Why is improving the diagnosis of tuberculosis drug resistance important?
Tuberculosis tests, like Xpert MTB/RIF, Xpert MTB/RIF Ultra, and Truenat, only diagnose rifampicin resistance, but do not provide information about resistance to other drugs used to treat tuberculosis. This information is needed to allow for effective treatment to be started quickly.
Not recognizing tuberculosis drug resistance when present (false negative, FN) may result in severe illness and death. An incorrect diagnosis of tuberculosis drug resistance (false positive, FP) may result in stigma and prolonged and unnecessary treatment with less effective drugs that have more side effects.
What is the aim of this review?
How accurate is Xpert MTB/XDR for detecting pulmonary tuberculosis and resistance to tuberculosis drugs (i.e. isoniazid, fluoroquinolones, ethionamide, and amikacin) in adults?
What was studied in the review?
Xpert MTB/XDR is a rapid test for detecting tuberculosis and drug resistance in one test, suitable for laboratories that do not require advanced skills and infrastructure. We assessed Xpert MTB/XDR accuracy against three reference standards.
What are the main results of the review?
We identified two multicentre studies reporting on six separate cohorts (groups of study participants), 1228 participants for pulmonary tuberculosis detection and 1141 participants for drug resistance detection.
For pulmonary tuberculosis detection, we included two studies (one reporting on two separate cohorts). We did not determine an overall summary of Xpert MTB/XDR accuracy.
If Xpert MTB/XDR were to be used in 1000 people with suspected tuberculosis of whom 100 have tuberculosis:
- an estimated 98 to 99 people would have an Xpert MTB/XDR result indicating tuberculosis: of these 1 to 2 (1%) would not have tuberculosis (FP); and 203 to 900 people would have a result indicating the absence of tuberculosis: of these 0 to 697 (0% to 77%) would have tuberculosis (FN).
Drug resistance detection
Of 1000 people detected as tuberculosis positive by Xpert MTB/XDR:
- where 50 have isoniazid resistance, an estimated 61 would have an Xpert MTB/XDR result indicating isoniazid resistance: of these, 14/61 (23%) would not have isoniazid resistance (FP); and 939 (of the 1000 people) would have an Xpert MTB/XDR result indicating the absence of isoniazid resistance: of these, 3/939 (0%) would have isoniazid resistance (FN);
- where 50 have isoniazid resistance, 61 (of 1000 people) would have an Xpert MTB/XDR result indicating isoniazid resistance: of these, 14/61 (23%) would not have isoniazid resistance (FP); and 939 (of 1000 people) would have a result indicating the absence of isoniazid resistance: of these, 3/939 (0%) would have isoniazid resistance (FN);
- where 50 have fluoroquinolone resistance, 66 would have an Xpert MTB/XDR result indicating fluoroquinolone resistance: of these, 19/66 (29%) would not have fluoroquinolone resistance (FP); and 934 would have a result indicating the absence of fluoroquinolone resistance: of these, 3/934 (0%) would have fluoroquinolone resistance (FN);
- where 300 have ethionamide resistance, 296 would have an Xpert MTB/XDR result indicating ethionamide resistance: of these, 2/296 (1%) would not have ethionamide resistance (FP); and 704 would have a result indicating the absence of ethionamide resistance: of these, 6/704 (1%) would have ethionamide resistance (FN);
- where 135 have amikacin resistance, 126 would have an Xpert MTB/XDR result indicating amikacin resistance: of these, 10/126 (8%) would not have amikacin resistance (FP); and 874 would have a result indicating the absence of amikacin resistance: of these, 19/874 (2%) would have amikacin resistance (FN).
How reliable are the results of the studies in this review?
For pulmonary tuberculosis detection, we did not consider the results reliable because around 90% of the participants had Xpert-detected pulmonary tuberculosis to begin with due to the way people were chosen to participate in the studies. For drug resistance detection, we were confident in the results, except for results for ethionamide resistance detection, where the reference standards were not ideal.
Who do the results of this review apply to?
People with suspected pulmonary tuberculosis and tuberculosis drug resistance living in countries with a high burden of tuberculosis drug resistance.
How up-to-date is this review?
We searched for studies up to 23 September 2021. Searches were limited to 2015 onwards as Xpert MTB/XDR was launched in July 2020.
Review findings suggest that, in people determined by Xpert MTB/XDR to be tuberculosis-positive, Xpert MTB/XDR provides accurate results for detection of isoniazid and fluoroquinolone resistance and can assist with selection of an optimised treatment regimen. Given that Xpert MTB/XDR targets a limited number of resistance variants in specific genes, the test may perform differently in different settings. Findings in this review should be interpreted with caution. Sensitivity for detection of ethionamide resistance was based only on Xpert MTB/XDR detection of mutations in the inhA promoter region, a known limitation. High risk of bias limits our confidence in Xpert MTB/XDR accuracy for pulmonary tuberculosis.
Xpert MTB/XDR's impact will depend on its ability to detect tuberculosis (required for DST), prevalence of resistance to a given drug, health care infrastructure, and access to other tests.
The World Health Organization (WHO) End TB Strategy stresses universal access to drug susceptibility testing (DST). DST determines whether Mycobacterium tuberculosis bacteria are susceptible or resistant to drugs. Xpert MTB/XDR is a rapid nucleic acid amplification test for detection of tuberculosis and drug resistance in one test suitable for use in peripheral and intermediate level laboratories. In specimens where tuberculosis is detected by Xpert MTB/XDR, Xpert MTB/XDR can also detect resistance to isoniazid, fluoroquinolones, ethionamide, and amikacin.
