Xpert MTB/RIF Ultra Is Highly Sensitive for the Diagnosis of Tuberculosis Lymphadenitis in a High-HIV Setting

ABSTRACT Tuberculosis lymphadenitis (TBL) is the most common extrapulmonary tuberculosis (EPTB) manifestation. Xpert MTB/RIF Ultra (Ultra) is a World Health Organization-endorsed diagnostic test, but performance data for TBL, including on noninvasive specimens, are limited. Fine-needle aspiration biopsy specimens (FNABs) from outpatients (≥18 years) with presumptive TBL (n = 135) underwent (i) routine Xpert MTB/RIF testing (later with Ultra once programmatically available), (ii) MGIT 960 culture (if Xpert or Ultra negative or rifampicin resistant), and (iii) study Ultra testing. Concentrated paired urine specimens underwent Ultra testing. Primary analyses used a microbiological reference standard (MRS). In a head-to-head comparison (n = 92) of an FNAB study Ultra and Xpert, Ultra had increased sensitivity (91% [95% confidence interval: 79, 98] versus 72% [57, 84]; P = 0.016) and decreased specificity (76% [61, 87] versus 93% [82, 99]; P = 0.020) and diagnosed patients not on treatment. Neither HIV nor alternative reference standards affected sensitivity and specificity. In patients with both routine and study Ultra tests, the latter detected more cases (+20% [0, 42]; P = 0.034), and false-negative study Ultra results were more inhibited than true-positive results. Study Ultra false positives had less mycobacterial DNA than true positives (trace-positive proportions, 59% [13/22] versus 12% [5/51]; P < 0.001). “Trace” exclusion or recategorization removed potential benefits offered over Xpert. Urine Ultra tests had low sensitivity (18% [7, 35]). Ultra testing on FNABs is highly sensitive and detects more TBL than Xpert (Ultra still missed some cases due in part to inhibition). Patients with FNAB Ultra-positive “trace” results, most of whom will be culture negative, may require additional clinical investigation. Urine Ultra testing could reduce the number of patients needing invasive sampling.

One potential test is the Xpert MTB/RIF Ultra (Ultra), which offers improved sensitivity over Xpert for pulmonary TB, partly enabled by, in addition to rpoB, amplification of multicopy insertion elements (IS6110, IS1081) (10). Data on Ultra for TBL are emerging: one retrospective evaluation tested 10 Xpert-negative, culture-positive FNABs and found half to be Ultra positive (11); another retrospective evaluation (n = 25) reported sensitivity and specificity of 94% (95% confidence interval: 71, 77)) and 100% (63, 72), respectively (12); and a prospective evaluation (n = 73) reported a sensitivity and specificity of 78% (40, 97) and 78% (66, 87), respectively (13). No studies included head-tohead Xpert and Ultra data. Additionally, since Ultra's advent, algorithms for TBL diagnosis remain essentially unchanged from the Xpert era-culture is still recommended in Ultra-negative patients. Whether this is needed or, conversely, if culture is needed to confirm positive Ultra results due to specificity concerns associated with the new trace semiquantitation category (10,14) requires investigation.
Lastly, FNABs are rarely collected in primary care; patients are referred to district or tertiary facilities, resulting in care cascade gaps (15). If an Ultra has high sensitivity and specificity on an easily accessible fluid like urine, the need for invasive sampling could be mitigated, potentially drastically reducing provider and patient economic and time costs, including those associated with referral. To our knowledge, urine Ultra testing for TBL specifically is unevaluated.
We evaluated the head-to-head diagnostic accuracy of Xpert and Ultra on FNABs and Ultra on urine specimens in patients with presumptive TBL in a tertiary hospital in a high-HIV setting in South Africa. We hypothesized that Ultra would show improved sensitivity compared to Xpert.

MATERIALS AND METHODS
Ethics statement. The study was approved by the Stellenbosch University Human Research Ethics Committee and Tygerberg General Hospital (TGH) (both N16/04/050).
Patient recruitment. One hundred thirty-five outpatients ($18 years) with presumptive TBL (swollen lymph nodes) undergoing routine referral and investigation at a tertiary referral clinic at TGH in Cape Town, South Africa, were consecutively recruited from 25 January 2017 to 12 March 2019 and provided FNABs and urine. Patients who received TB treatment #60 days prior were excluded.
