Company Spotlight: SafeBeat Rx App - AI for QT Intervals and Drug Dosing

Maya: Imagine an AI application designed to analyze an electrocardiogram, measure the precise intervals of the heart's electrical cycle, and then suggest exactly how to dose complex antiarrhythmic drugs based on the manufacturer label.

Sam: Today, we are looking at exactly that. This is the FDA 510(k) summary for a product called the SafeBeat Rx App, published by the U.S. Food and Drug Administration on February 6, 2026.

Maya: Before we begin, a quick reminder that our voices are AI-generated, and this episode is a summary of a publicly available FDA document, not company marketing material.

Sam: Right, this is part of our Company Spotlight series, and we are sticking purely to the facts in the 510(k) summary, specifically submission K251218. Let us start with the basics. Maya, what exactly is the SafeBeat Rx App?

Maya: It is classified as a Software as a Medical Device, or SaMD. It does not actually acquire the ECG from the patient directly. Instead, it takes raw digital ECG data from other acquisition and storage devices, analyzes it, and provides provisional ECG interval measurements.

Sam: So it is essentially a layer of software that sits on top of existing hardware in a hospital or clinic. How does the data actually get into the app if it is not touching the patient?

Maya: According to the summary, the data comes in either programmatically through an application programming interface, which is an API, or manually through a secure web interface upload. The document is very specific that it is only intended for traditional wet electrode inputs, and it strictly does not allow the analysis of ECG signals imported by images.

Sam: That makes sense, they want the raw digital signal, not a scanned picture of an ECG printout. What exactly is this machine learning algorithm measuring once it gets that digital signal?

Maya: It performs what the FDA calls provisional ECG analysis. This includes beat-level feature identification, like finding the R-peak, the QRS onset, the ST onset, and the T-wave offset. From there, it estimates several intervals. It measures heart rate, RR interval variability, QRS duration, QT interval, QT interval variability, and heart rate corrected QT, like the QTcF.

Sam: And the document says it is for adult patient populations, meaning 18 years and older. But one of the most interesting parts of this summary is what it does with those measurements. The document mentions optional non-device functions.

Maya: Yes, this is a key feature. The summary states that it provides suggested antiarrhythmic drug dosing consistent with the manufacturer drug label. It specifically lists five drugs: amiodarone, dofetilide, flecainide, sotalol, and IV sotalol.

Sam: That is a fascinating workflow. The software calculates the QT interval and then offers dosing suggestions for those specific medications. But the summary is also very clear about what this software is not. It says the SafeBeat App does not provide continuous cardiac monitoring.

Maya: Exactly. It explicitly states that it does not provide real-time ECG display, continuous monitoring, or alarm functions. Furthermore, it says it does not provide rhythm interpretation or diagnose cardiac arrhythmias, such as atrial fibrillation. It does not include automated rhythm analysis.

Sam: So if a hospital is buying this, they need to know it is not going to alert a nurse if a patient goes into a dangerous rhythm. It is purely an advisory tool for reviewing provisional ECG data. In fact, it says the analysis is not intended to be the sole means of diagnosis and must be reviewed by qualified healthcare personnel.

Maya: That is right. The physician is expected to view, edit, approve, or reject the analysis via the user interface. It also notes that the app is not validated for use in lead I alone.

Sam: I want to dig into the environments where this can be used, because there is a very specific warning in the indications for use about home monitoring versus professional environments.

Maya: Yes, the document states it can be used in a professional healthcare environment such as a hospital, clinic, or similarly equipped facility. However, it also mentions an optional long term monitoring workflow intended for monitoring and evaluating a patient's home acquired ECG.

Sam: Right, but then there is a critical sentence right after that. It says, the software workflow that is intended for use in the professional healthcare environment should not be used in the home environment to adjust QT prolonging medications as is contraindicated for applicable drugs.

Maya: That is a very important limitation. The document does not explain the exact clinical reasons behind that contraindication, it simply states that adjusting those QT prolonging medications at home using the professional workflow is contraindicated for the applicable drugs.

Sam: So a health system deploying this needs to have strict protocols about which workflows are being used where, especially if they are trying to manage patients outside the hospital. Let us talk about how this device actually got cleared. This is a 510(k) clearance, which means it had to prove it was substantially equivalent to an already marketed device.

