
Background and motivation: Congenital cytomegalovirus (cCMV) is the most common intrauterine infection. Approximately 20% of infected infants develop neurodevelopmental disabilities or sensorineural hearing loss (SNHL). Current prenatal assessment is limited, leading to difficult clinical and parental decisions. We aimed to investigate the prognostic capacity of our newly-identified amniotic fluid biomarkers chemerin and galectin-3-binding protein (Gal-3BP) in a large multi-center cohort and leverage machine learning to develop a robust prediction model for cCMV severity.
Specific questions: Can chemerin and Gal-3BP levels distinguish between severe and asymptomatic cCMV?
Can a machine learning algorithm combine these biomarkers into a robust prediction model?
Approach and results: Chemerin and Gal-3BP levels were measured by specific immunoassays in mid-gestation amniotic fluid from total of 207 infected fetuses: severe cCMV (N=64), isolated SNHL (N=19), and asymptomatic (N=124). Both biomarkers were significantly elevated in severe versus asymptomatic cases (AUC >0.95). Isolated SNHL demonstrated intermediate levels (chemerin AUC 0.86; Gal-3BP AUC 0.70). A Support Vector Machine SVM model (SVM), integrating both biomarkers, considerably enhanced the discrimination capacity between severe and asymptomatic cCMV. Moreover, SVM classification at the time of amniocentesis preceded the appearance of imaging abnormalities in most (19/20) severe cases with delayed imaging abnormalities. Isolated SNHL cases could not be discriminated by the SVM model.
Conclusions: This study demonstrates the high predictive accuracy of the new amniotic fluid biomarkers chemerin and Gal-3BP when integrated in a machine learning model, which could enable prenatal risk assessment, consultation for affected families, and targeted management of cCMV.