Fine-needle aspirate collection. FNABs were collected by multiple needle passes using a 23-gauge needle and a 10-ml syringe. While the needle was inserted, negative suction with a cutting motion was applied for aspiration. The first two passes were used for routine cytology. From each pass, two slides were prepared: the first was air dried for Rapidiff staining, and the second was spray fixed for Papanicolaou staining (;25 ml total volume used per pass) (Fig. 1). The remaining syringe contents were flushed into 1.5 ml TB transport medium (16). The third pass (5 to 50 ml) was collected into 700 ml 5% saline (Ysterplaat Medical Supplies, Cape Town, South Africa).
Xpert, Ultra, and culture. (i) Routine testing. Xpert (version 1; Cepheid, USA) was performed programmatically from 25 January 2017 to 9 April 2018 by the government programmatic laboratory (National Health Laboratory Service [NHLS]), who performed Ultra (version 1) thereafter (17). Sample reagent (2 ml; Cepheid, USA) was added to 500 ml of aspirate-containing 1.5 ml TB transport medium (4:1 ratio), and 2 ml of the mixture was used for Xpert or Ultra (18,19). Per the algorithm, if a specimen was Xpert or Ultra positive and rifampicin susceptible, culture was not done. If a specimen was Xpert or Ultra negative or was Xpert or Ultra positive and rifampicin resistant, 500 ml aspirate-containing TB transport medium was inoculated into a MGIT 960 liquid culture without NALC-NaOH decontamination (Fig. 1). If a nonactionable (not positive or negative) (14) Xpert or Ultra result occurred, the remaining 500 ml TB transport medium was used to repeat the test.
(ii) Study testing. The third pass in 700 ml saline was tested with Ultra (cartridge version 3; study Ultra) using a 2:1 sample-to-reagent ratio (19). Study Ultra was done irrespective of whether routine Xpert or Ultra was done.
(iii) MTBC typing and drug susceptibility testing. The MTBDRplus assay was done on culture-positive isolates for species identification and drug susceptibility testing.
Urine Ultra testing. Five to 20 ml urine stored at 280°C was centrifuged (1,811 Â g, 10 min, room temperature), and the supernatant was removed until 700 ml remained, which was tested with Ultra (2:1 sample-to-reagent volume ratio) (19).
Patient treatment and follow-up. Treatment decisions were programmatic without study involvement (no study results reported for patient management). Attempts were made to telephonically follow up patients at least 12 weeks after recruitment, at which point TB treatment initiation status was recorded and, if treatment started, treatment response was queried. Patients were lost to follow-up if at least two calls were unsuccessful and messages were unreturned for each time point.
Definitions. (i) Patient groups. Patients were designated definite, probable, or non-TB by use of different reference standards. For the microbiological reference standard (MRS), definite TB patients were culture positive and/or cytology positive on FNABs, and non-TB patients were culture and cytology negative on FNABs. Unclassifiable patients had no positive MRS test, culture was contaminated or not done, and cytology was not done. Table S1 in the supplemental material shows the extended microbiological standard (eMRS), composite reference standard (CRS), and probable TB definitions.
(ii) Other definitions. Xpert or Ultra actionable results for TB were MTBC detected and rifampicin susceptible, rifampicin resistant or rifampicin indeterminate, or MTBC not detected (14). For culture, actionable results were positive or negative for MTBC. For cytology, the presence or absence of granulomatous inflammation was recorded.
Statistical analysis. We included patients in head-to-head analyses if they had actionable routine index test (Xpert or Ultra), study Ultra, and culture results (or, if culture was nonactionable, a cytology result was available). Proportion tests (20) were done using STATA version 16.0 (StataCorp, College Station TX, USA) and GraphPad Prism version 8.0.1 (GraphPad Software, San Diego, USA). Venn diagrams were made with InteractiVenn (21). Differences in diagnostic accuracy metrics were calculated using proportion tests or McNemar's test as appropriate. STARD guidelines were followed (22). We excluded the probable TB patients from the primary analysis due to few patients meeting this definition.
Diagnostic accuracy and yield of study Ultra and routine Xpert on FNABs. (i) Overall. When Ultra was compared head-to-head to Xpert using the MRS (n = 92) (   resistance results occurred; however, numbers were small, precluding precise accuracy estimates (see supplementary results in the supplemental material).
(iv) Ultra PCR inhibition. An analysis of sample processing control (SPC) threshold cycle (C T ) values (see Fig. S1 in the supplemental material) (higher values indicate more inhibition) showed more inhibition in study Ultra positives than negatives (25. (v) Relationship with bacterial load. Neither study Ultra nor routine Xpert C T values correlated with bacillary load measured using culture time to positivity (Fig. S2) in FNABs.
Comparison of study Ultra true positives and false positives.  Table 3, and the false positives per patient information in shown in Table S3.
Urine Ultra yield, sensitivity, and specificity and nonactionable results. Urine Ultra testing had low sensitivity (18% [7,35]) and high specificity (98% [88, 100]) (headto-head comparisons with FNAB study Ultra are shown in Table S4). Of concentrated urine specimens tested (n = 84), 8% (7/84) were nonactionable and 100% (7/7) of these resolved to actionable when unconcentrated urine specimens were tested (one unconcentrated urine specimen was now Ultra positive). None of the 18 HIV-negative patients had any positive urine Ultra results. Twelve percent (7/57) of the HIV-positive patients were urine Ultra positive (six of seven detected by both positive MRS and study Ultra FNAB result) (Fig. 3C). In other words, when urine Ultra was attempted among HIV positives, 11% (7/64, 3 of which were trace) were positive, meaning that universal concentrated urine Ultra testing in HIV positives with presumptive TBL could reduce the number of FNABs required for TB diagnosis, as few are nonactionable.
Patient treatment status at follow-up. Ninety-six percent (130/135) of patients were followed up (median, 37 weeks [IQR, 16 to 65 weeks] since recruitment), and 52% (68/130) initiated treatment. Of the 48% (62/130) not initiated on treatment, 11% (7/ 62) were definite TB and 89% (55/62) were non-TB, of which 57% (4/7) and 29% (16/55) were study Ultra positive, respectively. In these definite-TB patients not on treatment but detected by study Ultra, 50% (2/4) were detected by routine Ultra or Xpert, meaning that study Ultra was the only rapid PCR test available in half of these cases (these two definite-TB patients were pretreatment lost to follow-up). Regarding the clinical status of patients initiated on treatment, 94% (64/68) reported treatment completion, and of these, 94% (60/64) reported being clinically well. Three percent of the patients followed up (4/130) died (one of these four was study Ultra positive [routine Ultra negative, Xpert negative]), and100% (4/4) of the decedents were non-TB and not placed on treatment.

DISCUSSION
Our key findings are the following. (i) Study Ultra on FNABs had, in comparison to Xpert, improved sensitivity and decreased specificity and outperformed routine Ultra (tests unaffected by HIV, alternative reference standards, and probable TBs). (ii) Approximately 3 in 10 study Ultra positives had not been placed on treatment, indicating opportunities to improve TBL treatment with Ultra. (iii) Exclusion of study Ultra trace results improved specificity (more so than reclassifying to negative) without large sensitivity costs relative to treating Ultra trace results as positive. (iv) Urine Ultra testing had low sensitivity but could reduce the proportion of presumptive TBL patients who require an FNAB in our setting. (v) Ultra false-negative results are associated with PCR inhibition. These data show high sensitivity of Ultra on FNABs for TBL with the inclusion of tracepositive results (without which sensitivity benefits over Xpert are not seen).
Ultra on FNABs had higher sensitivity than Xpert, suggesting that Ultra is a rapid initial test for TBL. Ultra still did not detect, however, approximately 1 in 10 TBL cases, indicating a sustained need for more sensitive tests (especially those that use noninvasive specimens) and a continued role for reflex tests for downstream testing of Ultranegative FNABs. Importantly, as was done previously for Xpert (23), we showed that one likely cause of Ultra false negativity is increased PCR inhibition (which could be caused by mucopurulent or viscous samples, as seen in sputum [23]), suggesting that optimized specimen processing workflows to better remove interfering agents are still needed to boost sensitivity. Notably, Ultra had suboptimal specificity (2 in 10 MRS-negative people were study Ultra positive). One reason may be that culture and cytology have limitations as reference standards for EPTB (8). Notably, this finding mirrors prior work on TBL that used tissue in addition to fluid biopsy specimens for Ultra, where a specificity of 78% versus culture was observed (13). However, when compared to an eMRS that included microbiological tests such as FNAB culture as well as culture and Ultra on non-site-of-disease fluids, FNAB Ultra specificity was 100% in that study. In contrast, we applied microbiological tests only to FNABs and did not exhaustively sample anatomical sites (24), which might underestimate specificity.
