Why is high potassium dangerous when you have kidney disease?
Potassium is a mineral your heart needs to beat steadily. Your kidneys get rid of any extra. In chronic kidney disease (CKD), a long-term loss of kidney function, potassium can build up in the blood. Doctors call this hyperkalemia. Some medicines that protect the kidneys and heart, such as ACE inhibitors, also raise the risk.
Hyperkalemia often causes no symptoms, yet it can trigger dangerous heart rhythm problems. Today, the only way to check is a blood test. According to the authors, this monitoring is badly underused: up to one third of patients on these medicines get no potassium checks at all. And when doctors read an ECG by eye, they spot hyperkalemia in only 34% to 43% of cases, based on the estimates they cite.
How did the researchers train the AI?
The Mayo Clinic team used a deep neural network, a type of AI that teaches itself to find patterns in complex data. They trained it on 1,576,581 ECGs from 449,380 patients recorded between 1994 and 2017, each paired with a blood test taken within 12 hours. Hyperkalemia was defined as a potassium level of at least 5.5 mEq/L, a common cutoff for starting treatment.
The AI reads only 2 ECG leads, or 4 in a second version. A lead is one viewing angle on the heart's electrical activity. Leads I and II can be recorded simply by touching electrodes with the hands and left leg. The AI was then tested in 61,965 patients with stage 3 or more advanced CKD at three Mayo Clinic sites: Minnesota, Florida, and Arizona.
What were the results?
Results are reported as area under the curve, a score from 0.5 (pure chance) to 1 (a perfect test). Using two leads, the AI scored 0.883 in Minnesota, 0.860 in Florida, and 0.853 in Arizona.
The researchers focused on a setting designed for screening: missing as few cases as possible. In that mode, the AI caught about 9 in 10 cases of hyperkalemia (a sensitivity of 88.9% to 91.3%). The trade-off was many false alarms. Its specificity, meaning how well it correctly clears people who don't have high potassium, was only 54.7% to 63.2%. Up to 42% of all results were false positives, while fewer than 1% were missed cases.
Key figures
- 0.853 to 0.883area under the curve using only 2 ECG leads
- 88.9% to 91.3%of hyperkalemia cases caught in screening mode (sensitivity)
- 42%at most, of all results were false alarms
Keeping track of your potassium checks with Mima
This AI is still research: today, only a blood test measures your potassium. But if you have kidney disease or take a medicine that can raise potassium, the study is a reminder of why regular checks matter.
Mima helps you keep track of your blood tests, prescriptions, and appointments, and gets your questions ready for your next visit, such as: "How often should I have my potassium checked on this medicine?"
What could this mean for patients?
The authors don't want to do away with blood tests. They see the AI as a first filter for patients who need their potassium watched. A "no hyperkalemia" result could reassure both patient and doctor and spare a blood draw. A "possible hyperkalemia" result would trigger a confirming blood test. Because the two leads can be recorded with the hands, the authors see a possible future home test.
For now, this is only a possibility: the study does not measure any health benefit for patients.
What are the limits of this study?
The data were retrospective, meaning they came from existing records, and the team had no information on patients' medicines or medical history. All ECGs came from standard 12-lead machines in clinics or hospitals. It is not yet known whether the AI works as well with a small portable device, which records a noisier signal, or at home.
Frequent false alarms could cause needless worry. A falsely reassuring result could delay treatment, which the authors see as the most serious concern. They call for prospective studies, which follow patients forward in time, to check whether the tool truly improves detection, care, and health outcomes.
Frequently asked questions
What is hyperkalemia?
Hyperkalemia means too much potassium in the blood. In this study, the cutoff was 5.5 mEq/L. It is common in people with kidney disease and can cause dangerous heart rhythm problems.
Can an ECG detect high potassium?
High potassium can change the shape of the ECG, but doctors reading by eye often miss it. In this study, an AI caught about 9 in 10 cases from two ECG leads, with many false alarms.
Can an ECG replace a potassium blood test?
No. The AI in this study is meant as a first filter. A result suggesting hyperkalemia still needs a blood test to confirm it.
Who needs regular potassium checks?
People with chronic kidney disease, especially those taking medicines such as ACE inhibitors or angiotensin receptor blockers, which can raise potassium. Your care team decides how often to test.
The original study
Galloway CD, Valys AV, Shreibati JB, et al. “Development and Validation of a Deep-Learning Model to Screen for Hyperkalemia From the Electrocardiogram” JAMA Cardiology, May 1, 2019. DOI : 10.1001/jamacardio.2019.0640
Read the full study in JAMA CardiologyThis article summarizes a scientific study and is not medical advice.