A Study to Detect Hyperkalemia Using Smartphone-enabled Electrocardiogram (EKG)

Understanding model's ability to predict hyperkalemia as determined by the area under the receiver operating characteristic

Detailed performance metrics of the algorithm (sensitivity, specificity, positive predictive value and negative predictive value) will be calculated using an optimized cutoff threshold determined from the primary outcome.

Following the detection of hyperkalemia by AI enhanced ECG time to initial hyperkalemia diagnosis (in minutes) by laboratory analysis following ambulatory emergency department presentation will be assessed.

Following outcome measure 3 for patients determined to have hyperkalemia, time to first treatment intervention of hyperkalemia (in minutes) will be assessed since presentation to the emergency department.

Patients who underwent AI enhanced screening for hyperkaliemia, and have a diagnosis of hyperkalemia by laboratory confirmation, will also be assessed for total time spent in the emergency department in hours.

Patients who underwent AI enhanced screening for hyperkaliemia, and have a diagnosis of hyperkalemia by laboratory confirmation, will have the frequency of hospital admission assessed.

Patients who underwent AI enhanced screening for hyperkaliemia, and have a diagnosis of hyperkalemia by laboratory confirmation, will have evaluation of survival at one year.

Patients who underwent AI enhanced screening for hyperkaliemia, and have a diagnosis of hyperkalemia by laboratory confirmation, will have evaluation of frequency of adverse events related to treatment of hyperkalemia (cardiac arrest, hypoglycemia, complications related to dialysis etc).

A previously developed AI algorithm to predict potential underlying cardiac pathology assess from 12 lead ECG via convolutional neural network will be adapted applied to the ECGs for patients who undergo screening for hyperkaliemia in the ED with 6L Kardia ECG device. This neural network uses PQRST complexes to yield a probability of heart failure which may not be readily apparent via manual review. Each recorded 6L Kardia ECG will undergo evaluation by this neural network and will produce a probability of heart failure (0-100%) for each individual patient.

A previously developed AI algorithm to predict potential underlying cardiac pathology assess from 12 lead ECG via convolutional neural network will be adapted applied to the ECGs for patients who undergo screening for hyperkaliemia in the ED with 6L Kardia ECG device. This neural network uses PQRST complexes to yield a probability of silent/paroxysmal atrial fibrillation which may not be readily apparent via manual review. Each recorded 6L Kardia ECG will undergo evaluation by this neural network and will produce a probability of silent/paroxysmal atrial fibrillation (0-100%) for each individual patient.

A previously developed AI algorithm to predict potential underlying cardiac pathology assess from 12 lead ECG via convolutional neural network will be adapted applied to the ECGs for patients who undergo screening for hyperkaliemia in the ED with 6L Kardia ECG device. This neural network uses PQRST complexes to yield a probability of aortic stenosis which may not be readily apparent via manual review. Each recorded 6L Kardia ECG will undergo evaluation by this neural network and will produce a probability of aortic stenosis (0-100%) for each individual patient.

A previously developed AI algorithm to predict potential underlying cardiac pathology assess from 12 lead ECG via convolutional neural network will be adapted applied to the ECGs for patients who undergo screening for hyperkaliemia in the ED with 6L Kardia ECG device. This neural network uses PQRST complexes to yield a probability of amyloidosis which may not be readily apparent via manual review. Each recorded 6L Kardia ECG will undergo evaluation by this neural network and will produce a probability of amyloidosis (0-100%) for each individual patient.

A previously developed AI algorithm to predict patient age from 12 lead ECG via convolutional neural network will be adapted applied to the ECGs for patients who undergo screening for hyperkaliemia in the ED with 6L Kardia ECG device. This neural network uses PQRST complexes to yield an ECG-predicted age. Each recorded 6L Kardia ECG will undergo evaluation by this neural network and determine "ECG age" for each individual patient.