Footnotes:

Quality of evidence was rated as high (no points subtracted), moderate (1 point subtracted), low (2 points subtracted), or very low (>2 points subtracted) based on five criteria: imitations, indirectness, inconsistency, imprecision, and publication bias. For each outcome, the quality of evidence started at high when there were randomized controlled trials or high quality observational studies (cross-sectional or cohort studies with diagnostic uncertainty and direct comparison of test results with culture) and at moderate when these types of studies were absent. One point was subtracted when there was a serious issue identified or two points when there was a very serious issue identified in any of the criteria used to judge the quality of evidence.

9 studies were included: 1 study evaluated both T-SPOT.TB and QFT-GIT, 2 studies evaluated T-SPOT.TB, the 6 remaining studies evaluated QFT-GIT(n=5) or QFT-G(n=1).

2 out of 9 studies were unpublished and quality indicators could not be assessed; remaining study populations were considered to be representative, however, only 1 of the remaining 7 studies reported that assessment of test results was performed blinded to other test results. Only 2/7 reported the blood draw had been performed prior to the TST.

33% (3/9) studies were done in low-income settings and the remaining 6 studies were done in middle-income settings. Some indirectness in the choice of reference standard was observed.

Serious heterogeneity in characterization of exposure gradient (some based on index case's smear status, some based on sleeping proximity, etc.) and in estimated effect.

Majority of studies had 200-300 participants, smallest study n=120. Estimated 95%CIs were relatively tight.

Data included in the review did not allow for formal assessment of publication bias using methods such as funnel plots or regression tests. Therefore, publication bias could not be ruled out. Although no points were deducted, it was assumed that some degree of publication bias is likely because: 1) literature on IGRAs is rapidly exploding and currently unpublished studies may come out in future (although an attempt was made to include unpublished studies, despite not being comprehensive); 2) there are anecdotal examples of unpublished negative studies on IGRAs; 3) because a sizeable proportion of IGRA studies have some level of industry involvement or support, the risk of unpublished negative studies (or delayed publication of negative studies) is not trivial.

6 studies were identified: 1 study evaluated both T-SPOT.TB and QFT-G, while 1 study evaluated T-SPOT.TB. The remaining 4 studies all evaluated QFT-GIT.

Only the 4 published studies could be assessed for quality, 50% reported on timing of blood draw prior to TST, 50% reported blinding had been done for assessment of test results and 50% reported industry involvement.

All studies, except one done in low-income setting were done in upper-middle income settings. Some indirectness in the choice of reference standard was noted.

Serious heterogeneity in characterization of exposure gradient (some based on index case's smear status, some based on sleeping proximity, etc.) and in estimated effect.

All but one large study (n=2211) had between 82-301 participants. Studies estimated wide 95%CI, and majority were not significant.

Data included in the review did not allow for formal assessment of publication bias using methods such as funnel plots or regression tests. Therefore, publication bias could not be ruled out. Although no points were deducted, it was assumed some degree of publication bias was likely because: 1) literature on IGRAs is rapidly exploding and currently unpublished studies may come out in future (although an attempt was made to include unpublished studies, despite not being comprehensive); 2) there are anecdotal examples of unpublished negative studies on IGRAs; 3) because a sizeable proportion of IGRA studies have some level of industry involvement or support, the risk of unpublished negative studies (or delayed publication of negative studies) is not trivial.

2 studies were included; both studies evaluated the QFT, one study using the QFT-GIT and the other the QFT-G.

1 study was unpublished and hence not suitable for quality assessment; the other study was a longitudinal study that followed HCWs after a nosocomial infection. Population was representative, blood draw was done prior to TST, and there was no industry involvement, however, blinding was not reported.

Both studies were done in Upper middle income settings, however one was a nosocomial outbreak involving health care workers and may not be generalizeable to other contact settings including household contacts, especially in low income settings. While we did not downgrade for reference standard, we acknowledge there is some indirectness in the choice of reference standard.

Serious heterogeneity between estimated ORs for exposure and conversions, one study shows a positive association between conversions and exposure, while the other shows a significant protective effect of exposure for conversions.

95% CIs are tight and significant for the large unpublished (n=2211), however, CIs range from 0.18-21.12 and 0.69-122.38 for the smaller hospital outbreak study (n=39)

Data included did not allow for formal assessment of publication bias using methods such as funnel plots or regression tests. Therefore, publication bias cannot be ruled out. Although we did not deduct points, we assumed some degree of publication bias is likely because: 1) literature on IGRAs is rapidly exploding and currently unpublished studies may come out in future (although we made an attempt to include unpublished studies, our attempt was not comprehensive); 2) there are anecdotal examples of unpublished negative studies on IGRAs; 3) because a sizeable proportion of IGRA studies have some level of industry involvement or support, the risk of unpublished negative studies (or delayed publication of negative studies) is not trivial.

2 studies included both IGRAs, 3 studies evaluated only T-SPOT.TB, while the rest evaluated a version of the QFT.

11/14 studies did not report on whether personnel assessing test results had been blinded to previous test results or reference standard and 5/14 studies reported industry involvement.

Studies were conducted in low and middle income settings. TB exposure gradient does not necessarily classify the target condition (LTBI) correctly.

47% of studies showed moderate agreement, while 26.5% showed poor agreement and 26.5% fair agreement. In 68% of comparisons, TST estimated a higher prevalence while in the remaining 32% IGRAs estimated a higher prevalence of LTBI.

Due to heterogeneity in effect estimates concordance could not be pooled. However, effects estimated for individual studies were frequently not significant.

Data included did not allow for formal assessment of publication bias using methods such as funnel plots or regression tests. Therefore, publication bias cannot be ruled out. Although points were not deducted, a degree of publication bias is likely because: 1) literature on IGRAs is rapidly exploding and currently unpublished studies may come out in future (although we made an attempt to include unpublished studies, our attempt was not comprehensive); 2) there are anecdotal examples of unpublished negative studies on IGRAs; 3) because a sizeable proportion of IGRA studies have some level of industry involvement or support, the risk of unpublished negative studies (or delayed publication of negative studies) is not trivial.