test results were reported. For the majority of participants, the required test information (spot count) was available and T-SPOT.TB test result classifications were calculated by the study team in accordance with the manufacturer’s algorithm (Figure 5)107 For 660 tests (7.9% of the 8387 participants, or of the 8366 participants with a duplicate test for 21 participants) in which the necessary information (T-SPOT.TB counts and positive and negative control values) was not provided, the classification of results reported from the laboratory was used in all subsequent analyses.

FIGURE 5.

T-SPOT.TB result classification algorithm.

Indeterminate results were treated as missing, borderline positive results were treated as positive and borderline negative results were treated as negative, in line with the interpretation of results in clinical practice. 1

Sensitivity analyses

Sensitivity analyses were carried out using all available test data rather than using data only from participants with results for all three tests. Analyses were also performed excluding those participants with BCG vaccination status assumed to generate a TST^b result and those participants with an uncertain progression status because of a short period between testing and diagnosis. Additional sensitivity analyses were performed using data from new entrants only and restricted to 2 years of follow-up, and from contacts of pulmonary TB cases only.

The decision problem

We developed an individual-level DES model to estimate the effect of different test strategies on risk prediction, use of preventative CPX and BCG vaccination, subsequent incidence of TB, and, hence, costs to the health-care system and health outcomes for patients. We considered a DES model to be most suitable as individual simulation and a bootstrapping approach preserved individual patient data and the correlation between baseline risk factors and the test results. The analysis followed the NICE reference case for public health interventions. Health outcomes were quantified using QALYs, including the impact of TB and treatment-related adverse effects on mortality and health-related quality of life (the utility). EuroQol-5 Dimensions, three-level version scores from the PREDICT study were valued using the UK general population tariff. Costs were estimated from a UK public sector perspective, including costs for case finding, prevention (BCG vaccination and CPX) and treatment of active TB and adverse effects. In the base case, we discounted costs and QALYs at 3.5% per annum and simulated outcomes over a lifetime horizon for a cohort of patients undergoing tests for LTBI.

Population

We evaluated the cost-effectiveness of LTBI testing strategies in the PREDICT cohort. The model used an individual patient data set comprising 4162 contacts and 3795 new entrants without active TB at baseline (total of 7957 participants) with sufficient data: age, sex, BCG vaccination status, baseline EQ-5D score (utility) and contact/new entrant status. We excluded participants with active TB at the baseline assessment, but included those with missing values for one or more baseline LTBI tests (TST, QFT-GIT or T-SPOT.TB) in order to model the impact of missing or indeterminate test results.

For our base-case analyses, we evaluated results for the whole cohort, and also for the two subgroups: new entrants to the UK and contacts. We had initially intended to present stratified analysis for other risk factors: age, BCG vaccination status, previous treatment status and baseline risk (defined by level of exposure or duration and nature of contact; or by country or region of birth for new entrants to the UK). However, an analysis of the PREDICT results did not show any important or statistically significant differences in incidence according to these risk factors (see Chapter 4). Instead, we investigated the sensitivity of results for populations at different a priori levels of risk in the absence of preventative treatment (BCG vaccination or CPX). This illustrated the possible effects of changing referral thresholds for LTBI tests, such as the threshold for defining a ‘close’ contact or ‘high-incidence’ country of recent residence.

In our modelling protocol, we stated that we would consider additional exploratory analyses in other subgroups, such as people infected with HIV or who have diabetes mellitus, but the numbers of participants in these groups were too low to support such analysis.

Interventions and comparators

We compare 11 approaches to LTBI testing:

1. 0. no test

2. 1. T-SPOT.TB

3. 2. QFT-GIT

4. 3. TST^a: positive if induration is ≥ 5 mm

5. 4. TST^b: positive if induration is ≥ 6 mm without prior BCG vaccination or ≥ 15 mm with prior BCG vaccination

6. 5. TST^a + T-SPOT.TB: positive if both TST^a and T-SPOT.TB are positive

7. 6. TST^a + QFT-GIT: positive if both TST^a and QFT-GIT are positive

8. 7. TST^a + IGRA: positive if TST^a is positive and either T-SPOT.TB or QFT-GIT are positive

9. 8. TST^b + T-SPOT.TB: positive if both TST^b and T-SPOT.TB are positive

10. 9. TST^b + QFT-GIT: positive if both TST^b and QFT-GIT are positive

11. 10. TST^b + IGRA: positive if TST^b is positive and either T-SPOT.TB or QFT-GIT are positive.

In the single-test strategies (1–4), participants who tested positive for LTBI were assumed to undergo diagnostic tests to rule out active TB, then to be offered appropriate treatment (CPX if aged ≤ 35 years), whereas those who tested negative for LTBI would be offered BCG vaccination if eligible (no previous BCG vaccination and age ≤ 35 years). Participants with missing test results for a strategy were assumed not to receive any preventative treatment (CPX or BCG vaccination).

In the dual-test strategies (5, 6, 8 and 9), the second test would only be carried out if the first test result was positive. Participants who tested positive in the second test (T-SPOT.TB or QFT-GIT) would be further evaluated to rule out active TB, and offered CPX or BCG vaccination if appropriate. Participants without a TST result, or with a positive TST result but missing T-SPOT.TB or QFT-GIT data, were assumed to not receive any preventative treatment.

In the three-test strategies (7 and 10), we assumed that participants with a positive TST result would have both IGRAs (T-SPOT.TB and QFT-GIT). Then, if either of the IGRAs were positive, they would be tested for active disease and offered treatment or CPX if appropriate. If both IGRAs were negative they would be offered a BCG vaccination if appropriate. Participants with an indeterminate result attributable to one or more missing test results would not receive preventative treatment; this included all participants with a missing TST result, and those with a positive TST result and either both IGRAs missing or one negative IGRA and one missing IGRA.

The ‘no test’ strategy was included as a comparator, which allowed us to investigate changes to the referral thresholds for LTBI testing.

