Introduction of the GeneXpert MTB/RIF (Xpert) assay has constituted a major breakthrough for tuberculosis (TB) diagnostics, providing a rapid and accurate way of identifying TB patients in high TB-burden, low-income countries [1, 2]. Nevertheless, post-implementation studies have identified several challenges [3–6], emphasizing the need for deeper understanding of clinical and operational factors affecting real-world performance [7]. Previous studies have shown that pauci-bacillary forms of TB are more commonly identified in patients who are HIV-seropositive [8–10] and in those who are sputum acid-fast bacilli (AFB) smear-negative, and may reduce the overall sensitivity of Xpert in reference to the standard of mycobacterial culture [11–13]. However, because microscopy is also less sensitive in these populations, these groups are also the ones most likely to benefit from Xpert, and in whom Xpert has been especially recommended [14, 15].

A recent systematic review identified no studies describing the effect of sputum quality on Xpert performance [16, 17]. This is surprising because international guidelines have long emphasized macroscopic sputum quality as an important determinant of performance of smear microscopy and culture [16, 18]. Furthermore, previous small studies have shown salivary samples may be unsuitable for Xpert testing [19, 20]. Therefore, we sought to determine the effect of sputum quality on the diagnostic performance of Xpert in a large cohort of AFB smear-negative, presumed pulmonary TB patients in Kampala, Uganda.

Specimen quality has long been assumed to be as an important predictor of the performance characteristics of microbiologic tests, particularly those used to diagnose lower respiratory-tract infections. Unfortunately, the amount and quality of evidence about how sputum quality affects the performance of TB diagnostic tests is limited. In this prospective cross-sectional study, we found no significant difference in diagnostic yield of Xpert testing between salivary and non-salivary specimens among adults with negative sputum AFB-smear examinations in a low-income country with high burdens of TB and HIV. In fact, we identified a strong trend towards a higher diagnostic yield in salivary than in non-salivary specimens. These differences were confirmed by a secondary comparison of diagnostic accuracy in reference to mycobacterial culture. This analysis showed significantly higher diagnostic sensitivity of Xpert on salivary samples as compared with the referent category, mucoid sputum samples, while blood-stained sputum was associated with significantly lower sensitivity.

Macroscopic quality has long been emphasized in guidelines on the use of smear microscopy in TB evaluation. Despite this emphasis, there is only one published study of sputum quality and smear microscopy, which demonstrated substantially higher sensitivity with purulent or bloody sputum as compared with mucoid or salivary sputum among 170 TB patients [20]. However, 40% had culture-negative TB, and it is unclear if these associations apply equally to patients with microbiologically confirmed TB, or be relevant to less pauci-bacillary populations. While a recent systematic review found no studies on the influence of sputum quality on the performance of Xpert [16], we identified two subsequently published studies addressing this question. One compared diagnostic sensitivity and specificity by sputum type among 136 culture-confirmed TB patients and 703 culture-negative non-TB patients in Kenya, but there were only a few modest differences by sample type and none of these reached statistical significance [19]. The second study enrolled over 21,000 household contacts in Vietnam but almost all samples collected were mucoid, which prevented meaningful comparisons with other sample types [17].

Thus, our finding that salivary sputum does not have lower but perhaps higher diagnostic yield when testing for TB with Xpert may have great clinical importance. Salivary sputum has been considered unsuitable for examination by smear microscopy, and therefore laboratory staff have historically been trained to discourage patients from producing and submitting salivary sputum samples in preference for other sample types. Our results suggest that the conventional assumptions that salivary sputum is of lower quality and bloody sputum of higher quality for smear microscopy do not apply to samples tested with Xpert. Given the observational study design, we were not able to explore reasons why salivary sputum may provide greater sensitivity than other samples. Potential reasons could include: a greater bacillary load of MTB DNA in saliva than in other sample types; a greater recovery of MTB DNA from salivary sputum than from more viscous sample types; or more efficient amplification of MTB DNA from saliva than from samples such as sputum that have a more complicated specimen matrix that could include inhibitors of amplification. The last explanation is unlikely because the Xpert assay includes a positive control to detect inhibitors in all samples.

