Lin AI-ECG (potassium)

AI-Enabled Electrocardiogram Alert for Potassium Imbalance Treatment: a Pragmatic Randomized Controlled Trial

Patient / Population Intervention / Exposure Comparison Outcome

In 14,989 patients cared for by 70 randomised emergency physicians at an academic medical centre and a community hospital, a real-time AI-ECG pop-up alert for moderate-to-severe hyperkalaemia or hypokalaemia did not improve rates of hyperkalaemia- and hypokalaemia-related treatment within three hours compared with usual care without alerts.

N
14,989
Design
Pragmatic open-label RCT with physician-level randomisation (70 emergency physicians, two hospitals)
Endpoint
Rate of hyperkalaemia-related treatment and rate of hypokalaemia-related treatment within three hours (two primary outcomes)
Relevance
2
ResultHyperkalaemia-related treatment 8.0% vs 7.7% (HR 1.05, 95% CI 0.94–1.17; P=0.420); hypokalaemia-related treatment 2.1% vs 2.4% (HR 0.91, 95% CI 0.74–1.13; P=0.392). Among patients flagged by AI-ECG as hyperkalaemic, treatment 69.1% vs 41.6% (HR 2.23, 95% CI 1.44–3.46; P<0.001).
Lin C, et al. Nat Commun. 2026;17(1):159. 10.1038/s41467-025-66394-4
Discussion & critique

A negative AI-ECG trial: at the population level the alert did not change treatment of either potassium disorder, and benefit appeared only in the subgroup the AI actually flagged. It teaches that an alert for a condition routine laboratory testing already catches quickly has little room to add at scale, and that subgroup gains do not translate into trial-level effects. Open-label, physician-level randomisation in two hospitals carries a risk of contamination, and the endpoints are process measures rather than outcomes.