Why the study?
Electrolyte imbalances significantly affect heart function, and evaluating AI models for ECG-based detection is needed to enhance early detection and improve cardiac care.
Do deep learning or conventional machine learning-based AI models accurately detect hyperkalemia and hypokalemia from ECGs in patients aged 15 years and older?
Meta-Analysis (n=671,062)
Do deep learning or conventional machine learning-based AI models accurately detect hyperkalemia and hypokalemia from ECGs in patients aged 15 years and older?
AI models demonstrate strong diagnostic accuracy for detecting potassium imbalances, particularly hyperkalemia, from ECGs, offering a potential non-invasive monitoring tool.
Background Electrolyte imbalances significantly affect heart function, making electrocardiography (ECG) a crucial non-invasive tool. This study systematically reviewed and meta-analyzed AI models’ diagnostic accuracy for detecting these imbalances from ECG, aiming to enhance early detection and improve cardiac care. Methods We searched 9 databases and reference lists. Two reviewers assessed bias via the Quality Assessment of Diagnostic Accuracy Studies-2 (QUADAS-2). Test performance data were extracted into 2×2 tables and pooled estimates of specificity, sensitivity, and diagnostic odds ratio (DOR) were calculated using a bivariate random-effects model, presented in forest plots and summary receiver operating characteristic curves. We explored heterogeneity by meta-regression, examining internal/external datasets and the number of leads. Results 21 studies addressing potassium, calcium, and sodium were included. A meta-analysis was conducted only on potassium imbalances (10 studies), covering over 600,000 ECGs from five countries, mostly 12-lead. Among eight studies focused on hyperkalemia, pooled sensitivity, specificity, and DOR were 0.856 (95% CI: 0.829–0.879), 0.788 (0.744–0.826), and 21.8 (17.8–26.7). For hypokalemia (six studies), pooled sensitivity, specificity, and DOR were 0.824 (0.785– 0.856), 0.724 (0.668–0.774), and 12.27 (9.15–16.47). QUADAS-2 assessment showed a 52% high risk of bias in patient selection, mainly due to inadequate sampling details and case-control approaches. Conclusion AI models can detect ECG-based electrolyte abnormalities, particularly hyperkalemia, and valuable in ICU settings requiring frequent electrolyte assessments and in home monitoring for patients with end-stage renal disease. However, larger retrospective and prospective studies across diverse clinical settings, hospitals, ethnic groups, countries, and regions are warranted.
No takes yet. Share an insight, caveat, or question.
Dasdelen et al. (2025) conducted a meta-analysis in Electrolyte imbalance (hyperkalemia and hypokalemia) (n=671,062). AI models for ECG-based electrolyte imbalance prediction vs. Reference standards (serum analysis) was evaluated on Diagnostic accuracy (diagnostic odds ratio) for hyperkalemia (DOR 21.8, 95% CI 17.8-26.7).
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: