Chest X-ray interpretation does not complement Xpert MTB/RIF in diagnosis of smear-negative pulmonary tuberculosis among TB-HIV co-infected adults in a resource-limited setting

Tuberculosis (TB) remains a significant public health problem globally and is the main cause of death among persons living with HIV [1]. Key global priorities for TB care and control emphasize improving earlier case detection and linkage to care. Globally, in 2018, a total of 10 million cases of TB were estimated, of which only 6.4 million (64%) were diagnosed and notified to national programs, the remaining 36% of the estimated TB cases were left unreported [1]. These unreported and hence untreated cases are a reservoir of infection with on-going transmission and contribute significantly to the persistently high TB prevalence and mortality particularly among HIV co-infected individuals [1‐3]. A recent study showed that in resource-limited settings (RLS), TB accounted for approximately 40% of facility-based HIV/AIDS-related adult deaths, and that in almost half of these, TB was undiagnosed at the time of death [4].

In 2018, Uganda had an estimated TB incidence of 86,000 cases with 24% deaths. In the same year, 34,000 of the new cases were HIV and TB co-infected with an estimated 11,000 HIV-infected individuals having died from TB, making Uganda one of the most affected countries worldwide for TB/HIV co-infection. Of the estimated new TB cases in Uganda, only 65% were notified to the national TB program in 2018 [1, 5]. Delayed and missed TB diagnosis in HIV-positive individuals is largely responsible for the high mortality reported in many RLS endemic for TB/HIV in sub-Saharan Africa (SSA) [6‐9].

In many RLS, sputum smear microscopy is still the most widely used up-front diagnostic method at most low-level health centers, and yet it only detects about half of TB cases especially in immunocompromised patients [10]. Low accuracy of sputum smear microscopy has resulted into high frequency of smear-negative TB and increased reliance on empiric decision-making to initiate TB treatment in HIV co-infected patients [11, 12].

In order to achieve the World Health Organization (WHO) goal of ending the TB epidemic by 2035 [13], we need to identify missing TB cases, link them and retain them in appropriate care. One of the strategies to improve identification of these missing cases is improved diagnosis including application of new TB diagnostic tools that have been found to have improved performance. The Xpert MTB/RIF and Ultra assays on the GeneXpert platform (Cepheid, Sunnyvale, CA) have shown the potential to provide rapid, sensitive diagnosis and have been rolled out in several countries in SSA [14, 15]. However, implementation of Xpert MTB/RIF is costly for most low-income countries and requires sophisticated infrastructure that may not be readily accessible in most resource-limited settings [16‐18]. Additionally, although the sensitivity of Xpert MTB/RIF assay for pulmonary TB detection was found to be approximately 98% for sputum smear-positive individuals, the sensitivity is much lower at 67–75% for sputum smear-negative HIV-positive individuals [19, 20]. This highlights the need for adjuvant tests and empiric decision-making for smear-negative HIV-positive individuals in particular.

The WHO still considers CXR an important piece in TB diagnosis particularly as an adjuvant test in smear-negative TB diagnostic algorithms [21, 22]. National TB guidelines for most RLS have included use of CXR as part of diagnostic algorithms to complement clinical diagnosis for TB especially for patients with negative sputum smear and/or Xpert MTB/RIF [5, 22]. However, implementation of CXR needs resources to meet operational costs related to electricity, X-ray films, CXR maintenance, and good real-estate facilities that are frequently not available in most RLS. Paucity of personnel that can correctly interpret CXR at most health facilities further challenges CXR implementation [23]. Moreover, the low specificity of CXR is often associated with over-diagnosis of TB, which results in unnecessary TB treatment, wastage of resource and unnecessary toxicity [23‐26]. In this Xpert MTB/RIF era, National TB programs in low-income countries need to make important policy decisions on allocation of resources and positioning of CXR within the TB diagnostic algorithms. There is limited published data available on the role of CXR in the Xpert MTB/RIF era to guide policymakers in RLS.

We evaluated the sensitivity, specificity, predictive values and likelihood ratios of CXR interpretation as stand-alone and when in combination with Xpert MTB/RIF, in the diagnosis of smear-negative TB among HIV co-infected patients presenting with symptoms of TB. The goal was to determine the additive value of CXR to smear or Xpert MTB/RIF test in the diagnosis of smear-negative TB among HIV-positive individuals in a high HIV/TB burden setting.

Chest X-Ray (CXR) remains a useful adjuvant test in smear-negative pulmonary TB to guide further patient management in high TB/HIV prevalence settings. Given the recent widespread roll-out of Xpert MTB/RIF in most low-income countries, we evaluated the role of CXR in the diagnosis of smear-negative TB among HIV-positive presumptive TB patients in the Xpert MTB/RIF era. Overall, we found poor performance of CXR interpretation in the diagnosis of culture- positive smear-negative TB among HIV-positive participants. We found that, although CXR increased sensitivity of Xpert MTB/RIF assay, this was at the expense of specificity. Given its high negative predictive value, a normal CXR could reliably exclude TB in HIV-positives with negative sputum smear, however, the applicability of CXR in settings with Xpert MTB/Rif is limited by the challenges of CXR application faced in most RLS. Our study further found that combining CXR with Xpert MTB/RIF reduced the specificity of Xpert MTB/RIF in the diagnosis of smear-negative TB. This implies that use of CXR as an adjuvant to Xpert MTB/RIF results into inaccurate diagnosis of TB among HIV co-infected presumptive TB patients with negative smear.