Outcomes

Strategies were evaluated in terms of the health-care costs and health outcomes measured using QALYs. For validation purposes, we also present disaggregated outcomes (numbers of incident TB cases; numbers given a BCG vaccination, CPX and active TB treatment; treatment-related adverse events; deaths from TB; and life-years) and costs [costs of diagnosis, preventative treatment (BCG vaccination and CPX) and active TB treatment (including treatment of adverse effects)].

Modelling methods

Model structure

Figure 6 is a process map of our model. The software used to develop the DES model was SIMUL8 Professional 2017 (SIMUL8 Corporation, Boston, MA, USA).

FIGURE 6.

Model process map. –ve, negative; +ve, positive; AE, adverse event; LTFU, lost to follow-up; Tx, treatment for active TB.

Cohort selection

The model was programmed so that a cohort of patients with chosen initial characteristics could be selected to run through the pathway, to allow flexible subgroup analysis. However, for this report we present results only for the whole cohort, and for the ‘new entrant’ and ‘contact’ subgroups.

Individual participants from the PREDICT cohort were randomly sampled (with replacement) from an individual-level data set from the PREDICT cohort. This data set included individuals’ demographic information and baseline characteristics: age, gender, prior BCG vaccination status, utility (EQ-5D scores) and LTBI test results (TST, T-SPOT.TB and QFT-GIT). For the purposes of this analysis, we excluded patients who were diagnosed with active TB at the beginning of the PREDICT study (n = 175). This left a data set of 7957 individuals for inclusion in the economic analysis.

The size of the sampled population in the model can be changed, but for the analysis presented below we used a bootstrapping approach; for each probabilistic iteration of the model, we sampled a simulation cohort (with replacement) of the same size as the individual patient data set (n = 7957). This scales uncertainty over the distribution of baseline characteristics in our probabilistic sensitivity analysis (PSA) to reflect sampling error in the PREDICT cohort.

Screening, diagnosis and treatment initiation

The model can be set to one of the 11 test strategies, and members of the cohort enter to be offered the chosen strategy. Treatment decisions, such as whether a patient is offered BCG vaccination, CPX or treatment for active TB, depend on the patient’s test results and history or demographic data. For example, under strategy 5 (TST^a + T-SPOT.TB), an individual aged ≤ 35 years with no prior BCG vaccination, no active TB, a TST induration of 7 mm and a negative T-SPOT.TB test result would be offered a BCG vaccination. Or, under the same strategy, a patient aged ≤ 35 years, who has had a BCG vaccination, with no TB, a TST induration of 18 mm and a positive T-SPOT.TB result would be offered CPX.

This first section of the model is treated as timeless; patients cannot experience adverse events and their utility is fixed at their baseline EQ-5D value (as observed in PREDICT), but they accrue costs based on choices that are made within the model as per LTBI testing, active TB diagnosis, BCG vaccination and initiation of CPX and treatment of active disease.

Health status

Having received tests and treatments, or having been lost to follow-up, participants flow into the second part of the model, in which time starts to elapse. At any time, each member of the cohort who is alive is in one, and only one, of the following health states:

• Active TB and on treatment – participants in this state have been diagnosed with active TB and are being treated. From this state, participants might successfully finish treatment, experience an adverse reaction to treatment and/or drop out of treatment, or they may die from TB or other causes.

• Active TB and not on treatment – this state includes participants with active TB who have not yet been diagnosed, as well as participants who have been diagnosed but have stopped treatment for some reason. From this state, participants might present clinically with symptoms and be diagnosed, or they might die from TB or other causes. Contacts of people with untreated active TB who are at heightened risk enter the model in the ‘no TB and not on CPX’ state.

• No TB and on CPX – these participants do not have TB but have tested positive for LTBI (according to the defined test strategy) and are taking prophylactic treatment. Future possibilities for this group are to complete CPX, have an adverse event and/or stop CPX, progress to TB or to die from other causes.

• No TB and not on CPX – this state includes all participants who do not have TB and are not taking CPX. This is a diverse group including participants who have tested positive for LTBI but did not start CPX or defaulted from CPX, those who have tested negative for LTBI and who may or may not have been vaccinated, as well as those who did not complete the LTBI test(s) or who had indeterminate results. All of these people are at risk of progressing to TB or they might die from other causes.

Note that LTBI is not modelled as a ‘disease state’ as such, but rather it means that each person who does not currently have active TB is at risk of developing active TB. Each person’s risk of an event is estimated using observed IRs in the PREDICT cohort, based on their current status, other characteristics and treatment effects. Then, for each possible event, a time to event is sampled.

Events and outcomes

The possible events are:

• Adverse events from treatment for active TB or CPX – this may result in death, in which case costs and QALYs are calculated. If the individual survives, the adverse event (drug-induced hepatotoxicity) may resolve with the participant continuing treatment, or symptoms may persist, in which case the participant is withdrawn from treatment.

• Dropout from CPX or treatment for active TB – for simplicity, we assume that dropout from CPX and treatment for active TB will take place halfway through the treatment period (after 1.5 months for CPX or 3 months for treatment for active TB). Possible events following dropout from CPX are death and progression to active TB. The next event after dropout from active TB treatment is either reinitiation of treatment (as patients with active disease are likely to re-present at some time), death from TB or death from other causes.

• Complete CPX or treatment for active TB – participants who complete CPX (after 3 months) may still progress to active TB, or die from other causes. Individuals who complete treatment for active TB (after 6 months) may be reinfected but we assume that this would be captured as secondary infection in a new incident cohort and is therefore outside the scope of this model. They may also suffer reactivation. The next event after completing treatment for active TB may therefore be death from other causes or active TB diagnosis if reactivation takes place.

• Progression to active TB – all individuals without current active TB have a risk of progressing to active TB. Those with positive LTBI test results are at higher risk. Once participants progress, they become undiagnosed cases of TB and the next event may be death or active TB diagnosis.