Our findings about the enhanced yield of salivary sputum may be of additional importance in high HIV-burden areas because of the inverse association we identified between CD4 count and the likelihood of producing salivary sputum. Because those with lower CD4 counts are also more likely to develop TB and to be smear-negative, it is crucial that they be tested with a diagnostic tool that has high sensitivity and high likelihood of obtaining a true positive result [15, 26]. Since salivary sputum had the highest diagnostic sensitivity of any specimen type tested, salivary samples should not be rejected for Xpert testing. Additional studies might explore whether saliva, particularly when obtained after coughing and prior to eating or oral care, can provide comparable diagnostic sensitivity to that of other sputum types [27].

In contrast, blood-stained sputum appears less desirable for Xpert testing in smear-negative populations. Xpert testing of blood-stained sputum missed twice as many TB cases as it diagnosed. A potential explanation for lower sensitivity could be that blood is a known inhibitor of DNA amplification, although this is less likely because by design Xpert inhibition should be detected by failed amplification of the internal positive control and reported as “Invalid”[28]. Nevertheless, our results do raise concern about the suitability of bloody sputum for molecular testing, and merit further investigation to identify the mechanisms underlying the low sensitivity. Until then, practitioners may consider attempting to obtain a non-bloody sample if a bloody specimen tests Xpert-negative.

Our study had some limitations. First, a majority of samples were mucoid (73%), resulting in small sample sizes for other sputum types and relatively large confidence intervals for all study measures. In particular, our primary analysis comparing diagnostic yield by specimen type is underpowered, because the 95% confidence intervals for yield differences include clinically important effects. Nevertheless, our sub-analyses were sufficiently powered to detect meaningful differences in sensitivity among three of four sputum types. Second, our sub-analyses could have been biased if the yield of sputum culture is also influenced by specimen quality. However, direct comparisons showed no difference in culture yield, and even if underpowered, the similarities in effect size and direction of our yield and accuracy analyses make this unlikely. Furthermore, we may have misclassified some culture-negative TB patients as not having TB, since we utilized solid rather than liquid culture media [29]. However, since our gold standard was rigorously determined utilizing multiple mycobacterial culture samples, we believe that this misclassification is minor and would not substantially bias our estimates. We may also have occasionally misclassified sputum quality. However, we utilized well-trained laboratory technologists to assess macroscopic sputum quality and provided them with visual aids to promote accurate readings. Therefore, any misclassification would likely be non-differential and unlikely to impact our results. Finally, our study focused on sputum smear-negative patients, but Xpert is now recommended as the first-line test for TB regardless of smear-status [15]. However, in many settings, Xpert testing continues to be limited to smear-negative patients due to resource constraints.

Our study also had many strengths. First, we carried out our study in a relevant population, possible TB patients in a low-income country with a high TB burden. This helps make our results generalizable to many other populations being tested with Xpert in high-burden, resource-limited settings. Second, our study is the first of sufficient size and power to provide meaningful comparisons of the effects of sputum quality on both Xpert diagnostic accuracy and Xpert diagnostic yield. We therefore believe our study fills a crucial gap in understanding Xpert testing.

In conclusion, for patients who are smear-negative, utilizing Xpert may provide a rapid diagnosis that might have otherwise been missed. As it replaces smear microscopy in an increasing number of high-burden countries, it has the potential to reduce the time and number of visits needed to obtain a diagnosis. Our findings, however, demonstrate the need to exercise caution in collecting sputum for Xpert and in interpreting results because sputum quality may impact test yield and sensitivity differently from what has been traditionally taught for smear microscopy. In particular, it may be wise to pursue additional testing should a blood-stained sputum test negative, especially in high TB-burden communities. In addition, laboratory staff should not reject salivary sputum for Xpert testing but accept it readily given its higher sensitivity and potentially higher yield than other sample types. Future studies attempting to replicate these findings and examining additional factors that may impact Xpert diagnostic performance are warranted to help enhance the yield and sensitivity of Xpert testing.