Performance of national HIV testing algorithms in 14 Population-based HIV Impact Assessment surveys: accuracy of HIV diagnosis using a two-test strategy, with or without a tie-breaker test, in different prevalence settings, 2015-2022.

UNLABELLED: Population-based HIV Impact Assessment surveys were conducted in 14 countries to measure HIV prevalence, incidence, and viral load suppression in children aged 0-14 years and adults aged ≥15 years. We examined the performance of rapid HIV testing algorithms in multiple countries. Survey participants received an HIV diagnosis based on two serial tests in Cameroon, Côte d'Ivoire, Eswatini, Haiti, Kenya, Lesotho, Malawi, Mozambique, Tanzania, and Zambia. Whereas in Ethiopia, Kenya, Namibia, Uganda, and Zimbabwe, diagnosis was based on a tie-breaker algorithm. Testers received intensive training on specimen collection, testing, and results reporting. HIV-positive and indeterminate results were confirmed by Geenius HIV-1/2 test (Bio-Rad). Pooled test 1 and 2 positive concordance, test 2 and 3 negative concordance, and the positive predictive value (PPV) of the algorithm were calculated. A total of 312,069 participants aged ≥18 months (except for Eswatini, Mozambique, and Uganda, where only adults aged 15+ years were included) were tested by either the two-test (n = 243,311) or tie-breaker (n = 98,757) algorithm for diagnosis. The combined test 1 and 2 positive concordance and PPV were 93.8% and 99.4%, respectively, for countries with a two-test algorithm. Similarly, test 1 and 2 positive concordance, test 2 and 3 negative concordance, and PPV were 87.5%, 84.4%, and 98.8%, respectively, for the tie-breaker algorithm. The PPV ranged from 94.2% to 99.9%, with higher PPV in high burden countries. The HIV testing algorithms performed with high accuracy, with PPV reaching 99% or higher. Adoption of continuous quality improvement is essential to ensure accuracy of HIV diagnosis in service delivery settings.

IMPORTANCE: HIV diagnostic testing in most African countries follows national algorithms that typically use two tests, with or without a tie-breaker. We assessed the accuracy of these algorithms using data from population-based surveys in 14 sub-Saharan African countries, where all HIV-positive results were further confirmed with the Geenius HIV-1/2 supplemental assay. Our findings show that inter-test concordance and positive predictive values (PPVs) varied by HIV prevalence, with higher PPVs observed in higher-prevalence settings. Overall, the PPV of HIV diagnosis was close to 99%, indicating that two-test algorithms can provide highly accurate results when testing is performed with strict adherence to quality standards and tester competency. These results underscore the importance of quality assurance (QA) and suggest that countries with lower HIV prevalence may benefit from adopting a three-test algorithm. However, such changes should be accompanied by careful attention to logistics, procurement, training, record keeping, and other QA measures.