• Active TB diagnosis – individuals with active TB who are not on treatment will develop symptoms and are likely to present clinically, leading to diagnosis and an offer of treatment. We assume a fixed time to diagnosis for these individuals based on the average time between onset of symptoms and diagnosis (3 months).

• Death from active TB or other causes – costs and QALYs are calculated at this point.

Model assumptions

The key modelling assumptions are:

• Only BCG-naive individuals who are aged ≤ 35 years are offered a BCG vaccination, a proportion of whom accept (94% in the base-case model). Any adverse effects of BCG vaccination are assumed to be negligible.

• Only those individuals aged ≤ 35 years are offered CPX; this reflects UK practice and guidelines86 during the study period. A proportion of patients who are eligible accept and start CPX (94% in the base case), although they may drop out or suffer an adverse event.

• Death from a CPX-related adverse event (hepatotoxicity) is possible, and is assumed to take place at the start of treatment. Fatal adverse reactions to CPX are very rare; thus, the timing of adverse events during treatment is unlikely to influence the results.

• If the patient survives and the adverse event resolves, the patient resumes CPX but may drop out later. We assume that if the patient resumes CPX they will not experience a second adverse event. If the adverse event is not resolved, the patient stops CPX and will either develop active TB or die from other causes.

• Patients who drop out from CPX are assumed to do so in 1.5 months and incur half of the cost of a complete course (3 months) of CPX. Incomplete CPX provides some protection against progression to active TB, but is less effective than complete CPX.

• After completing CPX patients may still progress to active TB (despite preventative treatment) or die from other causes.

• All individuals have a risk of developing active TB with a hazard conditional on their baseline test results (based on IRs estimated and sampled probabilistically from the PREDICT study). Hazards are estimated based on patients’ available test results, using the test or combination of tests with the greatest predictive value.

• Once an individual progresses to active TB they may be diagnosed, die from active TB or die from other causes. For patients who are diagnosed with active TB, the time from TB onset to diagnosis is assumed to be 3 months.

• All those diagnosed with active TB are assumed to start treatment, although they may drop out or suffer an adverse event.

• A treatment-related adverse event can only take place once. Death from an adverse event is possible and is assumed to take place at the start of treatment.

• If the patient survives, and the adverse event resolves, the patient resumes treatment but may drop out later. A second adverse event is not possible.

• If the patient survives but the adverse event is not resolved, the patient stops treatment and will either die from active TB or die from other causes. For simplicity, we have not modelled second and subsequent lines of treatment for patients who experience adverse reactions or resistance to the standard regimen.

• Patients who drop out from treatment for active TB are assumed to do so in 3 months and incur half of the cost of a complete course of 6 months. They do not get any treatment benefits from incomplete treatment for active TB and are still at risk of active TB death, as well as death from other causes. However, patients who have dropped out may be identified through opportunistic testing or clinical presentation, and will be offered a repeat course of treatment.

• Patients who complete treatment for active TB are assumed to be free of TB and will die only from other causes.

• Individuals come into the model with a utility estimated from their PREDICT EQ-5D data. For simplicity, we have assumed that an individual’s utility score remains constant as they age. As utility tends to fall with increasing age, QALY differences between strategies will be overestimated.

• Individuals’ utility scores are adjusted when they have active TB. A pre-treatment utility multiplier is applied as soon as the patient develops active TB to reflect the impact of emerging TB symptoms prior to treatment. A lower utility multiplier is applied during the first 3 months of treatment to reflect continuing symptoms, inconvenience and distress caused by isolation and adverse effects of treatment that are tolerated. The utility multiplier applied during the last 3 months of treatment is then higher as most patients will have less-severe symptoms by this time, and the prevalence of adverse effects is likely to be lower. After completion of treatment, patients are assumed to return to their initial utility value. 108

• It is assumed that CPX has a negligible impact on utility. 108

• For simplicity, our base-case model does not include transmission from members of the cohort who develop TB to their contacts. This will tend to underestimate the benefits (QALY gain and health-care cost savings) associated with TB prevention. We conducted a scenario analysis with a ‘cascade’ approach to model the impact of secondary transmission.

Model parameters

Test results

The QFT-GIT tests were carried out in four centres (NMRL, Imperial College London, Royal Free London NHS Foundation Trust and SYNLAB) and the T-SPOT.TB test was carried out in three centres (NMRL, Imperial College London and Oxford Immunotec Ltd). The results are summarised in Table 13. Positive test results (positive and borderline positive) were seen for participants when using the T-SPOT.TB test (n = 1571, 16.3%), QFT-GIT (n = 1892, 19.7%), TST^a (n = 3513, 36.6%) and TST^b (n = 1729, 18.0%). There were 6520 participants (67.8%) with a non-missing or indeterminate result (positive or negative) for the three tests (T-SPOT.TB, QFT-GIT and TST^a) and 6386 participants (66.5%) had a non-missing or indeterminate result (positive or negative) for all tests including TST^b. An ‘error’ was reported when a result could not be obtained as a result of the testing procedures, for example because there was not enough sample material.

Some participants had multiple tests carried out (21 participants had two T-SPOT.TB tests and 33 participants had two QFT-GIT tests).

In the following discussion, ‘borderline positive’ results are categorised as positive; ‘borderline negative’ results are categorised as negative; and ‘error’ and ‘indeterminate’ results are categorised as missing, as only positive and negative results can be used in decision-making.

Evaluation of the combination of results for the different tests showed that, when considering the results from the three tests, T-SPOT.TB, QFT-GIT and TST^a (6520 participants had available data), the results were positive for all three for 797 participants (12.2%) and were negative for all three for 2995 participants (45.9%), and the three results differed for 2728 participants (41.8%). When comparing the results for T-SPOT.TB and TST^a (6843 participants had available data), the results were both positive for 1020 participants (14.9%), were both negative for 3414 participants (49.9%) and disagreed for 2409 participants (35.2%). Comparing the results for QFT-GIT and TST^a (7041 participants had available data), 1113 participants (15.8%) had positive results for both, 3373 participants (47.9%) had a negative result for both and 2555 participants (36.3%) had discordant results.

A total of 6386 participants had a test result for each of T-SPOT.TB, QFT-GIT and TST^b. Of the 6706 participants with T-SPOT.TB and TST^b results, 743 (11.1%) had positive results for both, 4598 (68.6%) had negative results for both and the results disagreed for 1365 participants (20.4). For the 6897 participants with TST^b and QFT-GIT results, 785 (11.4%) had positive results for both, 4563 (66.2%) had negative results for both and the results for 1549 participants (22.5%) disagreed. Of the 6386 participants with all results, 620 (9.7%) had all positive results, 4026 (63.0%) had all negative results and the results for 1740 participants (27.2%) disagreed.

A comparison of the results for the two IGRAs only (data were available for 7594 participants) showed that 1176 participants (15.5%) had a positive result for both, 5588 participants (73.6%) had a negative result for both, and 830 participants (10.9%) had discordant results. Tables 14 and 15 show the cross-tabulations of test results, and Figure 8 shows the completeness of data for each testing combination. Further cross-tabulations of results showing the full categorisation of results (including missing, indeterminate and error) are presented in Appendix 2, Tables 44 and 45. Cross-tabulations for participants who did and did not progress to active TB are shown in Appendix 2, Tables 46–48.

FIGURE 8.

Test result combinations. *T-SPOT.TB, QuantiFERON + TST^a; ^†T-SPOT.TB, QuantiFERON + TST^b. TST^a was considered positive if participants had a skin induration measuring ≥ 5 mm. TST^b was considered positive if participants without a previous BCG vaccination had a skin induration measuring ≥ 6 mm following the standard TST; for participants with a previous BCG vaccination, a skin induration of ≥ 15 mm following the standard TST indicated a positive result.

Progression to active tuberculosis

Of the total number of participants, 97 (1.0%) developed active TB. Of the participants progressing to active TB, 66 (68.0%) were aged ≤ 35 years and 63 (64.9%) were recruited to the study because of contact with an individual with TB. The follow-up duration for participants in the study ranged from 21 days to 5.9 years (for non-progressing participants: range 0.3–5.9 years, mean 2.9 years, median 3.0 years; for progressing participants: range 21 days–4.1 years, mean 0.8 years, median 0.5 years) (Figure 9).

FIGURE 9.

Histogram of follow-up duration for participants who progressed to active TB and participants who did not progress to active TB.

Fifty-two out of the 97 participants (53.6%) who progressed to active TB developed TB within 6 months of follow-up (33 contact participants and 19 new entrant participants), and within 12 months 70 participants (72.2%) had progressed to active TB (46 contact participants and 24 new entrant participants). Twenty-seven participants (27.8%) developed active TB after 1 year of follow-up.

The characteristics of the 97 participants who progressed to active TB are summarised in Table 16. Forty-five of these participants had pulmonary TB, with (n = 7) or without (n = 38) TB disease at an extrapulmonary site; the remaining 52 participants had extrapulmonary TB only. Approximately two-thirds of participants were contacts and 48.5% were female. A total of 81 out of 97 participants (83.5%) were born outside the UK. The majority of participants (74.2%) had received a BCG vaccination. Sixteen participants (16.5%) reported contact with a patient with TB prior to the contact that led to their enrolment in the study. At the time of enrolment, four of the participants (4.1%) who went on to develop active TB reported problem drug use and three (3.1%) were homeless.

There were 14 index case/contact pairs with 24-locus MIRU-VNTR data available for both individuals’ infections: 11 of 12 pairs with data on ≥ 22 loci were identical at all available loci and one pair was identical at the 12 loci available.

T-SPOT.TB, QFT-GIT, TST^a and TST^b identified 56 (57.7%), 53 (54.6%), 77 (79.4%) and 60 (61.9%) of the participants developing active TB as positive, respectively. Participants who progressed to active TB accounted for 3.6%, 2.8%, 2.2% and 3.5% of the positive results for T-SPOT.TB, QFT-GIT, TST^a and TST^b (PPVs), respectively, and 0.5%, 0.5%, 0.3% and 0.5% of the negative results for each test, respectively (NPVs of 99.5%, 99.5%, 99.7% and 99.5%) (Table 17).

For the participants aged ≤ 35 years, individuals developing active TB accounted for 4.7%, 3.9%, 2.7% and 4.6% of the positive results for T-SPOT.TB, QFT-GIT, TST^a and TST^b, respectively; for participants aged > 35 years, this was 2.4%, 1.8%, 1.6% and 2.2%, respectively. For contact participants, individuals who progressed to active TB accounted for 4.3%, 3.5%, 2.5% and 3.7% of the positive results for T-SPOT.TB, QFT-GIT, TST^a and TST^b, respectively; for the new entrant participants, this was 2.9%, 2.1%, 1.7% and 3.2%, respectively. Tables 17–19 provide further information.

Primary analyses

For the primary analyses, the data for the 6386 participants who had all test data were used, which included 77 participants who developed active TB. As six of these participants did not have data enabling time in the study to be calculated, the final sample used for the primary analyses was 6380 participants, and 77 of these participants developed TB.

Incidence rates and incidence rate ratios for individual tests

When utilising all follow-up data, the IR of TB per 1000 participants per annum for those with a positive result for T-SPOT.TB was 13.2 (95% CI 9.9 to 17.4), and it was 1.5 (95% CI 1.0 to 2.2) for those with a negative result. For QFT-GIT, the IRs for participants with positive and negative results were 10.1 (95% CI 7.4 to 13.4) and 1.9 (95% CI 1.3 to 2.7). For TST^a, the IRs for participants with positive and negative results were 6.8 (95% CI 5.2 to 8.7) and 1.2 (95% 0.6 to 2.0), respectively. For TST^b, the IR for participants with positive results was 11.1 (95% CI 8.3 to 14.6), and it was 1.6 (95% CI 1.0 to 2.3) for those with negative results (Table 20).

