Performance evaluation of the Uganda National Algorithm for Diagnosis of tuberculosis in children at Mulago National Referral Hospital-Kampala, Uganda

Abstract

Background Diagnosing childhood tuberculosis (TB) is a challenge, this led the Uganda national tuberculosis control program (NTLP) to develop a clinical treatment decision algorithm for children. However, there is limited data on its accuracy and how it compares to new World Health Organization (WHO) treatment decision algorithms for children. Objective To evaluate the performance of the 2017 Uganda NTLP and 2022 WHO algorithms against the 2015 NIH Consensus classification (as a reference) for diagnosing Tuberculosis among children presenting at the Mulago National Referral Hospital Pediatric TB clinic. Methods This was a secondary data analysis of children 0-14 years from Kampala, Uganda who underwent an evaluation for pulmonary TB disease (physical examination, chest x-ray, tuberculin skin testing, HIV testing, and respiratory specimen collection for Gene-X-pert testing and culture) between September 2018 and November 2022. The diagnostic accuracy of the national TB diagnosis and WHO algorithms was determined using the National Institute of Health (NIH) consensus classification (as the reference standard). Results Overall, 732 children were included in this analysis with 64% under 5 years, 53% were male, and 6% had severe acute malnutrition. Approximately 11% were HIV positive, 56% had a history of TB contact, 62% had a normal CXR and 13% had positive Gene-X-pert. History of TB contact, weight loss and abnormal CXR were strongly associated with TB. The sensitivity and specificity of the NTLP algorithm were 90.3% (95% CI: 88.2 - 92.5) and 26.0% (95% CI: 22.7 - 29.2) respectively. The sensitivity and specificity of the WHO algorithm were 90.0% (95% CI: 87.7 - 92.3) and 30.2% (95% CI: 26.7 - 33.7) respectively. In settings where Gene-X-pert was available but no CXR, the NTLP algorithm had a PPV and NPV of 21.1% (95% CI: 18.1- 24.1) and 93.4% (95% CI: 91.6 – 95.3) respectively compared to PPV of 20.6% (95% CI: 17.7- 23.6) and NPV of 93.4% (95% CI: 91.6 – 95.2) in a setting with both CXR and Gene-X-pert. Both algorithms’ performance was comparable in settings with CXR. Conclusion The NTLP algorithm performs as well as the WHO algorithm, when compared with the composite NIH classification, but both are likely to lead to over diagnosis of TB in children 0-14 years due to their low specificity. Key words: Childhood TB, Diagnosis Algorithm, Performance.