Comparison of the performance of Ultrasound versus Chest X-ray in diagnosis of Pulmonary Tuberculosis in resource limited setting

Objective

to provide evidence for the use of LUS as a reliable alternative for evaluation of TB in settings without access to Chest X-ray (CXR) or where access to CXR is problematic. We evaluated the accuracy of LUS compared to CXR for diagnosis of PTB in patients with symptoms of TB in Papua New Guinea.

Methodology

This prospective diagnostic study included patients with presumptive PTB attending the MSF TB clinic at Gerehu Hospital in Port Moresby, Papua New Guinea. Patients underwent an initial clinical examination, sputum was collected for rapid molecular testing (GeneXpert), and digital chest radiography and lung ultrasound was performed. Presumptive PTB was defined as reported cough of two weeks or more. Patients were categorized as confirmed TB (GeneXpert positive), unconfirmed TB (treatment initiated based on clinical or radiological features) or non-TB (neither of the previous conditions). The clinician performing the LUS on site as well as blinded external expert reviewers examined the LUS images and recorded the presence of findings and their location (within 16 lung regions). The LUS findings recorded were: Subpleural nodules (SUN), consolidations, cavities, miliary pattern, and pleural effusion. Additionally, the presence of pericardial effusion was also assessed. For the main analysis, a LUS highly or possibly suggestive of TB was defined as the presence of at least one of the following: SUN in two regions, with at least one located in the apical or upper region; consolidations larger than 1 cm; pericardial effusion greater than 1 cm; a miliary pattern; cavities; or pleural effusion. The main analysis was done using bacteriologically confirmed TB as the reference standard (N=433).

Results

Between 16th May 2022 and 8th May 2023, 512 patients were enrolled. The median age was 30 [IQR 23-44] years; more than half were male (250; 55.3%). Half of the participants reported having contact with a TB case (51%), most commonly as household contacts, and 20% had previously undergone TB treatment. All patients reported cough. Other frequently reported symptoms were sputum production (89%), weight loss (89%), weakness (86%), difficulty to breath (74%), fever (74%) and night sweats (73%). Mycobacterium tuberculosis was detected by GeneXpert MTB/RIF in sputum in 149/488 (30.5%) patients with a test result. Among these, rifampicin resistance was identified in two patients. The 488 participants were classified as: 149 (30.5%) confirmed TB, 55 (11.3%) unconfirmed TB, and 284 (58.2%) non-TB. 

In total, 433 participants were included in the main analysis. Among them, the most common CXR findings associated with TB diagnosis were reticulonodular infiltrates (69.8%) and cavities (54.4%), which were often multiple (81.5%) and larger than one centimeter (96.3%). Other findings such as consolidation, miliary pattern, pleural effusion and lymphadenopathy, were also associated with TB but occurred less frequently. The LUS findings associated with TB diagnosis were the following: subpleural nodules (SUN, 63.8%), consolidation ≤1 cm (78.5%) and consolidation >1 cm (75.2%), and pleural effusion (40.3%). Cavities, miliary pattern and pericardial effusion were also associated with TB but were seen much less frequently. Specifically, TB was significantly associated with SUN showing the following characteristics: more than five nodules, nodules larger than 5 mm, location in the apical or apical/upper regions, and presence in at least two regions. 

The diagnostic performance of individual CXR findings in detecting TB showed that consolidation, cavity, miliary pattern and lymphadenopathy had the highest specificities ranging between 94% and 100%. However, these findings also had very low sensitivities ranging between 5% and 54%. Only reticulonodular infiltrates showed a balance between sensitivity (70%) and specificity (79%). The diagnostic performance of individual LUS findings in detecting TB showed that cavity, miliary pattern, pericardial effusion, and the presence of an SUN greater than 5 mm had the highest specificities (ranging between 94% and 99%). However, these findings had very low sensitivities ranging between 6% and 20%. None of the LUS findings exhibited an optimal balance between sensitivity and specificity. 

The overall diagnostic accuracy of CXR and LUS in detecting TB in the main analysis showed a sensitivity of 91.3% (95% CI 86.6–93.9) for CXR and 92.6% (95% CI 90.1–95.1) for LUS. The specificity of CXR and LUS was 65.1% (95% CI 60.7–69.6) and 46.8% (95% CI 42.1–51.5), respectively. Positive predictive values of CXR and LUS were 57.9% (95% CI 53.2–62.6) and 47.8% (95% CI 43.1–52.5), respectively. Negative predictive values were 93.4% (95% CI 91.1–95.8) for CXR and 92.3% (95% CI 89.9–94.9) for LUS.

The agreement between CXR and LUS in categorizing the imaging findings as highly or possibly suggestive of TB (N=433) was moderate (0.48). This moderate agreement indicates that while both CXR and LUS can be useful for identifying TB (when classified as highly or possibly suggestive of TB), each method may detect different findings related to TB or different individuals with TB. 

Conclusions

LUS and CXR both demonstrate high sensitivity, making them effective in detecting TB. Negative LUS and CXR excluded most of the non-TB cases. However, LUS and CXR specificities were low, particularly LUS, indicating a limited value of a positive result in differentiating TB from other diseases. Despite its moderate accuracy, LUS remains an alternative diagnostic tool, particularly in settings with limited access to X-ray.