A Systematic Review On Correlates of Risk of TB Disease In Children And Adults

BACKGROUND
Tuberculosis (TB) remains one of the leading causes of death in the world. Targeted treatment to prevent progression from TB exposure and infection to disease is a key element of WHO End-TB strategy. A systematic review to identify and develop correlates of risk (COR) of TB disease is timely.


METHOD
EMBASE, MEDLINE, PUBMED were searched using relevant keywords and MeSH terms published between 2000 and 2020 on COR of TB disease in children and adults. Preferred Reporting Items for Systematic reviews and Meta-analysis (PRISMA) framework was used for structuring and reporting of outcomes. Risk of bias was assessed using Quality Assessment of Diagnostic Accuracy Studies tool-2 (QUADAS-2).


RESULTS
4105 studies were identified. Following eligibility screening, 27 studies were quality assessed. Risk of bias was high in all studies. Broad variations in COR type, study population, methodology and result reporting were observed. Tuberculin skin test (TST) and interferon gamma release essays (IGRA) are poor COR. Transcriptomic signatures although promising require validation studies to assess wider applicability. Performance consistency of other CORs-cell marker, cytokines and metabolites are much needed.


CONCLUSION
This review identifies the need for a standardized approach to identify a universally applicable COR signature to achieve the WHO END-TB targets.

TB affects all age groups and once infected progression to disease is greater in the rst two years after being infected 3 . Children (< 5-years of age), individuals with HIV and those immune-suppressed are at increased risk of progression to TB disease 4 . Malnutrition and vitamin D de ciency have been associated with risk of TB disease progression (reference). Genetic de ciencies related to IFNγ and IL-12 are linked to an increased susceptibility to less virulent mycobacteria that causes severe disseminated and/ or recurrent infections termed as mendelian susceptibility to mycobacterial diseases (MSMD) 5 .
Current available TB screening tests, tuberculin skin test (TST) and interferon gamma release assays (IGRA) identify host immune responses suggestive of either an exposure to TB or evidence of a chronic non-progressive state. Uncertainty in screening is common, particularly when cross-reaction with other atypical mycobacterium occurs (TST) or an indeterminate IGRA result is encountered.
Bacteriological con rmation (culture or PCR) is the gold standard for TB disease diagnosis.
Mycobacterial Growth Indicator Tube (MGIT-BACTEC) a liquid culture is a sensitive method to isolate mycobacterium in comparison to the solid culture method (Lowenstein-Jenson medium) 6 . Xpert MTB/RIF (Nucleic Acid Ampli cation test) by Cepheid, was endorsed by WHO in 2011 as the next generation assay to diagnose TB disease. An advantage over culture is that it can detect Mycobacterium TB complex and resistance to rifampicin in less than two hours. Xpert MTB/RIF is reported to have a sensitivity of 64% in sputum samples and a speci city of 98% in smear positive samples 7 . The nextgeneration assay, Xpert MTB/RIF Ultra incorporates a PCR ampli cation technique and two additional molecular targets for Mycobacterium tuberculosis. In adults, Xpert MTB/RIF Ultra studies demonstrated an improved sensitivity both in smear positive and smear negative specimen 8 . Two accuracy studies in children conducted on cryo-preserved samples showed an increment in sensitivity by 11%, and a speci city of 97% 9 .
Considering a fourth of the world's population are latently infected, offering prophylactic treatment for such huge numbers is impractical. A consensus meeting report (2017) by WHO and the Foundation for Innovative New Diagnostics (FIND) on the development of a Target Product Pro le (TPP) for a test predicting the progression to TB disease outlined operational and performance characteristics for predictive tests and recommendations for standardization of future studies. A minimum sensitivity and speci city for a TB predictive test of ≥ 75% and speci city of ≥ 90% was set 10 .
To date a few systematic reviews on biomarkers with TB disease diagnostic potential have been published 11 12 . One systematic review /metanalysis on patient level pooled transcriptomic signatures for diagnosis of incipient TB state was published by Gupta et al. (2020). This evaluated only transcriptomic signatures as TB disease predictors on adult blood samples overlooking other potential serological markers.
In this systematic review, we aim to evaluate outcomes of studies on non-sputum-based COR of TB disease in children and adults, compare the results against the TPP targets for a TB prognostic test and identify the knowledge gaps for future exploratory studies.

