[ad_1]
The application of artificial intelligence to the ECG could help detect hyperkalemia in patients with chronic renal failure (CKD) without having them undergo a blood test, a revealed a study.
The researchers formed an in-depth learning model to predict serum potbadium levels of 5.5 mEq / L or greater (usual threshold for the treatment of hyperkalemia) and validated it compared to three regional datasets of patients with chronic renal failure at stage 3 or higher (n = 61,965).
The model appeared to be performing well with contributions from the ECG I and II diversions (AUC 0.883 for Minnesota, 0.860 for Florida and 0.853 for Arizona); the addition of the V3 and V5 versions has only slightly improved the model's performance, according to Paul Friedman, MD, of the Mayo Clinic of Rochester, Minnesota, and his colleagues, in their study published online in JAMA Cardiology.
A balance of sensitivity and specificity could be achieved, both between 75 and 82%. But when the model worked with a sensitivity of about 90%, its specificity was limited to 63.2% for Minnesota, 54.7% for Florida and 55.0% for Arizona.
"The ability to screen for non-invasive hyperkalemia using ECG data would represent a major advance in the care provided to patients with this life-threatening disease," the authors said. that they had found a negative predictive value of 99.0% to 99.6% for the model of deep learning.
This study appears to be AliveCor's fundamental research in the development of its KardiaK hyperkalemia detection software for two-lead ECGs from its Kardia portable systems.
Friedman's study did not include any data on out-of-hospital ECGs from wearable devices, although the same Mayo Clinic group reported at a conference that the artificial intelligence platform had a 90% accuracy 94% compared to these hospital data.
The model had been formed from more than 1.5 million ECG patients with chronic nephropathy who underwent at least one serum potbadium test within 12 hours before or after the ECG. Investigators excluded ECG with left branch block.
"Between 50% and 70% of patients in validation datasets did not have hyperkalemia predicted by DLM [deep learning model]with less than 1% of all test results being falsely negative; on the other hand, up to 42% of the test results were falsely positive, "noted Friedman's group.
In the end, their technique of detecting hyperkalemia is "virtually useless in the clinic" because "most doctors would consider almost unacceptable" that 90% of the sensitivity for such a test, commented Joel Topf, MD, from St. Claire Nephrology to Detroit.
Taking Minnesota's dataset as an example, he calculated that the positive predictive value only increased to 6.1%, the prevalence of hyperkalemia being 2.6% and a specificity of 63%, he said. MedPage today. "This means that 94% of these positive electrocardiograms will have a normal potbadium level.No one will worry about a screening test that will be a false alarm 19 times out of 20."
"Certainly, the performance of the test would increase as the prevalence of hyperkalemia increases and perhaps in a heart failure clinic, where potbadium is increased by commonly used drugs, such as lymphatic hyperemia. ; ACEi. [ACE inhibitors] "Although the rate of hyperkalemia is four times higher, the positive predictive value is only 10%, still useless."
Friedman's group argued that false negatives "are not necessarily false" because the blood test used as a reference is less "dangerous to health and risk of arrhythmia" than the potbadium derived from it. ECG.
"Another possibility is that there could be errors in blood tests in these patients," they suggested. "ECG-based tests are not sensitive to mechanics, temperature, contamination, or other potential errors related to blood processing."
Nevertheless, the investigators acknowledged that further improvement of the deep learning model was needed to reduce the false-positive rate.
The diagnosis of hyperkalemia remains for the moment a challenge since "the patients are often asymptomatic and that the monitoring of the level of potbadium in the blood according to the recommendations is extremely insufficient", according to the authors.
"The most interesting angle is to ask why they are focused on an intermediate outcome," Topf said. "Patients and cardiologists are not specifically interested in potbadium, but what causes hyperkalemia, which is a fatal cardiac arrhythmia." Perhaps the investigators' technique could be adapted to predict this result. . "
The study was funded by AliveCor and the Mayo Clinic as part of a sponsored research agreement.
The technology was developed in part by the Mayo Clinic, which has invested in licensed patent applications at AliveCor.
Friedman has disclosed a patent on potbadium detection technology.
2019-03-04T00: 00: 00-0500
Last update
Source link
