Key result
Uncalibrated single-lead AI-ECG transfer is predicted to inflate pediatric RHD referrals ~2-fold.
Why the study?
Portable single-lead ECG devices promise to extend RHD screening into schools lacking echocardiography, but lead placement varies with operator and child anatomy, so AI models trained on standard 12-lead recordings may not transfer.
Does uncalibrated transfer of AI-ECG models to portable single-lead devices under variable lead placement affect healthcare referral yield and positive predictive value in school-age children screened for rheumatic heart disease?
Does uncalibrated transfer of AI-ECG models to portable single-lead devices under variable lead placement affect healthcare referral yield and positive predictive value in school-age children screened for rheumatic heart disease?
Effect estimate: Referral volume inflation factor 1.5 to 2.3; PPV <20%
Uncalibrated transfer of 12-lead AI-ECG models to portable single-lead devices for school-based rheumatic heart disease screening can inflate referrals by 1.5- to 2.3-fold and drop positive predictive value below 20%, highlighting calibration as a critical criterion for deployment.
Uncalibrated AI-ECG transfer may inflate referrals and lower PPV in school screening; leaves open calibration strategies for portable devices.
Portable single-lead electrocardiogram (ECG) devices promise to extend rheumatic heart disease (RHD) screening into schools where echocardiography is unavailable, but their lead placement varies with operator and child anatomy, and artificial intelligence (AI) models trained on standard 12-lead recordings may not transfer. The framework separates lead-placement-induced distribution shift into geometric, impedance and operator components, and derives a referral-yield bound that depends on calibration slope and intercept under shift. Applied to published estimates, the framework predicts that uncalibrated transfer can inflate referral volume by a factor of 1.5 to 2.3 while reducing positive predictive value below 20% in low-prevalence school populations. The DHS analysis shows that household computer ownership, a proxy for digital infrastructure, remains low (mean 7.893%) and highly variable (range 1.50–22.00%), constraining where AI-ECG screening can be deployed with adequate connectivity. The contribution is a transferable evaluation protocol that makes calibration, not discrimination, the primary criterion for referral safety in RHD screening.
No takes yet. Share an insight, caveat, or question.
Gakwaya et al. (2026) studied Rheumatic heart disease. Uncalibrated transfer of 12-lead AI-ECG models to portable single-lead ECG devices vs. Calibrated transfer / standard 12-lead ECG recordings was evaluated on Healthcare referral volume and positive predictive value (PPV) (Referral volume inflation factor 1.5 to 2.3; PPV <20%). Uncalibrated transfer of AI-ECG models to portable single-lead devices is predicted to inflate RHD referral volume by 1.5 to 2.3-fold and reduce PPV below 20% in school-age children.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: