POC Testing Outperforms AI-ECG for Hypokalemia in Hemodialysis

Key Takeaways

• Among maintenance-hemodialysis patients at a Taiwanese center, point-of-care testing discriminated laboratory-confirmed hypokalemia significantly better than single-lead artificial intelligence–enabled electrocardiogram potassium regression.
• A separate binary ECG classifier had 77.9% sensitivity and 83.2% negative predictive value in the concurrent comparison.
• ECG potassium estimates had larger errors during hypokalemia and tended to overestimate low laboratory values.

Potassium can decline during dialysis, while intermittent blood sampling cannot resolve every shift. Continuously recorded single-lead electrocardiogram (ECG) signals might provide information between blood draws, but how accurately do they reflect low potassium? Blood-based measurements provided the reference for that comparison.

Researchers prospectively enrolled maintenance-hemodialysis patients at one Taiwanese center for their comparison of single-lead artificial intelligence (AI)–enabled ECG and point-of-care (POC) potassium testing. They trained the model in 45 patients and tested it in 29 different patients enrolled later. ECG signals were recorded continuously and compared with central-laboratory serum potassium and matched whole-blood POC measurements.

Central-laboratory potassium below 3.5 mmol/L defined hypokalemia. The head-to-head analysis used 231 simultaneous laboratory–POC–ECG observations, including 88 hypokalemic observations, rather than 88 different patients. Continuous ECG potassium regression supplied discrimination estimates; a separate binary ECG classifier supplied sensitivity and negative predictive value. Patient-level cluster bootstrapping accounted for repeated observations.

In the matched comparison, POC testing achieved an area under the receiver operating characteristic curve (AUC) of 0.996, versus 0.722 for ECG potassium regression, a statistically significant difference. The classifier measures reported in the takeaways came from the separate binary ECG output, not the continuous regression estimates used for the AUC comparison.

During hypokalemia, ECG regression had a mean absolute error (MAE) of 1.000 mmol/L. Its error was smaller outside hypokalemia, while POC measurements agreed more closely with central-laboratory potassium. ECG regression tended to predict higher potassium when laboratory values were low, potentially producing a falsely reassuring estimate at sampled reference points.

Testing involved a small independent cohort at one Taiwanese center, with repeated measurements and exclusions for certain cardiac histories and conduction-altering medications. Test-cohort laboratory potassium was sampled before and after dialysis, without independent mid-session reference measurements; continuously estimated intradialytic trajectories therefore lacked contemporaneous biochemical verification. Severe hypokalemia was too uncommon to assess, and dialysate potassium was not recorded. The authors proposed ECG trend surveillance as an adjunct, not a proven substitute for POC or laboratory confirmation; effects on clinical decisions, outcomes, and safely avoiding blood draws were not tested.

The authors interpreted the single-lead ECG approach as having preliminary screening potential, while its role in noninvasive trend monitoring remains unproven. They called for multicenter evaluation with serial intradialytic blood sampling and assessment of effects on clinical decisions.

Clinician Questions

Which hemodialysis patients were excluded from single-lead AI-ECG hypokalemia testing?

Researchers' prospective Taiwanese evaluation excluded adults receiving maintenance hemodialysis who had a previous myocardial infarction or coronary artery bypass grafting, a pacemaker, atrial fibrillation, or medications substantially altering baseline cardiac conduction. Applicability to patients with those histories or exposures was therefore not established.

What ECG changes contributed to overestimating low potassium during hemodialysis?

For single-lead AI-ECG potassium estimation during maintenance hemodialysis, investigators used T-wave slope and amplitude. One illustrated misclassified hypokalemic trace retained unexpectedly prominent T-wave amplitude despite low laboratory potassium, contributing to an inaccurately high estimate; that example does not establish a mechanism across patients.

This content is intended for healthcare professionals for educational purposes. The article and its image were developed with AI assistance and reviewed by members of the ReachMD team prior to publication. Please refer to the original source linked for the complete and most up-to-date information.