Performance of Xpert MTB/RIF Assay for Childhood Pulmonary Tuberculosis With Real World Evidence in China

BACKGROUND: Rapid and accurate notification of childhood pulmonary tuberculosis (PTB) is a worldwide challenge. Although the Xpert MTB/RIF assay (Xpert) has been endorsed as the initial test for suspected childhood pulmonary tuberculosis in many countries, limited studies have reported the real-world performance of Xpert for the detection of childhood PTB. OBJECTIVE: To evaluate the real-world performance of Xpert for the detection of childhood pulmonary TB in China. METHODS: We consecutively extracted the data of all patients ≤14 years with pulmonary disease through the electronic medical record (EMR) systems of Shanghai Public Health Clinical Center from January 2014 to December 2017. The clinical profile, the decision-making tests including AFB smear, solid/liquid culture, pathological examination and Xpert result were matched and assessed. The real diagnostic accuracy and the all-factors case notification rate for childhood PTB with the implementation of Xpert were evaluated. RESULTS: 519 cases ≤14 years with pulmonary disease were extracted from the data base. Of these, 145 had matched results, there were 374 non-matched cases including 346 with incomplete or unavailable data and 28 with NTM, BCG or an unidentified strain. For matched data, the overall sensitivity and specificity of the Xpert assay were 66.7% (32/48, 95%CI 0.52-0.80) and 87.6% (85/97, 95%CI 0.87-0.98) respectively against the gold standard; 34.6% (44/127, 95%CI 26.6-43.7) and 100% against the composite clinical reference standard (CCRS). The all-factors case notification rate by Xpert was 29%. CONCLUSIONS: Xpert/MTB RIF assay has acceptable sensitivity and excellent

specificity for rapid diagnosis of children with pulmonary TB as well as for the detection of RIF resistance in China. However, implementation of Xpert for the initial diagnosis of childhood PTB is inadequate to meet the urgent requirement for rapid and accurate detection of childhood PTB. tests and the responses to anti-TB treatment (ATT). About 10 to 15% of childhood TB is smear-positive because samples are paucibacillary [2]. The sputum culture is only able to detect approximately 30-40% of the cases with probable tuberculosis in children [3] and the tuberculin skin test (TST) and interferon-gamma release assay (IGRA) cannot differentiate latent TB from active TB [4][5]. In recent years, molecular tests for TB were introduced into clinical practice and a rapid test system, Xpert® MTB/RIF assay (Cepheid, USA) was endorsed by WHO for the detection of tuberculosis [6]. Xpert assays were reported to have better accuracy than sputum smear microscopy. For adults, this assay was at least as sensitive as culture methods [7]. For children, especially for those who experience difficulty in expectoration, the data for the accuracy of Xpert assay has been limited. Since 2018, Shanghai CDC has used the Xpert test as an initial method for early detection of PTB. We designed this retrospective study based on the electronic medical record (EMR) systems in order to evaluate the real diagnostic performance for childhood PTB among hospitalized patients in the context of an environment with a substantial TB burden.

METHODS
Non-interventional cohort data traced from EMR system We created a database from the EMR systems of Shanghai Public Health Clinical Center which is the exclusive tertiary hospital administered for the hospitalization of childhood tuberculosis in Shanghai, China. The EMR systems used in this study were the HIS (hospital information system, Kingstar Winning Software.co., Ltd, China) and the LIS (laboratory information system, Kingstar Winning Software.co., Ltd, China) systems which are widely used as the main systems in Shanghai. Case data were collected consecutively from January 2014 to December 2017. We searched all the cases with the following two criteria: (1) ≤14 years of age; (2) with pulmonary disease. The case information was analyzed for data integrity in the matching records from the EMR systems including the quality of samples, the results from Xpert MTB/RIF system, the sequencing for strain identification, the clinical profile and other laboratory examinations.
The real prevalence of PTB in the hospitalized children ≤14 years with pulmonary disease was estimated with an extended scale, based on comprehensive consideration of the EMR regarding clinical diagnosis, the follow-up records, the laboratory test results, the composite clinical reference standard (CCRS) and the experienced physicians' suggestions. With this estimation, we calculated the real notification rate of Xpert for suspected childhood PTB.

