Acute Rheumatic Fever (ARF) diagnosis requires advanced clinical and laboratory investigations that are not available in the primary care setting. This study explores the potential of a possible point-of-care device (POCD) for ARF screening.
We investigated antibody levels to host proteins (cardiac myosin, laminin, keratin, and tropomyosin) in two experimental models: (i) a rat autoimmune valvulitis model (30 control and 30 streptococcal M protein-injected rats); and (ii) human clinical samples (25 ARF patients and 50 healthy individuals). We implemented four machine learning methods (logistic regression, decision tree, random forest and AdaBoost) to predict whether an individual has ARF/RHD or not using the antibody levels detected by ELISA. Significant antibody level differences between both rat groups and human cohorts were observed. The random forest was the best model in terms of prediction accuracy for rat dataset (accuracy 100%, area under the receiver operating characteristic curve (AUC) = 1) and for human samples (accuracy 83%, AUC =0.87).
With currently available technology, multiple analytes can be detected on a single lateral flow assay (LFA) and using cost-effective smart-phone technology subjective interpretation of LAF results can be refined. We have shown with up to four already identified antibodies as biomarkers, algorithms can be trained on LFA images that could enhance sensitivity and reliability of LAF. Such POCD for screening ARF/RHD would have to be tested in multiple sites through international collaboration prior to being made available for use in resource-limited settings.