Our findings corroborate with earlier reports from high prevalence TB/HIV settings in SSA which also found poor performance, particularly low specificity, of CXR interpretation in the diagnosis of culture-confirmed TB among smear-negative HIV co-infected individuals [24, 26, 29‐31].

Poor performance of CXR in TB diagnosis in HIV co-infected individuals is often attributed to the non-specific CXR patterns and difficulty in interpretation [23, 25]. Results of a chest radiograph often depend on the severity and presentation of TB, which all depend on the stage of the HIV disease, with advanced HIV disease more likely to present with non-specific patterns compared to early-stage HIV disease. Moreover, CXR findings can be normal in up to 25–50% of patients with culture-confirmed TB [23, 25, 26]. Our study population comprised HIV co-infected individuals, majority of whom had advanced HIV disease (median CD4 count 112 cells/mm³) and this could potentially explain the non-specific patterns inaccurately interpreted as TB, and thus the low specificity seen in our study.

CXR interpretation in HIV co-infection is further complicated by the presence of other HIV-related pulmonary diseases that could radiologically mimic TB; commonly, bacterial pneumonia [23, 24]. Several studies from SSA [24, 26, 30] have reported presence of other HIV-related pulmonary conditions that radiologically mimic TB that have led to misinterpretation of chest radiographs as TB. This too could explain the poor CXR performance in this study that was performed among entirely HIV-positive individuals. However, our study did not evaluate other possible non-TB pulmonary conditions that could explain the inaccurate CXR interpretation.

Poor performance of CXR interpretation in this study could also be explained by poor quality of CXR films, which in turn is dependent on the skill of the technician, and the quality of the CXR equipment used. Studies have shown accuracy of CXR interpretation is dependent on the quality of the CXR film [23, 30]. In most peripheral health centres in RLS where CXR services are available, the quality of the radiographs produced from the poor quality CXR equipment is often not good enough to improve interpretation [23]. To improve accuracy of CXR interpretation in our study, CXR interpretation was performed by expert radiologists.

Our study further found that CXR interpretation did not complement Xpert MTB/RIF diagnostic performance in smear-negative TB diagnosis in this low-income country with high TB/HIV prevalence. Instead, we noted decreased accuracy with loss in specificity of Xpert MTB/RIF test when additively used with CXR interpretation i.e. Xpert MTB/RIF specificity dropped from 95.8% when used as stand-alone test to 51.6% when additively used with CXR interpretation. A similar trend was seen when results were stratified by setting (outpatients versus inpatients). Our findings are similar to an earlier report, which also found a reduction in specificity of Xpert MTB/RIF when additively used with CXR interpretation [31]. The reduction in specificity is attributable to the non-specific nature of CXR patterns in HIV disease, which results into inaccurate interpretation [23, 26]. These findings imply that use of CXR interpretation as an additive test to Xpert MTB/RIF test in centers where Xpert MTB/RIF is available as a TB diagnostic increases the likelihood of inaccurate diagnosis of TB, which consequently puts more financial strain to TB programs in RLS and leads to misuse of already scarce resources. However, based on our recent findings from the same patient population, CXR could complement urine TB Lipoarabinomannan (LAM) if Xpert MTB/RIF is not available [28]. The previous findings showed that an algorithm in which CXR is performed following urine TB LAM test significantly improves diagnosis of TB among smear-negative HIV-positive patients with a high negative predictive value, especially when CD4 counts are < 100cells/mm³ [28]. This was not the case when CXR was combined with Xpert MTB/RIF test in the diagnosis of TB in smear-negative HIV-positive patients.

Of note, our study found lower lung field infiltrates (41%) and pleural disease among the most frequent abnormalities reported by the expert radiologist. This is similar to earlier findings from other studies on CXR features among HIV co-infected individuals [23‐26, 30]. These atypical CXR features are not surprising in our study, which included entirely HIV co-infected patients, many of whom had immunologically advanced HIV disease. Advanced HIV disease could also explain the miliary TB disease CXR pattern seen in a few of our study patients. Cavities were rare among our study participants as is expected in HIV-positive patients with advanced disease [23, 25].

This study had strengths. First, our study included both sputum and blood MTB culture as the standard of reference, which improved the strength of our reference standard for TB diagnosis. Inadequacy of gold standard is often a limitation in TB diagnostics evaluations. Second, in order to improve CXR interpretation, we used two expert radiologists to interpret the chest radiographs, and in case of variability, a third radiologist was used as a tiebreaker. Third, our study population and setting well represent a real-world situation that is likely to be found in high prevalence HIV/TB settings in low-income countries. We studied a population of HIV-infected individuals with an increased risk of morbidity and mortality from TB that is targeted in the End TB strategy. Lastly, we used a relatively simple CXR evaluation form in the interpretation of the CXR, which is a more practical approach since hospitals in RLS often lack personnel that can interpret CXR with expertise and are often understaffed. By using a simple and easy-to-use CXR interpretation form, we assigned the accuracy of the CXR in this study more agreeable with resource-constrained settings. However, the study results cannot be generalized to populations that were not included in this study, particularly pregnant women and children.

In this high HIV/TB burden resource-limited setting, CXR interpretation had low diagnostic utility in HIV co-infected patients presenting with TB symptoms and negative smear. Addition of CXR to Xpert MTB/RIF increased sensitivity but reduced specificity in the diagnosis of smear-negative TB among HIV-positive patients. CXR interpretation may not add diagnostic value in settings where Xpert MTB/RIF is available as a TB diagnostic tool.