Case definition
The gold standard (microbiological reference standard) of PTB in this study was a composite of positive results for M. tuberculosis by culture method on a respiratory tract specimen (RTS) including induced or induced sputum (IS), nasopharyngeal aspiration (NA) or gastric aspiration (GA). In consideration of the low yield of MTB culture from childhood RTS, we also assessed the diagnostic performance against CCRS. In China CCRS involves the following 8 items [8]: (1) fever or cough for more than two weeks, weight loss or failure to gain weight in the previous 3 months, ( (III) non-TB (no TB evidence). According to the practice guideline, the sequencing method (Xpert was not included) and histopathological examination (HE) applied to lung biopsy specimens were used as definitive diagnostic methods. In the follow-up period, experts divided patients who had no definitely evidence into probable TB or non-TB by considering if they had effective response to anti-tuberculosis therapy (clinical features suggestive of tuberculosis disease that were present at baseline have improved, and there is no new clinical feature suggestive of tuberculosis) or no (clinical features suggestive of tuberculosis disease that were present at baseline have not improved).
Estimated prevalence at the site We estimated the real prevalence with four scales: EMR recorded diagnosis, the follow-up record, the microbiological test, the experts' experience. The first index "EMR recorded diagnosis" was established on clinical judgment and suspicions, which often over-estimated the real prevalence, representing the maximum estimation. The second index "follow-up record" was created by a sampling method and was potentially more accurate. The third index "microbiological test" represented the minimum estimation of prevalence. The fourth index "experts' experience" was the most likely correct, being consistent with all the clinical criteria. Although this estimation was not very exact, it was considered important to calculate the real prevalence of the site. Some accuracy indices, such as positive and negative predictive values can only be comprehended in association with the prevalence. The case notification rate was also calculated from these estimates.

Sample processing and records
Direct smear microscopy and MTB culture of each RTS were performed following the WHO guideline [10]. Cultures positive for growth of acid-fast bacilli underwent confirmation of M tuberculosis complex by MPT64/MPB64 antigen detection [11].
Phenotypic drug-susceptibility testing was done on positive cultures with the BACTEC MGIT 960 SIRE kit (BD, Franklin Lakes, NJ, USA) in accordance with the manufacturer's instructions [12]. The Xpert MTB/RIF test was performed on the basis of the manufacturer's protocol, Results were reported as the following four results: "negative", "TB detected, rifampicin resistance detected", "TB detected, no rifampicin resistance detected" and "invalid result". The semi quantitative scale for Xpert results was divided into four grades which were measured by cycle-threshold (Ct): very low (28<Ct ≤38), low (22<Ct ≤28), medium (16<Ct ≤22), or high (Ct≤16) [13].

Statistical analysis
The general sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated to evaluate the diagnostic accuracy of Xpert

Detection of RIF resistance
A total of 11 cases had recorded drug-resistance via phenotype drug sensitivity test (DST) and had matched Xpert MTB/RIF test results. 5 were resistant to HRES (isoniazid, rifampin, ethambutol and streptomycin), 3 were resistant to HR, 1 was resistant to HE, and 2 were mono-resistant to H. The 8 cases with phenotypic rifampin-resistance were also detected by Xpert assay. The consistency of culturebased DST and Xpert assay for rifampin resistance was 100%.