Method
A systematic review on COR TB disease in children and adults published between January 1st, 2000 to May 25th, 2020 was conducted. EMBASE, MEDLINE, PUBMED were searched for relevant publications. In addition, key journals were searched. English language, year of publication and age groups [infants-(0-1 year), children (1-10 years), adolescents (10-19 years) and adults (19-65 years)] were used as lters. For each database, the search terms were transposed as appropriate.

Eligibility criteria
To ensure the selection of relevant studies for this review, the study selection was guided by the eligibility criteria as speci ed under the inclusion/exclusion criteria.

Inclusion criteria
We included studies that met the following criteria: 1. Study question-studies that addressed the review question i.e. COR of TB disease progression were included. 2. Study type-Case-control, longitudinal cohort studies that utilized new study samples/subjects and studies that included prior recruited study samples/subjects were included.
3. Study subjects-studies with human subjects were included. And ages between 0-65 years were included.
4. COR-studies that included COR derived from host immune response to TB mycobacterium were included.

Exclusion criteria
Studies on correlates of treatment (COT) and correlates of diagnosis (COD) of latent and TB disease were excluded. Cross-sectional studies, conference papers and review articles were excluded. Studies on animal subjects were excluded.

Study Selection
A structured preparation and reporting of systematic review in line with Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines was performed 13 .
The data extraction criteria were developed electronically using google forms. The Table-2 shows the data extraction form that was used.
Titles of the studies were screened by one of the authors (PS) and exclusions were made based on-target condition (non-LTBI /TB infection and non-TB disease were excluded), and publication type (review articles, meta-analysis and conference abstracts were excluded). Abstracts and full-text studies were reviewed for eligibility by PS and RR independently. Inter-reviewer disagreements were resolved through consensus decision process overviewed by SW and RS. Data extraction was completed by PS utilizing google forms. For the purpose of standardization and consistency, one signalling question was selected in each domain of the QUADAS-2 framework and this process was completed by PS. In domain 1 (patient selection), signalling question on consecutive or convenience sampling was considered relevant to assess the sampling strategy. Under domain 2, signalling question on the conduct and interpretation of the index test (i.e. blinding) was chosen to be relevant. "Index test" indicates the biomarker test of prognostication. Under domain 3, signalling question on conduct and interpretation of the reference standard was considered relevant. "Reference standard" indicates either a culture proven or an agreed composite standard to classify latent TB and/ or TB disease states was used. Under Domain 4, signalling question on patient ow was regarded relevant. The terms "Yes", "No" and "Unclear" were stated for each study for each domain.
If all responses in all four domains were "yes", then the risk of bias was considered "low" and "low applicability concerns", however if any of the responses were "no" or "unclear" then the risk of bias was considered "high" and hence "high applicability concerns". 2)Index test -(Signalling question-was the conduct and interpretation of the index test blinded?) Overall, eight studies clearly stated and implemented the process of blinding 30 31 32 33 34 35 . No clarity on the process of blinding in either the conduct or interpretation in ve studies 14 17 28 32 . Fourteen studies had no documented process of blinding in their conduct and interpretation.
3)Reference standard: (Signalling question-the reference standard likely to correctly classify the target condition (i.e. LTBI/ TB infection and TB disease). Twelve studies speci ed use of microbiological culture or a composite standard as a reference 19  Hence a "high risk" of bias was observed in the studies and hence a "high applicability" concerns. (Figure:3)

Characteristics of studies
Year of publication All studies included were published between 2007 and 2020. Nineteen (70%) studies were published in the last ve-years (2016-2020). Thirteen (48%) were conducted in sub-Saharan Africa, seven in Europe, six in Asia and one in South America. Based on the WHO classi cation of TB endemicity-twenty (74%) studies were conducted in high TB incidence countries (Incidence rate >40 per 100,000 population) and seven (26%) studies were conducted in low TB incidence countries (incidence rate <40 per 100,000 population).

