The research, led by an international team spanning institutions in the United States, Europe, and West Africa, pooled participant-level data from three field studies conducted in 2023 in Mozambique, Ghana, and Benin. Rather than evaluating each test in isolation, the investigators brought together individuals tested across all three countries, creating a combined dataset large enough to support sophisticated statistical modeling. The tests under scrutiny included the commercially available SD BIOLINE Ov16 rapid diagnostic test, which has been widely deployed in onchocerciasis mapping exercises, and two novel assays developed more recently: the DDTD rapid diagnostic test and the GADx rapid diagnostic test. All three are designed to detect antibodies to the Ov16 antigen, a marker of exposure to the parasitic worm that causes river blindness.

The central methodological challenge the researchers confronted is one that plagues much of diagnostic science in neglected tropical diseases: there is no perfect reference standard against which a new test can be judged. The gold standard for confirming an O. volvulus infection involves detecting microfilariae in skin snips, a procedure that is invasive, technically demanding, and critically insensitive in areas where infection levels have been driven down by years of treatment. In low-prevalence settings, a skin snip examination can miss the vast majority of true infections, making it a poor benchmark. Serological tests that detect antibodies offer a more practical alternative, but evaluating their accuracy without a definitive gold standard requires a different analytical approach altogether.

Enter the Bayesian latent class model, the statistical engine at the heart of this study. Latent class models treat the true infection status of each participant as an unobserved, or latent, variable, and then use the patterns of agreement and disagreement among multiple imperfect tests to infer both the true prevalence of infection and the sensitivity and specificity of each test simultaneously. The Bayesian framework adds prior information, formal mathematical expressions of what is already known or believed about test performance before the data are examined. By running the models under multiple specifications, including different priors reflecting laboratory-based and field-based expectations, the researchers could assess how sensitive their conclusions were to the assumptions embedded in the analysis. This approach, while computationally intensive, avoids the circular reasoning that arises when an imperfect test is treated as a perfect reference.

The headline finding concerns the GADx rapid diagnostic test, which emerged as the clear sensitivity leader across every model specification the team examined. Posterior median estimates of its sensitivity ranged from 92.0 percent to 92.8 percent, consistently clearing the 89 percent threshold recommended by the World Health Organization for tests used in decisions to stop mass drug administration. In practical terms, this means the GADx assay missed fewer than one in ten true antibody-positive cases in the studied populations, a level of performance that brings it meaningfully closer to the reliability that elimination programs require. The consistency of this result across alternative model assumptions lends considerable weight to the conclusion, since a finding that survives multiple analytical specifications is far more robust than one that depends on a single set of choices.

The performance of the other two tests tells a more nuanced story. The SD BIOLINE Ov16 test, the only commercially available option among the three, produced posterior median sensitivity estimates above the 89 percent threshold in most models, but fell to 83.2 percent under the field-based prior, a specification that reflects performance expectations drawn from real-world deployment conditions rather than idealized laboratory settings. This inconsistency echoes earlier reports that the SD BIOLINE test shows variable sensitivity when used on whole blood in field conditions. The DDTD rapid diagnostic test, a multi-antigen assay, demonstrated lower sensitivity overall, with posterior median estimates ranging from 86.6 percent to 88.4 percent, placing it just below the recommended threshold. Notably, the researchers tested alternative definitions of a positive result for the multi-antigen DDTD test and found that these did not meaningfully alter its estimated performance, suggesting the result was not an artifact of how positive readings were scored.

Specificity, the ability of a test to correctly identify people who are not infected, presented a different kind of problem. All three tests achieved consistently high specificity estimates, with the highest posterior median reaching 98.8 percent. That figure sounds impressive, and in most clinical contexts it would be. But the WHO-recommended threshold for tests supporting decisions to stop mass drug administration is 99.8 percent, a demanding standard that reflects the consequences of false positives in low-prevalence settings. When transmission has been reduced to very low levels, even a small false-positive rate can generate a misleading picture of ongoing transmission, potentially triggering unnecessary rounds of treatment, wasted resources, and delayed declarations of elimination. None of the evaluated tests, including the high-performing GADx assay, reached that specificity bar on its own.

The implications of this gap are significant for the global elimination agenda. Onchocerciasis programs increasingly rely on serological surveys of children, who should show little or no antibody evidence of exposure if transmission has been interrupted, to guide decisions about stopping treatment. A test that combines high sensitivity with extremely high specificity is essential for these surveys to be trustworthy. The study’s findings suggest that while the novel rapid tests, particularly the GADx assay, represent genuine progress over the existing commercial option, no single test currently available can shoulder the entire burden of an elimination decision. The authors conclude that improved diagnostic tools or complementary testing strategies will be needed, which could involve combining tests in sequence, using different sample types, or integrating serological data with entomological surveillance of the blackfly vectors that transmit the parasite.

The geographic breadth of the underlying data strengthens the study’s relevance across the African continent, where the vast majority of onchocerciasis burden is concentrated. Field studies in Mozambique, Ghana, and Benin captured diverse epidemiological settings, from areas with historically intense transmission to those where control efforts have been sustained for many years. Pooling participant-level data, rather than combining summary statistics from separate analyses, allowed the Bayesian models to account for heterogeneity across sites and to borrow statistical strength from the full dataset. This design choice matters because diagnostic performance can vary with factors such as infection intensity, population age structure, and the prevalence of other parasitic infections that might cross-react with the Ov16 antigen.

For the elimination community, the study offers both reassurance and a call to action. The demonstration that rigorous, assumption-aware evaluation of rapid diagnostic tests is feasible using Bayesian latent class methods provides a template for assessing future diagnostics, not only for onchocerciasis but for other diseases approaching elimination where gold standards are similarly lacking. At the same time, the finding that no evaluated test meets the specificity threshold on its own underscores that the final mile of the river blindness campaign will demand continued investment in diagnostic development. The GADx test’s sensitivity performance marks a milestone worth celebrating, but the road to confident, evidence-based decisions about stopping mass drug administration will require tools that are both more sensitive and more specific than anything currently in the field. Until then, program managers must weigh these imperfect instruments carefully, knowing that the health of millions and the credibility of the elimination enterprise rest on getting the answer right.

Subject of Research: Diagnostic accuracy of Ov16 rapid diagnostic tests for onchocerciasis elimination using Bayesian latent class models

Article Title: Assessing diagnostic accuracy of onchocerciasis rapid diagnostic tests using Bayesian latent class models

Article References: Norman, J., Bassabi-Alladjie, N.-M., Boko-Collins, P. M., de Souza, D. K., Gass, K., Hamill, L., Langa, J., Moore, C., Mucato, N., Nala, R., Sullivan, S., & Giorgi, E. (2026). Assessing diagnostic accuracy of onchocerciasis rapid diagnostic tests using Bayesian latent class models. PLOS Neglected Tropical Diseases, 20(10), e0013954. https://doi.org/10.1371/journal.pntd.0013954

Image Credits: AI Generated

DOI: 10.1371/journal.pntd.0013954

Keywords: onchocerciasis, river blindness, rapid diagnostic tests, Bayesian latent class models, Ov16 antigen, diagnostic accuracy, mass drug administration, neglected tropical diseases, GADx, SD BIOLINE, elimination, serology