To assess the diagnostic accuracy of Xpert MTB/XDR for pulmonary tuberculosis in people with presumptive pulmonary tuberculosis (having signs and symptoms suggestive of tuberculosis, including cough, fever, weight loss, night sweats).
To assess the diagnostic accuracy of Xpert MTB/XDR for resistance to isoniazid, fluoroquinolones, ethionamide, and amikacin in people with tuberculosis detected by Xpert MTB/XDR, irrespective of rifampicin resistance (whether or not rifampicin resistance status was known) and with known rifampicin resistance.
We searched multiple databases to 23 September 2021. We limited searches to 2015 onwards as Xpert MTB/XDR was launched in 2020.
Diagnostic accuracy studies using sputum in adults with presumptive or confirmed pulmonary tuberculosis. Reference standards were culture (pulmonary tuberculosis detection); phenotypic DST (pDST), genotypic DST (gDST),composite (pDST and gDST) (drug resistance detection).
Two review authors independently reviewed reports for eligibility and extracted data using a standardized form. For multicentre studies, we anticipated variability in the type and frequency of mutations associated with resistance to a given drug at the different centres and considered each centre as an independent study cohort for quality assessment and analysis. We assessed methodological quality with QUADAS-2, judging risk of bias separately for each target condition and reference standard. For pulmonary tuberculosis detection, owing to heterogeneity in participant characteristics and observed specificity estimates, we reported a range of sensitivity and specificity estimates and did not perform a meta-analysis. For drug resistance detection, we performed meta-analyses by reference standard using bivariate random-effects models. Using GRADE, we assessed certainty of evidence of Xpert MTB/XDR accuracy for detection of resistance to isoniazid and fluoroquinolones in people irrespective of rifampicin resistance and to ethionamide and amikacin in people with known rifampicin resistance, reflecting real-world situations. We used pDST, except for ethionamide resistance where we considered gDST a better reference standard.
We included two multicentre studies from high multidrug-resistant/rifampicin-resistant tuberculosis burden countries, reporting on six independent study cohorts, involving 1228 participants for pulmonary tuberculosis detection and 1141 participants for drug resistance detection. The proportion of participants with rifampicin resistance in the two studies was 47.9% and 80.9%. For tuberculosis detection, we judged high risk of bias for patient selection owing to selective recruitment. For ethionamide resistance detection, we judged high risk of bias for the reference standard, both pDST and gDST, though we considered gDST a better reference standard.
Pulmonary tuberculosis detection
- Xpert MTB/XDR sensitivity range, 98.3% (96.1 to 99.5) to 98.9% (96.2 to 99.9) and specificity range, 22.5% (14.3 to 32.6) to 100.0% (86.3 to 100.0); median prevalence of pulmonary tuberculosis 91.3%, (interquartile range, 89.3% to 91.8%), (2 studies; 1 study reported on 2 cohorts, 1228 participants; very low-certainty evidence, sensitivity and specificity).
Drug resistance detection
People irrespective of rifampicin resistance
- Isoniazid resistance: Xpert MTB/XDR summary sensitivity and specificity (95% confidence interval (CI)) were 94.2% (87.5 to 97.4) and 98.5% (92.6 to 99.7) against pDST, (6 cohorts, 1083 participants, moderate-certainty evidence, sensitivity and specificity).
- Fluoroquinolone resistance: Xpert MTB/XDR summary sensitivity and specificity were 93.2% (88.1 to 96.2) and 98.0% (90.8 to 99.6) against pDST, (6 cohorts, 1021 participants; high-certainty evidence, sensitivity; moderate-certainty evidence, specificity).
People with known rifampicin resistance
- Ethionamide resistance: Xpert MTB/XDR summary sensitivity and specificity were 98.0% (74.2 to 99.9) and 99.7% (83.5 to 100.0) against gDST, (4 cohorts, 434 participants; very low-certainty evidence, sensitivity and specificity).
- Amikacin resistance: Xpert MTB/XDR summary sensitivity and specificity were 86.1% (75.0 to 92.7) and 98.9% (93.0 to 99.8) against pDST, (4 cohorts, 490 participants; low-certainty evidence, sensitivity; high-certainty evidence, specificity).
Of 1000 people with pulmonary tuberculosis, detected as tuberculosis by Xpert MTB/XDR:
- where 50 have isoniazid resistance, 61 would have an Xpert MTB/XDR result indicating isoniazid resistance: of these, 14/61 (23%) would not have isoniazid resistance (FP); 939 (of 1000 people) would have a result indicating the absence of isoniazid resistance: of these, 3/939 (0%) would have isoniazid resistance (FN).
- where 50 have fluoroquinolone resistance, 66 would have an Xpert MTB/XDR result indicating fluoroquinolone resistance: of these, 19/66 (29%) would not have fluoroquinolone resistance (FP); 934 would have a result indicating the absence of fluoroquinolone resistance: of these, 3/934 (0%) would have fluoroquinolone resistance (FN).
- where 300 have ethionamide resistance, 296 would have an Xpert MTB/XDR result indicating ethionamide resistance: of these, 2/296 (1%) would not have ethionamide resistance (FP); 704 would have a result indicating the absence of ethionamide resistance: of these, 6/704 (1%) would have ethionamide resistance (FN).
- where 135 have amikacin resistance, 126 would have an Xpert MTB/XDR result indicating amikacin resistance: of these, 10/126 (8%) would not have amikacin resistance (FP); 874 would have a result indicating the absence of amikacin resistance: of these, 19/874 (2%) would have amikacin resistance (FN).