A train-the-trainer curriculum to scale up artificial intelligence-supported echocardiography for rheumatic heart disease screening in a public health care system

Doreen Nakagaayi, Jafesi Pulle, Jenifer Atala, Linda M. Oyella, Sarah de Loizaga, Neema W. Minja, Nicholas J. Ollberding, Ndate Fall, Miriam Nakitto, Rachel Sarnacki, Joselyn Rwebembera, Emmy Okello, Andrea Beaton, Craig Sable

Abstract

Background: This study aimed to test a train-the-trainer model for scaling up echocardiographic screening for rheumatic heart disease (RHD).
Methods: This was a two-phased, prospective, pilot cohort study conducted in Lira, Uganda. In phase one, four nurse centres underwent a three-day training on echocardiographic screening for RHD. Thereafter, nurses integrated screening of patients aged 5-40 years into their routine clinic workflow. Echocardiograms were uploaded to a cloud-based server for review by an expert, who served as the gold standard for interpretation. In phase two, 12 nurses were trained by trainees included in phase one.
Results: A total of 16 nurses (four in phase one and 12 in phase two) were recruited. All nurses achieved the target of acquiring 100 echocardiograms, resulting in a total of 1620 screening echocardiograms (n = 406 in phase one and n = 1214 in phase two). Over 95% of the studies were of diagnostic quality. There were 44/1473 (3%) screen-positive participants, 40 with RHD. There was a decline in sensitivity from 70% to 50% (p = 0.33) and an increase in specificity from 86% to 94% (p < 0.0001) between phases one and two. All four cases of moderate/severe RHD were correctly identified. In phase one, 56 cases were incorrectly classified as left ventricular dysfunction using the auto ejection fraction function; this option was omitted for phase two.
Conclusion: This pilot study showed that utilising a train-the- trainer model to implement echocardiographic screening for RHD into primary health care in a low-resource setting is feasible.