DISCUSSION
In this study, the sensitivity of the Xpert test varied with the AFB smear status in culture-proven cases. Xpert showed an overall sensitivity of 66.7%, which was similar to AFB smear (30/48, 62.5%). The sensitivity was higher in cases where both culture and AFB smear were positive (76.7%) and lower in cases that were culturepositive but AFB smear negative (50%). This finding was consistent with other reports [14][15] and easy to understand: the trend for an Xpert-positive yield is associated with the bacterial load in specimens. In this study, we estimated the real prevalence of childhood TB to be 67%, but the real case notification rate was much lower. One reason of this situation was that the sensitivity of neither Xpert, culture nor smear was good enough for the notification of childhood PTB cases (34.6%, 37.8% and 23.6% against CCRS, respectively); another reason was that about 1/4 of the cases did not have valid specimen test results. However, the causes could not be fully assessed because this was a retrospective study. These results showed that the performance of Xpert remains suboptimum in a real-world setting.
For establishing a definitive diagnosis of childhood PTB, multiple or repeated decision-making tests (such as MTB culture, nucleic acid test or sequencing) on RTS are often required. The added diagnostic yield from combined tests seems much higher than repeated test [15][16], although this was not extensively assessed in our study. A study from South Africa suggested that the incremental increase in sensitivity from testing a second specimen was 27.8% for MTB/RIF in smearnegative cases [17]. But another study indicated that the collection of one sputum specimen should be adequate in establishing a diagnosis of TB; the sensitivity of MTB/RIF for MTB detection in suspected TB cases from whom only one specimen was analyzed did not differ significantly from the sensitivity achieved by testing two or three specimens [18]. In this study, we included data from only one specimen from each child, further studies would be needed to confirm that two specimens give a higher detection rate than one sample. Some studies indicated that sputum induction can be effectively performed and is well tolerated and safe even in infants and this induction was better than GA for the isolation of M. tuberculosis in both HIV-infected and uninfected infants and children [17,19]. In this study, children under five years old underwent GA, over five years old had IS or NA, Xpert performed equally well with the different respiratory specimens (IS, NA and GA). The sensitivity of Xpert varied significantly in different age groups, in 1-5 years group the sensitivity of Xpert was 36.4%, which was similar to the sensitivity of child-wide population against culture, it indicated GA was a desirable method to instead sputum in infant and young children. 5-10 years group had the lowest sensitivity (15.4%), we analyzed it was due to children among this group couldn't exhaust sputum well. 10-14 years group had the highest sensitivity (61.5%), which was similar to adult [7], it indicated children of this age had the power to cough and could acquire high-quality specimens.
In a real-world setting, NTM and BCG vaccine infection cannot be pre-excluded before laboratory test. We noticed that if NTM and BCG vaccine infection were included in the analysis, the specificity of Xpert against CCRS would decrease from 100% to 80.8%. The added false positive was from BCG vaccine infection. BCG was derived from Mycobacterium bovis and the sensitivity to anti-TB drugs is different.
The BCG strain has low to medium level resistance to isoniazid and is intrinsically resistant to pyrazinamide which is much different from the routine MTB strains [20]. The Xpert MTB/RIF assay detects rifampicin-resistance mutations in the rpoB gene, based on the finding that 95% of all rifampicin-resistant M tuberculosis strains contain mutations localized within the 81bp core region of the bacterial RNA polymerase β subunit (rpoB) gene, which encodes the active site of the enzyme [21]. Rifampicin resistance is frequently associated with concomitant isoniazid (INH) resistance [22], so rifampicin resistance is strongly indicative of MDR tuberculosis and Xpert was used as the initial screen for MDR tuberculosis in this study. There were 11 cases of drug-resistant tuberculosis, 8 of whom were of MDR-TB, 3 were resistance to INH but sensitivity to RFP. The MTB/RIF assay was positive to all MDR-TB and negative to all rifampin-sensitive, isoniazid-resistant tuberculosis patients. It indicated Xpert can find out MDR-TB with high sensitivity and specificity.
Our study has some limitations. Firstly, it was a retrospective study and a singlecenter study an endpoint that we were unable to analyze due to our retrospective design. Secondly, Incorporation bias can arise when the unknown strain identification results influences the diagnosis of active tuberculosis.

CONCLUSION
Compared with the poor sensitivity of AFB smear and the long turnover time of culture, Xpert is a point-of-care diagnostic tool for rapid detection of childhood PTB with high sensitive and excellent accuracy for RIF resistance. However, using Xpert or culture alone can lead to the missed diagnosis of childhood tuberculosis. The

Ethics approval and consent to participate:
This study is a retrospective study and the informed consents are not possible to acquire from participants' parental guardians. The tests performed in this study were routine clinical practice according to domestic practice guidelines and benefited the participants. The study procedure was approved by the Ethics hospital staff for their invaluable support and contribution during patient enrollment and data collection. Figure 1 Flowchart of the participants. Legends: a) Of these cases, 45 were confirmed with TB before

Supplementary Files
This is a list of supplementary files associated with the primary manuscript. Click to download.