The transition to Xpert MTB/RIF ultra: diagnostic accuracy for pulmonary tuberculosis in Kampala, Uganda

From April 2018 to April 2019, we conducted a cross sectional study of patients with presumed pulmonary TB at Kiruddu Hospital pulmonary ward and outpatient chest clinic, Mulago Hospital TB ward, and Kisenyi Health Center IV outpatient clinic in Kampala, Uganda. As previously described [7], we included adults (> 18 years) who were referred by clinicians for sputum-based TB testing (i.e., Xpert). We excluded participants who received anti-TB treatment or antibiotics with anti-TB activity such as fluoroquinolones in the prior 12 months, or who refused or were unable to provide informed consent. To select participants, we reviewed the facility Xpert testing log daily. On each day, we enrolled all participants with Xpert-positive results and a random 1:1 sample of Xpert-negative patients identified from the log. The study and full protocol (available on request) was reviewed and approved by the Makerere University School of Medicine Research and Ethics Committee, the Uganda National Council for Science and Technology and the University of California San Francisco Committee on Human Research. This study was performed according to the Standards for Reporting of Diagnostic Accuracy Studies (STARD) guidelines [8].

After consent, all eligible participants completed a survey on demographics and medical history. Each participant provided up to two sputum samples: one for Ultra testing (all participants) and one for solid (Lowenstein-Jensen [LJ]) and liquid (BACTEC MGIT 960, Becton Dickinson, USA) mycobacterial cultures (Xpert-negative participants only). Positive cultures were confirmed using Ziehl-Neelsen (ZN) staining and M. tuberculosis (MTB) complex speciation testing (MPT64TB Ag kit, SD Bioline, South Korea). Trained laboratory technologists conducted all TB testing following standard protocols for Xpert, Ultra [2], and mycobacterial culture [9]. Laboratory staff were blinded to Xpert results at the time of Ultra or culture testing. Participants with negative or unknown HIV status underwent HIV testing per national guidelines, and CD4 cell count was measured for HIV-positive participants. To assess any additional microbiological evidence for false-positive results, Determine TB LAM (Alere, Waltham, USA) was performed on urine samples following standard protocols [10].

Descriptive statistics were used to summarize the demographic and clinical characteristics of study participants. The sensitivity and specificity of sputum Xpert Ultra was calculated with exact binomial 95% confidence interval (CI) using our reference standard definition of TB. We compared sensitivity and specificity by HIV status, CD4 cell count category, prior history of TB, and inpatient status using Chi-squared or Fisher’s Exact test as appropriate. We conducted a sensitivity analysis on Ultra trace results, comparing diagnostic accuracy if trace results were considered positive for all participants or no participants using McNemar’s paired test of proportions. We used STATA 15 (Stata Corp, College Station, TX, USA) to perform all analyses.

During the 12-month study period, we identified 475 patients with positive Xpert results in the facility testing logs. Of the 475 patients, 311 (65%) met eligibility criteria and were enrolled (Fig. 1). We randomly selected and enrolled 387 Xpert-negative participants, of whom 25 were subsequently found to be culture-positive for MTB. Thus, 698 participants (336 with and 362 without TB) were included in the final analysis. The median age was 32 years (IQR 25–40), 436 (62%) were male, over a third (42%) had BMI < 18.5, and 211 (30%) were HIV-positive, of whom 46 (22%) had CD4 cell count < 100 cells/μL (Table 1). Most participants were identified in the outpatient setting (n=620, 89%), and the median Karnofsky score was 80 (IQR 70–90).

Sputum Ultra had a sensitivity of 90.5% (95% CI 86.8 to 93.4) and specificity of 98.1% (95% CI 96.1–99.2) (Table 2). Xpert and Ultra were concordant in 670/698 (96%) participants. Of the 28 discordant results, 14 (50%) were Ultra-positive and Xpert-negative (7 culture-positive), while 14 were Ultra-negative but Xpert-positive. Consequently, the sensitivity of Ultra was not significantly different from that of Xpert (90.5% vs. 92.6%, difference − 2.1, 95% CI − 5 to + 0.1, p=0.1). However, the seven false-positive Ultra results led to a small reduction in specificity of 1.9% (95% CI 0.2 to 3.6, p=0.01) compared to Xpert. Of note, 18 participants with positive cultures (5% of 336 TB cases) had negative Xpert and Ultra results.

Among patients with positive Xpert and Ultra results, rifampicin resistance results were concordant in 285/297 (96%) patients. Three patients had rifampicin resistance detected by Ultra but not Xpert, and one patient had rifampicin resistance detected by Xpert but not Ultra. Rifampicin resistance results were indeterminate in two patients by Ultra (all due to trace results, Xpert RIF negative), and in one patient by Xpert (Ultra RIF negative). Two of the discordant results were missing on Ultra and were not resistant on Xpert, while the three results missing on Xpert were not resistant on Ultra.

As shown in Table 2, Ultra sensitivity and specificity did not differ by HIV status (sensitivity difference 0.96, 95% CI − 6.1 to 8.0, p=0.79; specificity difference − 2.3, 95% CI − 6.0 to 1.33, p=0.1). Ultra sensitivity was also similar among HIV-positive patients with CD4 cell count < 100 cells/μL vs. higher CD4 cell counts (difference 8.6, 95% CI − 2.0 to 19.3, p=0.2). There was also no difference among patients with and without a prior history of TB (sensitivity 85.7% vs. 90.9%, difference − 5.2, 95% CI − 18.5 to 8.2, p=0.37; specificity 95.0% vs. 98.7%, difference − 3.7, 95% CI − 9.3 to 2.0, p=0.06) or among inpatients vs. outpatients (sensitivity 87.2% vs. 90.9%, difference 3.7, 95% CI − 7.3 to 14.7, p=0.46; specificity 100% vs 97.8%, difference 2.2, 95% CI 0.6 to 3.7, p=0.35).

Of the seven patients with false-positive Ultra results, four were HIV-positive (including one with CD4 cell count < 100 cells/μL), of which one had a prior history of TB and all had trace results. Of the remaining three HIV negative participants, two had a prior history of TB.

There were 21 trace Ultra results. Of the 11 HIV-positive patients with trace results (Table 3), eight had microbiological evidence of TB (two Xpert-positive with a very low semi-quantitative result, five culture-positive and one urine Determine TB LAM-positive). Of the 10 HIV-negative patients with trace results, 8 had microbiological evidence of TB (four Xpert positive with low [N=1] or very low [N=3] semi-quantitative results, three culture-positive, and one urine Determine TB LAM-positive). Thus, 16 of 21 (76%) patients with trace results had microbiologic evidence of TB.

We examined how classification of trace results impacted Ultra accuracy (Table 4). When trace results were defined as positive for all patients, Ultra sensitivity and specificity were 92.6% (95% CI 89.2 to 95.1) and 97.2% (95% CI 95.0 to 98.7), respectively, representing a 2.1% increase in sensitivity (95% CI 0.3 to 3.9, p=0.01) and a non-significant 0.8% reduction in specificity (95% CI − 0.3 to 2.0, p=0.08) compared to when trace results were considered to be positive only in HIV-positive patients as per WHO guidelines. When trace results were defined to be negative for all patients, Ultra sensitivity and specificity were 88.4% (95% CI 84.5 to 91.6) and 99.2% (95%CI 97.6 to 99.8), respectively, representing a 2.1% reduction in sensitivity (95% CI − 3.9 to − 0.3, p=0.01) and non-significant 1.1% increase in specificity (95% CI − 0.2 to 2.5, p=0.05) compared to the current guidelines.