Ultra false positivity was more frequent in patients with less mycobacterial DNA, and in contrast to pulmonary TB, FNAB Ultra false positivity was not associated with prior TB (14). The true nature of these Ultra "false positives" in EPTB requires clarification and is an important topic for future research (in our setting, most "false-positive" patients with presumptive pulmonary TB remain well without treatment) (25,26). Such "false-positive" results could be caused by M. tuberculosis in FNABs that is not culturable using conventional methods like MGIT 960. For example, in animal models, M. tuberculosis DNA in lymph nodes is detectable during reactivation of TB, despite no pathological evidence of disease and no culturability. M. tuberculosis is hypothesized to then disseminate throughout the body from the lymph nodes (27). Moreover, we observed no correlation in bacterial load measured between Ultra and culture, further supporting the presence of M. tuberculosis DNA in the absence of culturability.
Critically, if Ultra trace results were excluded or reclassified to elevate specificity, Ultra would lose sensitivity benefits in comparison to Xpert; however, this sensitivity loss was less for the former than the latter strategy, suggesting that exclusion is the preferred strategy for handling trace results.
When routine and study Ultra concordances were analyzed, study Ultras had a higher yield. This may be due to specimen processing (e.g., more sample reagent is used for routine Ultras than for study Ultras) or cartridge version differences (which includes extending product stability and improving product manufacturability without affecting assay performance) but is overall indicative of an area to improve the diagnosis of TBL within the program. Few studies examined Ultra on urine specimens (28)(29)(30) and none in patients investigated for TBL. Urine Ultra may obviate invasive sampling (and hence referral to a specialized facility and associated costs and delays). Despite concentration (31), low yield and sensitivity were observed for urine Ultra, suggesting it could marginally reduce FNAB collection (approximately 1/10). Such a strategy is undermined by elevated nonactionable result rates and cost effectiveness, including the number needed to test, and would require prospective investigation and modeling; however, we expect the utility of such an approach to be further enhanced with better urine tests (we did not have access to Fujifilm SILVAMP) (32).
These results have strengths and limitations. Our study was pragmatic, routine culture was not always done, and although our MRS included cytology, multiple cultures (including on specimens from other anatomical sites) may improve specificity estimates. Furthermore, multiple FNAB passes were done to obtain adequate volumes that could have introduced sampling variation; however, FNABs were collected using a standardized protocol by a single health care worker. Additionally, a third of the FNABs collected for routine testing were used for Ultra and had more sample reagent (SR) buffer added for testing (a 4:1 ratio of SR buffer to sample, the programmatic standard of care) than that used for study Ultras, in which the full sample collected was used and a 2:1 ratio of SR buffer was added. A different cartridge version (version 1) was used in routine testing, whereas study Ultra used a later version (version 3); however, in an internal head-to-head version evaluation done by the programmatic laboratory (as part of separate study), no sensitivity differences between different versions were observed (most changes between versions were done to improve stability and optimize manufacturing), meaning that it is unlikely that these version differences in routine versus study Ultra accounted for any meaningful performance differences. Thus, the differences observed primarily appear to be due to sampling and specimen processing differences, which improved sensitivity and yield; however, our study was not designed to quantify the contribution of each component that differed between routine and study Ultra testing.
In conclusion, in a routine clinical setting in patients with presumptive TBL, Ultra detects more TBL than Xpert and would result in more people being placed on treatment. This is driven by the added benefit of trace results. Furthermore, programmatic Ultra testing can be optimized on the diagnostic laboratory front, as study Ultra had better performance than that done routinely. Urine Ultra testing could reduce invasive sampling and associated delays, but there remains a need for better urine-based tests for TBL. We recommend that a positive FNAB Ultra result be used to initiate treatment; however, patients with a negative Ultra result still require confirmatory testing, and many patients with a trace-positive result will be culture negative. Our study supports Ultra's use for TBL diagnosis.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, PDF file, 0.4 MB.