Our aim was to use the T-SPOT.TB assay to determine M.tb infection prevalence among M.tb exposed and unexposed five year old children in Entebbe, Uganda. We anticipated that positive responses would be confirmed with a second T-SPOT.TB assay and a TST. However, we have demonstrated a high level of instability in positive T-SPOT.TB responses between baseline and a three week follow-up and poor agreement between T-SPOT.TB and TST responses, making the categorisation of children as LTB-infected or LTB-uninfected difficult.
Analysis of SFUs showed that positive responses were not concentrated around the diagnostic test cut-off level for children whose T-SPOT.TB result varied between the first and second tests. Instead, some children showed large changes in response – both increases and decreases in SFUs – between the two assessments. In contrast, ESAT-6 and CFP-10 spot counts correlated well at each time point. Together these findings suggest that our results may reflect a true change in the immune response to the M.tb
antigens in peripheral blood. Several other studies, albeit with longer term follow-up, have observed variations in responses between baseline and follow-up 
. In a bid to explain changes on follow-up when using IGRAs, Hill et al.
suggested that IGRA responses are not long-lived and generally require sustained, continuous exposure to M.tb
antigens to maintain high frequencies 
. They hypothesized that the decline in response following exposure to M.tb
may be a reflection of the lifecycle of M.tb
and the dynamic interaction with the host immune system. As the mycobacteria enter a state of dormancy, secretion of ESAT-6 and CFP-10 may decline leading to a decrease in circulating memory T cells specific for the antigens used in the assay 
. The weak association between T-SPOT.TB and TST in our study raises a further question as to whether the responses observed indicate LTBI at all. It may be that the unstable positive T-SPOT.TB responses in this young age group reflect a weak (perhaps transient) response to M.tb
antigens in BCG-immunised children who resist the establishment of latent infection. However, the short time between the two assessments in our study make it unlikely that a change in TB exposure could account for the effects observed.
Some studies have described a rather high rate of changes from negative to positive IGRA response during follow-up in high-risk populations in endemic settings 
. Incident infections might explain this, especially in studies involving contacts of TB patients, or, again, the intermittent secretion of ESAT-6 and CFP-10 by M.tb
. However, in our study, changes from negative to positive were very few compared to changes from positive to negative.
Fluctuation in IGRA results upon serial testing has sometimes been attributed to technical factors. These factors may include blood volume (for QuantiFERON assays), different staff performing the assay, preanalytical delays and reagents. The T-SPOT.TB assay uses blood volumes as low as 2 ml; we collected 4 ml on average. Most of the assays were performed by the study laboratory technologists (GN and JEL), and any other staff performing the assay were given specialist training and worked under the supervision of the study laboratory technologists. Doberne et al.
recently demonstrated that preanalytical delay resulted in increased positive-to-negative reversions in as little as six hours. Such delays were not characteristic of our study (median time after sample collection
1.4 hours, IQR 50 min-2.2 hours), so time to processing is unlikely to explain our observed fluctuations. A recent systematic review 
showed that tuberculin skin testing has a boosting effect on IGRA responses, but this cannot explain the observed fluctuations in our study because we drew blood before performing the TST.
Longitudinal assessments for IGRAs in published literature have mainly been in contacts of TB patients 
. Our study investigated children, most of whom had no known TB contacts. We hypothesize that the intense exposure that TB contacts experience, compared to non-contacts, may explain the more stable responses observed in those studies. In keeping with this hypothesis, we found that household TB contacts in our study showed better agreement between the baseline and repeat test although the number of contacts was too small to provide good power for sub-group analyses.
T-SPOT.TB has been dubbed the “100-year upgrade” to the well-established TST for the diagnosis of TB infection 
. However, we report a high level of disagreement between the TST and T-SPOT.TB in our cohort of children. Previous studies have shown that levels of agreement are varied depending on the study and outcome measurement. For example, in four investigations that analysed agreement between the two tests, the κ scores ranged from -0.15 to 0.76 
. In our group, recent unpublished data from adult women in Entebbe showed that T-SPOT.TB performed better as an indicator of LTBI among adults. Among 23 women who tested T-SPOT.TB positive, 21 (91%) were TST positive as well. All these findings support the perception that agreement between the TST and the IGRA in the diagnosis of TB infection might vary depending on several factors such as age, history of previous BCG vaccination 
, and infection with other mycobacteria.
The immunological inferences that can be drawn from the observed discordance between T-SPOT.TB and TST in our study are unclear. TST results may be falsely negative in children due to the influence of factors such as malnutrition, concurrent viral and/or parasitic infections, and concurrent medical conditions and diseases 
. However, these factors were not characteristic of our participants. For example, we had TST data for six of the 13 HIV positive participants, and none of these had discordant T-SPOT.TB and TST results. The children in this study were BCG-immunised, making it possible that the discordances observed were due to false TST positives, but this is unlikely because we observed fewer positives by TST than by T-SPOT.TB. Furthermore, there was a general lack of intermediate sized TST responses, which have been attributed to BCG vaccination and infection with mycobacteria other than tuberculosis 
. The discordances may therefore be due to the unstable T-SPOT.TB responses between baseline and follow-up, rather than a result of the known shortcomings of the TST.
Tuberculin surveys carried out in the 1970s suggested that the annual risk infection (ARI) with TB in Uganda was approximately 3%, although rates were lower than this in young children 
. Recent national reports and other studies continue to quote this estimate 
although some surveys have indicated a steady decline in ARI – for example, a survey in northern Uganda in 1994 estimated ARI at 1.4% 
. Our results suggest that this may now be a more realistic figure for young children in Central Uganda also. However, our study cohort was derived from a small area, and those who participated in this study showed some biases. New M.tb
infection surveys of broader scope may be warranted for Uganda.
The principal limitation of our study was the change in protocol, resulting in variation in sample sizes for T-SPOT.TB at five years (n
907), repeat T-SPOT.TB at follow-up (n
405) and TST (n
319) making it difficult to compare the three tests directly. Secondly, the proportions of children followed up were different depending on the child’s T-SPOT.TB result at five years: this was expected with Protocol 1 but not with Protocol 2, where 85.7% of the children who were initially positive were followed up compared to 63.6% of those who were initially negative. Data analysis was therefore biased towards children with a positive first T-SPOT.TB result. We attempted to overcome this limitation by presenting results from Protocol 1 and Protocol 2 separately where it was relevant, and by comparing the participants’ characteristics between the two protocols.
Our data provide a valuable insight into the usefulness of IGRAs in the diagnosis of TB infection among children living in endemic settings. It has been suggested that recommendations on use of IGRAs in children younger than five years and in immunocompromised children should be taken with caution because of a lack of adequate published data on their efficacy in these groups 
. Our study has contributed to the increasing evidence that IGRAs may not be superior to TST in children in high incidence settings and cannot be used alone to diagnose TB infection in these settings. Diagnosis of TB infection and estimating TB infection prevalence among children in high incidence settings remains a challenge; better diagnostic tests are still needed.