Maya: Correct. The primary predicate device is the CardioLogs ECG Analysis Platform, under clearance K170568. The summary notes that SafeBeat Rx App has the same overall technological characteristics as CardioLogs, including providing measurements and analysis of ECGs for physician review.

Sam: But CardioLogs does not use machine learning for QT interval measurement, right? The summary says because the SafeBeat app uses a machine learning-based approach for that specific task, they had to bring in a second device for comparison.

Maya: Yes, they included a reference device, which is the AliveCor QT Service, under clearance K212662. The summary states that AliveCor similarly utilizes a machine learning-based algorithm to compute the QT interval.

Sam: The FDA summary explains that using this reference device provides an appropriate technological benchmark. It demonstrates that using machine learning for QT interval calculation is well established in FDA-cleared devices and does not introduce new questions of safety or effectiveness.

Maya: It is a great example of how software developers build their regulatory strategy. You use the primary predicate for the overall workflow and indication, and a reference device to validate a specific new technological feature, like a machine learning algorithm for a specific measurement.

Sam: Let us get into the actual clinical validation. How did they prove this machine learning algorithm actually works? What does the data look like?

Maya: The clinical testing relied on a retrospective approach using publicly available clinical datasets. They assembled data from multiple independent sources, and compared the software-generated measurements for QT, QRS, and heart rate against annotations performed by board-certified cardiologists.

Sam: And the document gives us a very detailed breakdown of the demographics of that testing cohort. The mean age of the patients was 40.9 years, with a range from 18 to 87.

Maya: For sex, it was balanced, with 51.5% male representation. But the race and ethnicity breakdown is where we see some significant gaps in the data.

Sam: Yes, the document states the cohort was 60.9% White, 10% Black or African American, 4.8% Hispanic or Latino, 4.1% Other, and 3.8% Asian. But a large chunk of the data, exactly 16.5%, was classified as Unknown individuals.

Maya: We see a similar issue with the body mass index, or BMI, data. The summary categorizes patients as normal, which is less than 25, at 18.9%. Overweight, which is greater than or equal to 25 and less than 30, was 22.3%. Obese, meaning greater than or equal to 30, was 22.6%.

Sam: And the unknown category for BMI is massive. The document states that 36.2% of the patients had an unknown BMI.

Maya: This is a reality of using retrospective, publicly available datasets. You often cannot go back and fill in missing demographic variables. While the summary states that the validation dataset encompassed healthy individuals, patients in critical care settings, and patients with known arrhythmias, anyone evaluating this AI should be aware of those missing demographic fields.

Sam: The testing did ensure the algorithm saw a wide variety of complex heart signals. The summary notes it included clinically relevant T-wave, U-wave, and T-U wave fusion morphology, as well as drug-induced QTc prolongation.

Maya: They also ran the software through standard international testing databases. For QTc and QRS measurements, they used the Common Standards for Electrocardiography, or CSE, dataset. The document specifies this test used exactly 100 ECGs to demonstrate agreement with manual reference measurements.

Sam: And for arrhythmia detection, meaning heart rate and beat-segment detection, they used three established databases referenced in the standard IEC 60601-2-47. Those were the MIT-BIH Normal Sinus Rhythm dataset, the AHA database, and the MIT-BIH Noise Stress Test dataset.

Maya: According to the FDA summary, the testing on these databases demonstrated effective beat-segment and QRS detection, as well as heart rate and R-R interval estimation.

Sam: So, looking at the big picture, this 510(k) summary tells us about an app that can ingest digital ECG data from an existing hospital system, use machine learning to calculate intervals like the QT, and then provide dosing suggestions for specific antiarrhythmic drugs based on the label.

Maya: But it requires a qualified healthcare provider to review the data, it is not a real-time monitor, it does not diagnose arrhythmias, and it comes with clear warnings about not using the professional workflow in the home environment to adjust QT prolonging medications.

Sam: It is a fascinating look at how machine learning is being applied to very specific, highly regulated tasks in cardiology. The combination of interval measurement and drug dosing suggestions is a unique workflow, even if it is entirely provisional and reliant on physician oversight.

Maya: That wraps up our look at the SafeBeat Rx App. As always, a link to the full FDA 510(k) summary is in our show notes. Please read the source document for yourself, and remember, this podcast is for informational purposes only and is not medical advice.