An evidence-based algorithm for the rapid diagnosis of tuberculosis in HIV positive patients presenting to emergency centres

Abstract

Background Tuberculosis remains a prevalent and deadly global disease. Diagnostic delays are partly due to reduced diagnostic performance of tuberculosis tests in HIV-positive people. The use of reliable pointof-care and near-patient diagnostic tests (e.g. urine lipoarabinomannan and point-of-care ultrasound) are increasingly being used and would benefit patients presenting to emergency centres by rapidly diagnosing HIV-associated tuberculosis. Methods Two studies were done: i) A systematic (Cochrane) review was done to determine the diagnostic accuracy of abdominal ultrasound for detecting abdominal tuberculosis or disseminated tuberculosis with abdominal involvement in HIV-positive individuals, and ii) A cross-sectional diagnostic study to derive a multi-parameter clinical decision tree, incorporating clinical information, point-of-care ultrasound features, chest x-ray and urine lateral flow lipoarabinomannan. The cross-sectional study was performed at the emergency centre of Khayelitsha Hospital, a South African district-level hospital in a high HIV-prevalence community, and resulted in three different publications. Consecutive HIV-positive adults presenting with ≥1 WHO tuberculosis symptoms were enrolled over a 16-month period (June 2016 to October 2017). Demographic and clinical information was recorded on a standardized data collection form. Point-of-care ultrasound was performed according to a standardized protocol. Urine lipoarabinomannan assays were done at point-of-care by emergency physicians and repeated in the laboratory. Chest x-rays were reviewed by a single radiologist using a standardized assessment form. The reference standard was a positive tuberculosis culture or Xpert MTB/RIF test on sputum, or appropriate extra-pulmonary samples. We compared diagnostic accuracy and reproducibility of urine lipoarabinomannan between point-ofcare readers and laboratory readers. We determined the diagnostic accuracy of individual point-ofcare ultrasound features, performed an external validation of the focused assessment with sonography for HIV/TB (FASH) protocol, and determined independent point-of-care ultrasound predictors of HIV-associated tuberculosis. We derived the decision tree model from multivariable logistic regression models. Results Abdominal ultrasound had a pooled sensitivity of 63% (95%CI 43-79; 5 studies, 368 participants; very low-certainty evidence) and a pooled specificity of 68% (95%CI 42-87; 5 studies, 511 participants; very low-certainty evidence) for bacteriologically confirmed tuberculosis. We screened 556 patients in the cross-sectional study of whom 414 (74.5%) were enrolled. The prevalence of microbiologically confirmed tuberculosis was 41.5% (n=172). Point-of-care and laboratory-performed urine lipoarabinomannan had similar sensitivity (41.8% vs 42.0%, P=1.0) and specificity (90.5% vs 87.5%, P=0.23). Moderate agreement was found between point-of-care and laboratory testing (k=0.62), but there was strong agreement between point-of-care readers (k=0.95) and between laboratory readers (k=0.94). Sensitivity and specificity of ≥1 individual point-of-care ultrasound feature were 73% (95%CI 65-79) and 54% (95%CI 47-60), and of the FASH protocol 71% (95%CI 64-78) and 57% (95%CI 50-63). Independent point-of-care ultrasound predictors identified were intra-abdominal lymphadenopathy of any size (aDOR 3.7; 95%CI 2.0-6.7), ascites (aDOR 3.0; 95%CI 1.5-5.7), and pericardial effusion of any size (aDOR 1.9; 95%CI 1.2-3.0). Two or more independent point-of-care ultrasound predictors had 33% (95%CI 27–41) sensitivity and 91% (95%CI 86-94) specificity. The best performing model included WHO screening symptoms ≥2, antiretroviral therapy use, urinary lipoarabinomannan, independently predictive point-of-care ultrasound features (ascites, any size pericardial effusion, any size intra-abdominal lymphadenopathy), and chest x-ray (c-statistic 0.82; 95%CI 0.78–0.86). Adding CD4 cell count did not improve the performance of the model. Classification And Regression Tree (CART) analysis positioned urinary lipoarabinomannan as the optimal screening test after WHO symptoms (75% true positive rate, representing 17% of participants). Conclusion An evidence-based algorithm for the rapid diagnosis of tuberculosis in HIV-positive patients presenting to an emergency centre was developed. Urinary lipoarabinomannan can be reliably performed at the point-of-care since there was no diagnostic accuracy advantage in laboratory-performed versus pointof-care–performed tests. The role of ultrasound in diagnosing HIV-associated tuberculosis had limitations. The low sensitivity of ultrasound (63% in the systematic review; 73% in the cross-sectional study) and the moderate discrimination (specificity 91%) of the presence of ≥2 independent point-ofcare ultrasound predictors indicate that point-of-care ultrasound results should be interpreted in combination with other diagnostic information. The derived decision tree can facilitate the immediate initiation of anti-tuberculosis treatment in about a quarter of patients among whom 75% would have a definitive diagnosis of tuberculosis regardless of CD4 cell count. The 30% false negative rate indicates that the algorithm should not be used to exclude tuberculosis. The performance of the decision tree needs to be further evaluated in settings with a different prevalence of HIV-associated tuberculosis.