Artificial Intelligence-Enabled Electrocardiography for Potassium Abnormality Detection and Estimation: A Systematic Review

Laboratory measurement of serum, plasma, or whole-blood potassium remains the reference standard for diagnosing hyperkalemia and hypokalemia, yet results are not always immediately available when rapid clinical decisions are required. Artificial intelligence-enabled electrocardiography (AI-ECG) has been proposed as an early, non-invasive tool that may identify potassium abnormalities while laboratory confirmation is pending. This systematic review evaluated adult studies published between 2016 and 2026 in which AI or machine-learning (ML) techniques were applied to ECGs to detect hyperkalemia or hypokalemia, classify potassium status, or estimate continuous potassium concentration, always against a paired laboratory potassium reference standard. We searched multiple biomedical, preprint, trial, and citation databases through June 18, 2026, following PRISMA-DTA, PRISMA 2020, and PRISMA-S reporting principles. Methodological quality was assessed with QUADAS-2. Of 366 records assessed for eligibility, 33 peer-reviewed primary AI/ML studies were included. Hyperkalemia detection dominated the evidence base, with many externally or internally validated studies reporting area under the receiver operating characteristic curve (AUROC) in the high 0.8 to mid-0.9 range, whereas evidence for hypokalemia detection, categorical potassium-status classification, and continuous potassium estimation remained overlapping and more heterogeneous. Although reported performance was often encouraging, overall risk of bias was low in only four studies, unclear in 23, and high in six. Considerable heterogeneity in potassium thresholds, ECG modalities, validation strategies, reporting units, and performance metrics precluded meta-analysis. Overall, the available evidence suggests that AI-ECG is a promising investigational adjunct for triage, monitoring, and clinical decision support, primarily serving as a rule-out tool rather than a replacement for laboratory potassium testing. Its clinical utility may be particularly relevant in higher-prevalence populations, such as patients with chronic kidney disease or those receiving dialysis, where positive predictive value and monitoring yield may be more favorable. Prospective external validation, calibration, subgroup reporting, reproducibility, and workflow-impact studies are needed before routine clinical implementation.