Study population
Mean sample size in the twenty-seven studies was 745(n). Mean number of study subjects who developed TB disease during the study period was 42 (n). Nineteen (70%) studies included adult, children and adolescent groups. Five studies included the Adolescent Cohort Study (ACS) with 6363 healthy adolescents aged between 12-18years in South Africa 30 33 35 36 38 . Four studies included the Grand Challenges 6-74 (GC6-74) cohort with 4462 healthy household contacts of 1098 index TB cases in Africa 33 .Median duration of studies was two years (IQR-1.5 -5 years). Twenty (74%) studies evaluated >1 COR for TB disease (Table-1).

A. Analysis based on COR type i)Tuberculin skin test (TST) based
Five cohort studies reported TST as a COR of TB disease 20 21 23 27 31 . The risk of bias was high in all studies. Four were conducted in low endemic countries and one in a high endemic country. Despite a large sample size, the proportion developing TB disease was low (range 0.9-2.5%). As a COR for TB disease, a pooled sensitivity of 94% and speci city of 44% was calculated.
ii)IGRA based Ten cohort studies reported on IGRA 17 19 20 21 23 24 25 29 31 40 . The risk of bias was high in all studies.
Seven were conducted in low endemic countries. Despite a large sample size, the proportion developing disease was low (Range-0 -3.1%). As a COR for TB disease, a pooled sensitivity of 90% and speci city of 86% was calculated. PPV was reported to be <4% in four studies.

iii)Cytokine based
Nine studies on cytokines were identi ed 15 16 17 18 26 28 30 37 38 . Five were case-control and four were cohort studies. Eight were conducted in high TB endemic regions. A high risk of bias was noted in these studies. Three studies had large sample sizes, however the percentage who progressed to have TB disease was <1%. Four studies were on IL-10 based on children and adults and demonstrated no consistent correlation to TB disease progression 40

v)Antibody based
Two antibodies-based studies in adults were identi ed. Both studies had high risk of bias 17 22 . The proportion who developed TB disease at the end of the study period were negligible. Hence outcomes reported were not signi cant.

vi)Transcriptional signature based
Eight transcriptional signature studies were conducted in high TB endemic countries 14

B. Performance of COR in high-risk groups
Twelve studies included high risk groups-ten studies on infants and children, one on HIV infected adults, one on HIV exposed and un-infected infants, one on malnourished adolescents and adults and one on Rheumatoid Arthritis (RA) adults. A lack of correlation (not statistically signi cant) of IFN against Heparin binding haemagglutinin antigen (HBHA) in HIV +ve adults was observed. A higher neopterin levels in RA patients who developed TB was observed (median -24.5pg/ml and 23053pg/ml respectively) and non-RA TB patients (12.2pg/ml and 9633pg/ml respectively) compared with QFT-Gold converters without TB (3pg/ml and 2720pg/ml respectively both p<0.01). Alpha-2 macroglobulin, sero-transferrin, haptoglobin levels in malnourished endemic population demonstrated a two-fold increase in levels, however the number/s followed up were negligible.

C. COR and WHO TPP targets
WHO and FIND set an optimal TPP for a TB predictive test with a sensitivity and speci city of ≥90% or a minimum of ≥75% (within 2-years of exposure) and a high PPV with low NNT. Outcomes of ve included studies could be compared against the WHO TPP targets (cytokine signature-3PR, TRM-5 30 , transcriptomic signatures-16-gene 33 , 3-gene 36 , PREDICT-29 14 and metabolomic signature-TB-healthy 34 ) The performance of cytokine signatures as COR were comparable in the test sets but fell short in the validation sets. Similarly, the performance of 16-gene signature and PREDICT29 signature fell short of comparison in the validation sets. The 3-gene transcriptomic signature showed consistency in its performance in the nested cohort (validation set) and was comparable to the targets set for a TB predictive test. (Figure:4)