When restricting the follow-up of participants to just 1 year, 59 of the 77 participants (76.6%) who progressed to active TB in the total study duration had progressed at this stage. With the 1-year restricted follow-up, the IR of TB per 1000 participants per annum for participants with a positive T-SPOT.TB result was 34.0 (95% CI 24.4 to 46.2), and it was 3.5 (95% CI 2.1 to 35.1) for participants with a negative T-SPOT.TB result. For participants with a positive QFT-GIT test result, the IR was 25.4 (95% CI 17.8 to 35.1), and it was 4.7 (95% CI 3.0 to 7.0) for those with a negative result. For participants with a positive TST^a result, the IR was 17.1 (95% CI 12.7 to 22.6), and it was 2.6 (95% CI 1.2 to 5.0) for those with a positive result; for TST^b, the IR for participants with a positive result was 28.9 (95% CI 2.08 to 39.1), and it was 3.5 (95% CI 2.0 to 5.6) for those with a negative result (Table 21).

When extending the follow-up period to 2 years, the IR decreased. For participants with a positive T-SPOT.TB result, the IR was 20.5 (95% CI 15.1 to 27.2), and it was 2.1 (95% CI 1.3 to 3.2) for participants with a negative result. For participants who had a positive QFT-GIT result, the IR was 15.6 (95% CI 11.3 to 21.0), and it was 2.7 (95% CI 1.8 to 4.0) for those with a negative result. For TST^a, the IR for participants with a positive result was 10.0 (95% CI 7.6 to 13.0), and for those with a negative result it was 1.8 (95% CI 0.9 to 3.2); for TST^b, the IR for participants with a positive result was 16.6 (95% CI 1.2 to 22.1), and for those with a negative result it was 2.3 (95% CI 1.5 to 3.5).

The IRR (the incidence of TB for those with a positive test result compared with those with a negative test result, with adjustment for person-years) for T-SPOT.TB was 8.8 (95% CI 5.5 to 14.2), for QFT-GIT it was 5.4 (95% CI 3.4 to 8.5), for TST^a it was 5.8 (95% CI 3.2 to 10.6) and for TST^b it was 7.1 (95% CI 4.4 to 11.4). Similar IRRs were produced for each test when restricting the follow-up duration to 1 year and 2 years (see Tables 20 and 21).

Incidence rates and incidence rate ratios for test strategies

For the two-step strategies, the IRs for positive and negative results were 15.9 (95% CI 11.7 to 21.1) and 1.7 (95% CI 1.1 to 2.4) for TST^a + T-SPOT.TB, respectively; 13.2 (95% CI 9.6 to 17.8) and 2.0 (95% CI 1.4 to 2.7) for TST^a + QFT-GIT, respectively; and 13.4 (95% CI 10.0 to 17.6) and 1.6 (95% CI 1.0 to 2.3) for IGRA + TST^a, respectively. For the two-step strategies including TST^b, the IRs for positive and negative results were 19.6 (95% CI 14.2 to 26.4) and 1.9 (95% CI 1.3 to 2.6) for TST^b + T-SPOT.TB, respectively; 16.4 (95% CI 11.6 to 22.6) and 2.1 (95% CI 1.5 to 2.9) for TST^b + QFT-GIT, respectively; and 17.4 (95% CI 12.7 to 23.2) and 1.8 (95% CI 1.2 to 2.5) for TST^b + IGRA, respectively (Tables 22 and 23).

For the two-step strategies using results from each IGRA with the TST^a result, the calculated IRRs were 9.6 (95% CI 6.1 to 15.3) for TST^a + T-SPOT.TB, 6.7 (95% CI 4.3 to 10.5) for TST^a + QFT-GIT and 8.6 (95% CI 5.4 to 13.8) for IGRA + TST^a. For the two-step strategies using results from each IGRA with the TST^b result, the calculated IRRs were 10.6 (95% CI 6.7 to 16.6) for TST^b + T-SPOT.TB, 7.7 (95% CI 4.9 to 12.0) for TST^b + QFT-GIT and 9.8 (95% CI 6.2 to 15.4) for IGRA + TST^a (see Tables 22 and 23). See Appendix 2, Figures 12 and 13, for tables showing the results for each step of the investigated two-step strategies.

Comparative performance of tests and test strategies

Association of participant risk factors and progression to active tuberculosis

For the 6380 participants included in the main analysis, further analyses to identify associations of participants’ demographic characteristics and risk factors with progression to active TB identified that an increased body mass index (BMI) was associated with a lower IR of progression to active TB (IRR 0.92, 95% CI 0.86 to 0.97; p = 0.003). A borderline significant association was seen with haematological malignancy; however, only one participant with haematological malignancy progressed to active TB (Tables 26 and 27).

Test performance in new entrants and contacts

The test performance for new entrants and contacts (separately) is shown in Table 28; CIs for the IRRs among contacts and new entrants overlapped for all tests. Point estimates of IRRs were higher among new entrants than contacts for T-SPOT.TB, QFT-GIT and TST^b, but not for TST^a. The difference was more marked for T-SPOT.TB for new entrants (IRR 15.7, 95% CI 5.9 to 41.6) than for contacts (IRR 7.2, 95% CI 4.1 to 12.7).

Test performance by demographic group

Table 26 provides further subgroup analysis by age, gender and ethnic group, showing the number of participants progressing to active TB and the IRR for each demographic group. Although there is variation by subgroup, the observed differences were not significant.

Use of Mantoux testing

The difference between the two outcomes of Mantoux testing (TST^a and TST^b) is shown further in Appendix 2, with the combination of results highlighting the additional number of positive results for TST^a compared with TST^b.

Sensitivity analyses

Sensitivity analyses were performed excluding those participants with assumed BCG vaccination status, excluding participants with an unclear progression status, using all available test results rather than only using participants with all test results available, using data from new entrants with 2 years of follow-up only and using data for contacts of pulmonary TB patients only. The results of these analyses are provided in Appendix 2, Tables 50–55.

Stability of results by number of iterations

For the base-case analyses, we performed 5000 PSA iterations for each strategy, using bootstrap samples of 4162 patients for contacts and 3795 for new entrants for each iteration (reflecting the sample sizes available for economic analysis).

Figure 10 shows the incremental effects (mean discounted QALYs per patient for each strategy compared with the ‘no test’ strategy) for contacts and new entrants at increasing numbers of PSA iterations. The estimates and ranking of test strategies by incremental effect stabilise after around 3500 iterations in the contacts subgroup. Results are less stable in the new entrant subgroup, and there is still some swapping of position among strategies at the cut-off point of 5000 runs. Owing to the very small QALY differences between strategies, longer runs are required to rank strategies confidently according to effectiveness alone, but, at a runtime of about 2 hours per strategy for 5000 iterations, additional runs are computationally prohibitive.

FIGURE 10.

Incremental effectiveness (compared with ‘no test’) by number of iterations. (a) Contacts; and (b) migrants.

Figure 11 shows incremental net benefit (INB) estimates for the contacts and new entrant subgroups; these are the mean monetary values per patient for each strategy compared with the ‘no test’ strategy, with QALYs valued at £20,000 (the lower end of the cost-effectiveness threshold range used by NICE). It can be seen that INB is much more stable than QALYs, resolving after about 250 iterations. This suggests that INB differences between strategies are largely driven by cost differences rather than by QALYs.

FIGURE 11.

Incremental net benefit (compared with ‘no test’) by number of iterations. (a) Contacts; and (b) migrants.

Clinical outcomes

Key health outcomes from the base-case analysis are shown in Tables 29 and 30 for the contacts and new entrant subgroups, respectively. Strategies are ranked by undiscounted QALYs. It can be seen that without testing or preventative treatment (no test), the estimated mean life expectancy was approximately 47 years for contacts and 44 years for new entrants; these estimates reflect the lower mean age of contacts at baseline in the PREDICT cohort. The mean numbers of QALYs per participant in the contacts and new entrant subgroups were 46.01 and 42.96, respectively. The small differences between QALY and life-year estimates can be explained the high baseline EQ-5D scores for PREDICT participants. The simulated number of TB cases (over the lifetime horizon) per 10,000 people without testing was 189 for contacts and 179 for new entrants. This equates to an annual rate of approximately 42 cases per 100,000 people. For comparison, the rate of TB in the non-UK-born population of England in 2015 was 51 cases per 100,000 people. 116 However, the use of a fitted Weibull distribution in our model to estimate the long-term incidence from PREDICT data means that the simulated incidence was much higher in the first year and levelled off over time.

The QALY estimates were higher in all of the LTBI testing strategies than for the ‘no test’ strategy. This was generally because of the lower number of TB cases predicted, attributable to the use of preventative treatments (CPX). It should be noted, however, that QALYs are not always directly related to the number of TB cases because of QALY losses as a result of adverse reactions to CPX; for example, in the new entrant subgroup, although there are three fewer TB cases with TST^a than with T-SPOT.TB, there are also four more CPX-related adverse events, and, hence, mean QALYs are slightly lower with TST^a than with T-SPOT.TB.

A further ranking of strategies by the number of TB cases that were prevented (compared with ‘no test’), cases in which CPX was started and cases in which BCG vaccination was offered is presented in Table 31.

Cost-effectiveness

Tables 32 and 33 show the estimated mean discounted costs and QALYs per participant by strategy under the base-case analysis, for the two subgroups. The ‘no test’ strategy had the lowest costs and QALYs. The incremental costs for the LTBI testing strategies compared with ‘no test’ were between £156 (TST^b + QFT-GIT) and £273 (TST^a) for contacts and between £155 (TST^b + QFT-GIT) and £247 (TST^a) for new entrants. The additional cost yielded few extra QALYs; there was a mean gain of around 0.0002 per patient. Although none of the strategies under this base-case scenario appeared to be cost-effective compared with ‘no test’, the primary objective of this study was to assess the most cost-effective test or combination of tests and not to assess the cost-effectiveness of contact tracing or migrant tracing. Other QALY benefits will accrue from contact tracing and migrant screening through the detection of active TB and from the prevention of subsequent transmission, which we have not accounted for in the base case. Nevertheless, the comparison of strategies remains valid as the probable subsequent benefit was directly proportional to the extent to which each testing strategy prevented incident cases.

The most cost-effective of the LTBI strategies was TST^b + QFT-GIT because, although the QALY gain was lower than for some of the other strategies (e.g. TST^b alone in both subgroups), it was cheaper. The estimated INB differences between the LTBI test strategies were small. In particular, in both subgroups, very similar INB estimates were obtained for the three most cost-effective test strategies: (1) TST^b + QFT-GIT, (2) TST^b + T-SPOT.TB and (3) QFT-GIT alone. Despite this, the conclusion that TST^b + QFT-GIT was the most cost-effective of the LTBI test strategies was relatively robust in the PSA. In the contacts subgroup, the estimated probability that TST^b + QFT-GIT was the most cost-effective of the LTBI test strategies was 64%, followed by QFT-GIT alone (with 36% probability). Using a higher cost-effectiveness threshold of £30,000 per QALY gained, the estimated probabilities were 59% for TST^b + QFT-GIT and 39% for QFT-GIT alone. In the new entrant subgroup, the estimated probability that TST^b + QFT-GIT was the most cost-effective of the LTBI test strategies was 94% at a cost-effectiveness threshold of £20,000 per QALY and 87% at a cost-effectiveness threshold of £30,000 per QALY.

Doubled baseline incidence of tuberculosis in both subgroups

When the baseline incidence of TB was doubled for contacts (Table 34), compared with the base-case analysis (see Table 32) only the TST^a + QFT and TST^b + IGRA strategies swap positions. In the new entrants’ subgroup, doubling the baseline incidence of TB does not alter the position of strategies and the difference in probability of cost-effectiveness is marginal (Table 35). Although the increasing costs observed with doubling the TB incidence accrues from treating more TB cases, the lowered INBs suggest that the LTBI test strategies become more cost-effective at higher TB incidences.

Decreased cost per patient of T-SPOT.TB to equal that of QuantiFERON TB Gold In Tube

Compared with the base-case cost-effectiveness results, Tables 36 and 37 show that when the costs of T-SPOT.TB and QFT-GIT are equal, at £41, the T-SPOT.TB -based LTBI strategies become optimal. Among contacts, T-SPOT.TB was the most cost-effective strategy, whereas TST^b + T-SPOT.TB was the most cost-effective strategy among new entrants.

Further increased baseline incidence by multiples of 2 for TST^b + QuantiFERON TB Gold In-Tube compared with no test

In Tables 38 and 39, the baseline incidence of TB is increased by multiples of 2 and the comparison is only between TST^b + QFT-GIT and no test. This analysis was carried out to explore a TB incidence threshold or range for cost-effectiveness and for model validation, rather than to explore cost-effectiveness among LTBI test strategies. TST^b + QFT-GIT appeared to become cost-effective at a lower incidence in new entrants than in contacts. However, the high incidences that are required to achieve cost-effectiveness in both subgroups (compared with no test) are probably overestimated, as the analysis did not account for the benefits from LTBI testing, because the number of individuals who were offered CPX or a BCG vaccination was fixed at for all scenarios.

Changed chemoprophylaxis uptake to 59.9% and chemoprophylaxis completion rate to 57%, as stated in the Tuberculosis in England 2016 report2

We tested lower proportions of CPX uptake (59.9%) and CPX completion (57%) as reported in the Tuberculosis in England 2016 report. 2 Tables 40 and 41 indicate that, with fewer individuals being offered CPX and completing CPX, QFT-GIT was the most cost-effective LTBI strategy with a probability of cost-effectiveness of 98% for contacts and 84% for new entrants.

Used a 20-year time horizon

Compared with a lifetime horizon in the base case, Tables 42 and 43 show an almost identical ordering of strategies. However, there is a huge drop in QALYs, from about 22.0 to 14.0 in contacts and from about 21.0 to 13.6 in new entrants. Compared with QALYs, the drop in costs from the base case is negligible.

Allowing for transmission

The results for a modelled transmission scenario are reported in Appendix 2, Tables 58 and 59. Our approach to modelling transmission was based on the assumption that, for every case of active TB in the original cohorts (4162 in the contacts subgroup and 3795 in the new entrants subgroup), four new secondary contacts were traced. The traced contacts, drawn randomly from the original contacts cohort, were introduced into the model and subsequently followed the same pathway as individuals from the original cohort. For instance, of 4162 participants in the original cohort, 79 developed active TB (i.e. there were 79 index cases, each accounting for four secondary cases randomly drawn into the model) (see Appendix 2, Table 58). The number of secondary cases is slightly higher than the sum of 4 and 79 because some of the secondary contacts also develop active TB, resulting in additional participants being introduced into the model.

Diagnostic accuracy

We assessed the ability of 12 screening strategies to predict the development of active TB through latent TB detection that used IRRs comparing those with positive and negative screening results. The point estimate of the IR for active TB per annum was 10-fold or more higher in those with a positive IGRA result or in whom the TST^b strategy was applied. TST^a was less predictive, at about a sevenfold higher IR when comparing those with a positive test result and those with a negative test result. Although T-SPOT.TB and combinations of T-SPOT.TB with various thresholds of TST appeared to perform better, the confidence limits for all strategies overlapped.

The TST^a approach identified more progressors than any other strategy. The IRs in the TST^a-positive and TST^a-negative groups were lower than in the corresponding positive and negative groups identified by the other individual tests, although we did not find evidence that these differences were real rather than chance occurrences. The poorer predictive value of the TST^a approach than of IGRAs was consistent with the greater number of false-positive results because of a previous BCG vaccination; this was expected in this TST strategy that did not take prior vaccination into account. In the secondary analysis, assessing contacts and migrants separately, progression rates were higher among contacts than among migrants, although, again, this may have been attributable to chance.

Although no test was clearly superior to other tests or test strategies in all test dimensions, for PPV, all strategies appeared to be better than TST^a alone. The combination of T-SPOT.TB alone, QFT-GIT alone and TST^b (in that order) appears to be the most favourable option for predicting the development of active TB. Although the effect was more modest, combining TST^b with any of the IGRAs also appeared to be better than TST^b alone. The approaches using T-SPOT.TB or QFT-GIT alone were also not as good in their PPVs as strategies that combine TST with the respective IGRA.

For the NPVs of tests and strategies, all tests appeared to have high predictive values. Compared with TST^a + QFT-GIT, TST^a, T-SPOT.TB and TST^a + IGRA appeared to be the best approaches. Compared with TST^b + QFT-GIT, TST^a, TST^b, T-SPOT.TB, TST^a + T-SPOT.TB, TST^a + IGRA and TST^b + IGRA were better.

Cost-effectiveness

We developed a DES model to estimate the cost-effectiveness of alternative testing strategies for LTBI in two high-risk subgroups: contacts of people with active TB and recent migrants from high-incidence countries. The model used individual data from the PREDICT cohort to characterise baseline risk factors and test results in these subgroups, simulated the LTBI screening process and, when appropriate, preventative treatment with CPX or a BCG vaccination. The model accounted for missing and indeterminate test results, as observed in PREDICT, and simulated imperfect uptake and completion of preventative treatment and related adverse events. TB incidence was simulated over the individuals’ lifetimes, using available information on their actual test results and estimates of the prognostic value of tests from the statistical analyses, with adjustment for the effects of any preventative treatment. Public sector costs were estimated, including LTBI testing, preventative treatment and treatment of adverse events and active TB. Health outcomes were estimated in the form of QALYs, calculated from simulated survival and quality-of-life impact of incident TB and adverse events. The costs and QALYs were both discounted at the current recommended rate of 3.5% per year.

