Human versus machine: Assessing the Allplex™ bacterial vaginosis plus assay (Seegene®) using Ison-Hay criteria and cut-off values interpretations in pregnant population

Bacterial vaginosis (BV) is a common dysbiotic condition associated with adverse pregnancy outcomes, including preterm delivery, and its diagnosis remains challenging due to limitations of conventional methods and the lack of standardized molecular criteria. In this cross-sectional study, we evaluated the diagnostic performance of the Allplex™ Bacterial Vaginosis Plus assay, a multiplex real-time quantification polymerase chain reaction (qPCR), compared with Gram stain using the Ison-Hay criteria in a population of pregnant female, and explored optimal quantitative cut-off values for BV-associated bacteria (BVAB). A total of 90 pregnant female with a median age of 37 (95% CI: 35-38) years were included during the BV screening process. According to the Ison-Hay criteria, 6 patients (6.7%) were classified as BV (grade III), whereas the Allplex™ BV Plus assay identified BV in 19 patients (21.1%). Using Gram staining as the reference standard and considering intermediate flora/Gram grade II as negative, qPCR demonstrated a sensitivity of 100% (95%CI: 78-100%) and a specificity of 89.3% (95%CI: 79-95%), with a positive predictive value (PPV) of 40% (95%CI: 23-59%), and a negative predictive value (NPV) of 100% (95% CI: 92-100%). Agreement between methods was moderate (κ=0.53, 95%CI:0.35-0.71), and a statistically significant difference was observed (p < 0.05). Quantitative analysis showed that Gardnerella vaginalis (AUC = 0.93), Atopobium vaginae (AUC = 0.99), and Megasphaera type 1 (AUC = 0.87) exhibited strong discriminatory capacity. Lactobacillus spp. also showed good overall performance (AUC = 0.88) with high sensitivity but lower specificity, whereas BVAB2 (AUC = 0.55) and Mobiluncus spp. (AUC = 0.61) demonstrated poor diagnostic performance. The molecular assay tended to classify more cases as both normal flora and BV compared to microscopy, reflecting differences in diagnostic frameworks. Overall, qPCR demonstrated excellent performance for ruling out BV but a tendency to overestimate BV cases. These findings highlight the complexity of interpreting multiplex molecular results and support the use of combined diagnostic approaches. In clinical practice, qPCR may be particularly useful as a rule-out tool, while BV-positive results should be interpreted cautiously and in conjunction with microscopy and clinical findings. Further studies are needed to assess the clinical relevance of molecular diagnosis in relation to pregnancy outcomes.