Point-of-care artificial intelligence-enabled ECG for dyskalemia: a retrospective cohort analysis for accuracy and outcome prediction

Dyskalemias are common electrolyte disorders associated with high cardiovascular risk. Artificial intelligence (AI)-assisted electrocardiography (ECG) has been evaluated as an early-detection approach for dyskalemia. The aims of this study were to determine the clinical accuracy of AI-assisted ECG for dyskalemia and prognostic ability on clinical outcomes such as all-cause mortality, hospitalizations, and ED revisits. This retrospective cohort study was done at two hospitals within a health system from May 2019 to December 2020. In total, 26,499 patients with 34,803 emergency department (ED) visits to an academic medical center and 6492 ED visits from 4747 patients to a community hospital who had a 12-lead ECG to estimate ECG-K+ and serum laboratory potassium measurement (Lab-K+) within 1 h were included. ECG-K+ had mean absolute errors (MAEs) of ≤0.365 mmol/L. Area under receiver operating characteristic curves for ECG-K+ to predict moderate-to-severe hypokalemia (Lab-K+ ≤3 mmol/L) and moderate-to-severe hyperkalemia (Lab-K+ ≥ 6 mmol/L) were >0.85 and >0.95, respectively. The U-shaped relationships between K+ concentration and adverse outcomes were more prominent for ECG-K+ than for Lab-K+. ECG-K+ and Lab-K+ hyperkalemia were associated with high HRs for 30-day all-cause mortality. Compared to hypokalemic Lab-K+, patients with hypokalemic ECG-K+ had significantly higher risk for adverse outcomes after full confounder adjustment. In addition, patients with normal Lab-K+ but dyskalemic ECG-K+ (pseudo-positive) also exhibited more co-morbidities and had worse outcomes. Point-of-care bloodless AI ECG-K+ not only rapidly identified potentially severe hypo- and hyperkalemia, but also may serve as a biomarker for medical complexity and an independent predictor for adverse outcomes.


Supplementary Figure 1 | Scatter plots of ECG-K + and Lab-K + at an academic medical center and a community hospital. The
x-axis indicates the Lab-K + . The y-axis presents the ECG-K + . Red points represent the highest density, followed by yellow, green light blue, and dark blue. Perfect model performance would fall only along the diagonal line. We presented the Pearson correlation coefficients (COR) and mean absolute errors (MAE) to compare ECG-K + estimated via information from 12 leads and each lead.

Supplementary Figure 2 | Stratified analyses for the performance of ECG-K + for detecting mild to severe hypo/hyper-kalemia.
The sensitivity and specificity are tabulated across a series of stratified conditions. The diagnostic OR, which is the ratio of positive likelihood ratio (sensitivity/(1- specificity)) to the negative likelihood ratio ((1-sensitivity)/specificity), as well as the associated 95% CI, is shown for each situation. All analyses were checked for problematic zero counts, which were remedied by adding a fixed value of 0.5 to all cells where the problem occurred. The ECG-K + is estimated via information from 12 leads in this analysis. The pinteraction was the significance test of strength of association, with an adjusted significance level of 0.001 based on Bonferroni correction. Based on this correction, the only significant strength of association was for detecting mild hypokalemia (p < 0.001).

Supplementary Figure 3 | Performance ranking of ECG-K + via information from 12 leads and each lead individually for
detecting mild to severe hypo/hyper-kalemia. All analyses were conducted for an academic medical center and a community hospital simultaneously. The y-axis presents the area under of receiver operating characteristic curve (AUC) based on the definitions of cases and controls in the subtitles.

Supplementary Figure 4 | Distributions of patient characteristics in each ECG-K + and Lab-K + group.
Bars represent the mean or proportion where appropriate and corresponding 95% conference intervals, which are adjusted by hospital.

Supplementary Figure 5 | Risk effect analysis of patient characteristics on outcomes of interest.
Univariable and multivariable analyses were conducted by Cox proportional hazard model and logistic regression, respectively. All analyses were adjusted by hospital site, including univariable analysis. Continuous variables are standardized by mean and standard deviation, so the units of each continuous variable were 1 standard deviation. The selected variables in multivariable analyses were based on stepwise process for each outcome.

Supplementary Figure 6 | Patient characteristics in different ECG-K + groups and Lab-K + groups. Bars represent the mean or
proportion where appropriate and corresponding 95% conference intervals, which are adjusted by hospital and Lab-K + via linear or logistic regression (*: p < 0.05; **: p < 0.01; ***: p < 0.001).

Supplementary Figure 7 | Risk contribution analysis of additional consideration of ECG-K + .
The baseline model of combined analysis is adjusted to each hospital site and based on Cox proportional hazard model or logistic regression as appropriate for each outcome. C-index and AUC are used as the performance assessment where appreciate. Model 1 includes significant demographic data (All-cause mortality: gender, Age, SBP, and DBP; Hospitalization: gender, age, BMI, DBP, and smoke; ED revisit in 30 days: gender). Model 2 includes the variables in model 1 and additional significant disease histories (All-cause mortality: HLP; Hospitalization: HLP, STK, and HF; ED revisit in 30 days: DM, CAD, STK, and COPD). Model 3 includes the variables in model 2 and additional significant laboratory tests (All-cause mortality: Hb, HCO3, Blood pH, Na, AST, ALT, Alb, CRP, pBNP, and D-dimer; Hospitalization: WBC, Hb, PLT, HCO3, PH, Na, Cl, tCa, GLU, AST, CK, Alb, CRP, TnI, and D-dimer; ED revisit in 30 days: Hb and Na). Abbreviations: *, p < 0.05; **, p < 0.01; ***, p < 0.001.
Supplementary Figure 8 | Performance comparison before and after COVID-19 pandemic for detecting mild to severe hypo/hyper-kalemia. All analyses were conducted for an academic medical center and a community hospital simultaneously. The yaxis presents the area under of receiver operating characteristic curve (AUC) based on the definitions of cases and controls in the subtitles. We used February 6, 2020 to distinguish the start of the pandemic period as fever screening stations began on that date per the central preventive policy in Taiwan.

Supplementary Figure 9 | Continuous association of ECG-K + and Lab-K + on adverse outcomes in included and excluded patients.
The upper and lower panels presented the analyses of Lab-K + and ECG-K + . The solid line and dashed line are point estimation and corresponding 95% conference interval, respectively. The baseline model of combined analysis is adjusted to each hospital site and based on Cox proportional hazard model or logistic regression as appropriate for each outcome.