Evaluation of six different tests for Schistosoma haematobium diagnosis in a near-elimination setting: A prospective observational diagnostic accuracy study

Accurate diagnostic tools are needed in schistosomiasis elimination settings to determine prevalence thresholds for assigning or stopping interventions, guide pre- and post-elimination surveillance, and verify whether elimination has been reached. We assessed the accuracy of six different diagnostic tests in Pemba, Tanzania, a setting approaching Schistosoma haematobium elimination.

A prospective diagnostic accuracy study was conducted from February to April 2025. From an initial cross-sectional single-day urine filtration (UF)-microscopy screening of 784 students, 69 S. haematobium-positive and 212 negative students were randomly selected for longitudinal follow-up. Four additional urine samples collected over four different days, were available from 262/281 participants and subjected to UF-microscopy. One sample per participant was analysed in parallel with five additional diagnostics: microscopy-based artificial intelligence (AI) scanner, Schistosoma-ITS-2 qPCR, S. haematobium-Dra-1 recombinase polymerase amplification (RPA), Hemastix reagent strips, and up-converting particle lateral flow circulating anodic antigen assay (UCP-LF CAA). We assessed the sensitivity and specificity of the different diagnostics, using 5-day UF-microscopy as reference test.

A total of 85/262 participants were S. haematobium-positive using 5-day UF-microscopy. Directly compared with the reference test, the sensitivity for single-sample examination was: AI scanner: 76.7% (95% confidence interval (CI): 71.0-82.5%), qPCR: 76.0% (95% CI: 70.1-81.4%), UF-microscopy: 61.2% (95% CI: 55.3-67.1%), RPA: 56.1% (95% CI: 50.0-62.2%), Hemastix: 44.6% (95% CI: 38.5-50.7%), and UCP-LF CAA: 30.6% (95% CI: 24.9-36.3%). Sensitivity increased with increasing infection intensity. The specificity of all investigated diagnostics was >92%, except for qPCR and RPA.

In near-elimination settings, multiple-day urine examination with standard UF-microscopy substantially improves case detection but is operationally challenging. For single-sample testing, among the six diagnostics investigated, the AI scanner proved to be the most accurate. Hence, the AI scanner might offer a promising alternative for research, clinical and programme use, but requires further validation in other settings and cost-effectiveness analyses. Trial registration: clinicaltrials.gov, NCT06808750. Registered 08 January 2025, https://clinicaltrials.gov/study/NCT06808750.