Discussion
We conducted a systematic review on blood-based, host-derived COR for TB disease in children and adults in accordance with PRISMA framework. We intended to analyze the outcomes reported by studies on TB COR and compare their performance against the WHO/FIND TPP targets. TST and IGRA are poor COR of TB disease progression. A meta-analysis reported a PPV of 2.4% for TST and 6.8% for IGRA, implying the number needed to treat (NNT) to prevent one case of TB disease to be as high as 67.3 for TST and 37.3 for IGRA 25 . Among the cytokine studies, IL-10 (anti-in ammatory) was analyzed in ve studies that demonstrated poor co-relation to TB disease progression 1718293738 . Complement components 3038 and M:L ratio 272939 demonstrated a good correlation to TB progression, however larger prospective studies in diverse populations are required. Activated T-cell as a COR can be confounded by several factors that include environmental triggers and childhood immunization and hence this is a poor predictor in children 28 . Studies on transcriptomic signatures have demonstrated the most encouraging results to date 14323335363738 . However, their predictive performance when validated in geographically diverse cohorts reported inconsistencies. In addition, their predictive performance improved closer towards TB disease diagnosis which reduces the usefulness for early detection of incident TB for targeted preventive TB treatment. A meta-analysis on eight transcriptomic signatures reported comparable results (at pre-test probability of 2%), PPV between 6.8-9.4% over 24 months and 11.2-14.4% over 3-months before TB disease diagnosis 41 . The recent CORTIS trial, a randomised open label study that tested the prognostic accuracy of RISK-11 signature (derived from Zak-16 signature) and e cacy of targeted TB preventive treatment reported that the signature performed well as a COR, however, the targeted preventive TB treatment in comparison to signature positives and negatives showed poor e cacy in preventing TB disease during the 15-month follow-up 42 . Though this study incorporated diverse population in South Africa, it raises a few questions on the accuracy of the signature as a predictor of TB disease, the targeted preventive TB treatment approach and tackling the distinct complexities in high and low TB endemic regions. This review has some potential limitations. The restricted search terminologies might have resulted in a narrow search. Only pulmonary TB studies were included in the search terminology hence that excluded studies on extra-pulmonary TB. All blood-based biomarkers were included in this review, making this review very broad. However, this review consolidates information on most potential COR studied till date. A narrative analysis was entailed due to the heterogenous nature of the studies which highlights the need for a standardized reporting for future studies in this eld. A large number of studies incorporated highrisk groups, however the outcomes cannot be extrapolated into clinical practice due to the relatively small sample sizes and even smaller proportion of those evolving into TB disease. This highlights the practical challenges in conducting large longitudinal observational studies.

Conclusions
A worldwide accelerated effort to derive at a blood-based, host-derived COR for TB disease has been observed in the last two decades. TB disease progression is marked by a cascade of events and the studies on cell markers, metabolomic and transcriptomic signatures identi ed in this review have shown good potential as COR. As future prospects, validation studies on combination of these COR (as a signature) would be ideal.

Declarations Con icting interest
The authors declare that they have no potential con ict of interest regarding the submitted manuscript.

Funding statement
This research received no speci c grant from any funding agency in the public, commercial, or not-forpro t sectors.
Ethical approval and consent to participate Not applicable.

Consent for publication
All authors have read the nal manuscript and provided their consent for publication.
Author's Contribution P.S, S.W and R.S envisaged the idea for the review and its purpose. P.S and P.P conducted the systematic literature search. P.S and R.R screened the studies as per PRISMA framework. P.S assessed the quality of the studies based on the QUADAS II framework overviewed by S.W and R.S. P.S and A.C wrote the manuscript text and prepared all gures and tables. All authors reviewed the manuscript   Sensitivity and Speci city for COR compared against WHO TPP Targets

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download.