Balancing HIV testing efficiency with HIV case-identification among children and adolescents (2-19 years) using an HIV risk screening approach in Tanzania.

To optimize HIV testing resources, programs are moving away from universal testing strategies toward a risk-based screening approach to testing children/adolescents, but there is little consensus around what defines an optimal risk screening tool. This study aimed to validate a 12-item risk screening tool among children and adolescents and provide suggested fewer-item tool options for screening both facility out-patient and community populations by age strata (<10 and ≥10 years). Children/adolescents (2-19 years) with unknown HIV status were recruited from a community-based vulnerable children program and health facilities in 5 regions of Tanzania in 2019. Lay workers administered the screening questions to caregivers/adolescents; nurses enrolled those eligible for the study and tested all participants for HIV. For each screening item, we estimated sensitivity, specificity, positive predictive value and negative predictive value and associated 95% confidence intervals (CI). We generated a score based on the count of items with a positive risk response and fit a receiver operating characteristic curve to determine a cut-off score. Sensitivity, specificity, positive predictive value (PPV; yield) and number needed to test to detect an HIV-positive child (NNT) were estimated for various tool options by age group. We enrolled 21,008 children and adolescents. The proportion of undiagnosed HIV-positive children was low (n = 76; 0.36%; CI:0.29,0.45%). A screening algorithm based on reporting at least one or more items on the 10 to 12-item tool had sensitivity 89.2% (CI:79.1,95.6), specificity 37.5% (CI:36.8,38.2), positive predictive value 0.5% (CI:0.4,0.6) and NNT = 211. An algorithm based on at least two or more items resulted in lower sensitivity (64.6%), improved specificity (69.1%), PPV (0.7%) and NNT = 145. A shorter tool derived from the 10 to 12-item screening tool with a score of "1" or more on the following items: relative died, ever hospitalized, cough, family member with HIV, and sexually active if 10-19 years performed optimally with 85.3% (CI:74.6,92.7) sensitivity, 44.2% (CI:43.5,44.9) specificity, 0.5% (CI:0.4,0.7) PPV and NNT = 193. We propose that different short-tool options (3-5 items) can achieve an optimal balance between reduced HIV testing costs (lower NNT) with acceptable sensitivity. In low prevalence settings, changes in yield may be negligible and NNT may remain high even for an effective tool.