In these populations (contacts of people with active TB and recent migrants from high-incidence countries), although none of the test strategies was cost-effective, our base-case assumptions and parameters limited the application of this analysis to a comparison of strategies with no testing for the following reasons:

• The model did not reflect transmission of infection, and so would underestimate health gain and cost savings from TB prevention.

• The model did not account for cases of active TB detected at baseline, which, if asymptomatic, may be detected through further investigation following a positive LTBI test.

Partly attributable to these assumptions, the estimated QALY gains of 0.0002 per person were small, despite LTBI testing and treatment preventing TB cases, for an additional cost of between £150 and £270. The model predicted that LTBI testing and preventative treatment would prevent between 27 and 44 cases of TB in a population of 10,000 people over the lifetime horizon, depending on the test strategy. To achieve this, between 361 and 2192 people would need to start CPX and between 522 and 1146 people would need to have a BCG vaccination. In general, strategies with a higher number of individuals starting CPX had fewer patients developing active TB. However, diminishing returns were observed, with fewer additional cases prevented from successive increases in the number of individuals given CPX.

The most cost-effective LTBI test strategy was QFT-GIT following a positive TST^b result, which had an estimated cost per QALY gained (compared with no testing) of > £790,000 in new entrants or > £870,000 in contacts. The less discriminating strategies (i.e. the strategies that generate the greatest number of LTBI-positive results, such as TST^a), and consequently those in which the greatest numbers of patients received CPX, were generally the least cost-effective, suggesting that CPX costs are the main driver of cost-effectiveness. Although less obvious, the cost of the IGRAs does also influence cost-effectiveness. In particular, we noted that TST^b + QFT-GIT and TST^b + T-SPOT.TB prevented about the same number of TB cases and yielded similar QALY increases, but TST^b + QFT-GIT appeared to be more cost-effective as it was cheaper.

Head-to-head interferon gamma release assay versus tuberculin skin test comparison studies

This study provides a comprehensive comparison of latent TB testing strategies. Prior comparison studies have largely been in high-incidence countries, where the risk of reinfection limits the applicability of results to the UK. Four head-to-head comparison studies23,80,81,104 have been published in low-incidence countries. Only one of these studies81 compared the TST with each of the commercially available IGRAs; the other three23,80,104 utilised QFT-GIT alone. Sample sizes ranged from 339 to 1335 people, and all four studies included ≤ 15 individuals who progressed to active TB. None directly compared rates of TB development between individuals with positive and negative baseline test results.

Our study is the largest study of the predictive values of IGRAs published to date, in terms of both the number of participants and the number of progressions to active TB in a low-incidence setting. This has allowed us to quantify the prognostic value of IGRAs, in direct comparisons with the TST, with a high degree of precision and taking into account follow-up duration. The results are directly relevant to clinical practice in the UK because the study population comprised individuals who are being targeted for screening in national policies.

Of the four previous studies, two reported point estimates of PPV for IGRA that were higher than those for the TST (17% for IGRA vs. 2% for TST23 and 13% for IGRA vs. 5% for TST104). The other two studies reported a similar PPV for IGRA and TST (3% for IGRA vs. 2% for TST80 and 3% for IGRA vs. 3% for TST81). The TST threshold that was used in these studies differed. We found only limited differences between the PPVs of T-SPOT.TB, QFT-GIT, TST^a and TST^b, which ranged from 2.2% for TST^a to 3.6% for T-SPOT.TB. Our estimates of NPV were high (≥ 99.5%) for all four tests; this was consistent with the four previous studies, all of which reported NPVs of ≥ 98%.

Comparison with previous estimates of interferon gamma release assay predictive value

The pooled IRR (comparing individuals with a positive IGRA result with those with a negative result) from a previous meta-analysis was 2.11 (95% CI 1.29 to 3.46) for IGRA, 1.60 (95% CI 0.94 to 2.72) for TST with a 10-mm cut-off point and 1.43 (95% CI 0.75 to 2.72) for TST with a 5-mm cut-off point. 16 Results for IGRAs published since the systematic review117 included hazard ratios ranging from 2.054 to 42.4,29 odds ratios of 2.10–14.935,76 and rate ratios of 2.12–73.9. 24,75 Our estimates for T-SPOT.TB and QFT-GIT were 8.8 (95% CI 5.5 to 14.2) and 5.4 (95% CI 3.4 to 8.5), respectively; these were higher than the pooled estimate and broadly consistent with the less extreme estimates from the later studies.

Our estimates of PPV were at the lower end of those previously reported (see Table 1), but were fairly similar to the pooled estimate of 2.7% (95% CI 2.3% to 3.2%) from a meta-analysis of risk groups. 15 The PPV for the development of active TB depends in part on the incidence of TB in the study population, being higher when the incidence is higher; consistent with this, the pooled estimate among high-risk groups was 6.8% (95% CI 5.6% to 8.3%). 15 Although our study focused on high-risk groups within the UK, the incidence here is still relatively low compared with the high-burden countries in which many of the previous studies were conducted. Our estimates of NPV were similar to those previously published; for example, the meta-analysis produced a pooled estimate of 99.7% (95% CI 99.5% to 99.8%). 15

The PREDICT study therefore confirmed the relatively low PPV of IGRA for the development of active TB disease. Several factors contributed to this relatively poor performance. First, the test detects an immune response to M. tuberculosis antigens, rather than the presence of the organism itself. A positive result therefore did not necessarily indicate that the bacterium was present in the body and able to reactivate to cause disease. Second, although there were accepted cut-off values to define positive and negative IGRA results, categorical results may vary within the same individual, particularly if they were close to the cut-off value. 118 Further analysis of the PREDICT data will help to determine the utility of quantitative (rather than categorical) IGRA results in predicting active TB.