Biobeat

Artificial intelligence is claimed for the deterioration score, and the company's own published description of that score is a rule set. The in hospital monitoring page and the company's CE mark announcement describe an automated, customizable early warning score system that "incorporates advanced health-AI-based algorithms" analyzing aggregated patient data to identify deterioration and provide predictive analytics. The company led paper that developed the score (Eisenkraft and colleagues, Frontiers in Physiology, 2023; five of seven authors at Biobeat) describes the Multi-Parameter Real-Time Warning Score as expert defined: a panel set cutoff values for nine parameters, each scored 0 to 3 against predefined ranges, with decision rules aggregating them and an alert requiring two consecutive measurements over ten minutes. That is a rule set, not a model estimated from data. The cleared measurement algorithms are described in the FDA summaries as signal processing: pulse wave transit time from the optical sensor calibrated against a cuff, respiration from waveform variation, and a waveform derived stroke volume and cardiac output.

What Biobeat sells is a complete product without any learned component: a disposable chest patch and wrist monitor measuring seven parameters, a cloud platform and a rules based warning score.

AI is claimed, and no learned component can be found in the score or the measurements.

Automated alerts are claimed and the oversight around them is asserted. The platform provides medical staff with real time data and alerts and an automated early warning score to identify deterioration, while the cleared Platform-2 is described as a spot checking and trend device that does not contain any alarms.

No threshold, alert volume, operating characteristic or statement of who governs the score is published, and nothing describes what a clinician should do when the score and the hospital's own early warning process disagree. In a setting whose defining failure is alerts being ignored, a further automated alert stream carries a burden to show its volume.

The sensor is described in detail and the score is only called AI. The clearance summaries describe reflective photoplethysmography with four LEDs at 650 and 880 nm, blood pressure from pulse wave transit time calibrated against a cleared cuff, respiration from cyclic variation in the waveform, thermistor temperature, and a waveform derived algorithm for stroke volume and cardiac output, with stated accuracy ranges.

The early warning score is described only as incorporating advanced health AI based algorithms that analyze aggregated patient data: no model class, inputs, scoring basis, thresholds, version or relation to a published framework such as NEWS. Sensor transparency does not answer what produces the score clinicians act on.

The data path is described and no party on it is named. The clearance summaries describe readings passing from the patch by Bluetooth to a gateway, then to the cloud for display in a web application, with the early warning score analyzing aggregated data on the platform. No hosting provider, subprocessor list or model supplier for the score is named, and the company has described onward integration with at least one care at home partner platform without naming what it receives. For a continuous physiological stream crossing borders, the host and each onward recipient matter for breach scope.

Measurement is validated against invasive reference standards, and the deterioration score has a company led peer reviewed evaluation. The Platform-2 clearance summary reports a study in 90 ambulatory heart failure patients undergoing right heart catheterization (80, 77 and 82 in the analyses), comparing the patch's stroke volume and cardiac output with the Fick method and thermodilution, and states accuracy of 2 percent for oxygen saturation, 3 percent for pulse rate and 5 mmHg for blood pressure after cuff calibration. The warning score was evaluated in Frontiers in Physiology (Eisenkraft and colleagues, 2023; five of seven authors at Biobeat) in 361 hospitalized high risk patients and 160 healthy controls, with readings every five minutes for up to 72 hours: 39 patients deteriorated (ABCDE criteria, resuscitation, intensive care transfer or death), and the score flagged 30 of the 39 against 6 for NEWS, with a mean lead time of about 43 hours; the company's release reports a positive predictive value of 77 percent against 20 percent for NEWS. The study is single program and company led, and no named hospital deployment reports results.

Published research short of independent validation, for both the readings and the score.

A compliance assurance is published and nothing answers what happens to the data. The in hospital monitoring page describes a secure, HIPAA and GDPR compliant cloud based patient management system, and the clearance summaries describe the path: patch to a Bluetooth gateway to the cloud to a web application.

No retention period, de-identification practice or training use statement is published, and the early warning score is described as analyzing aggregated patient data without saying whether that aggregation trains it. The data is dense: a continuous multi parameter physiological stream from patients in hospitals, long term care and homes, processed by an Israeli company selling in the United States, Europe and Latin America, so processing location and transfer are live questions. Safety engineering, such as how a corrupted reading or a disconnected patch is caught, is not published.

Compliance is claimed and no document behind it is published. The in hospital monitoring page describes a secure, HIPAA and GDPR compliant cloud based patient management system; no business associate agreement, contracting entity, permitted use terms or subprocessor list is published on the pages describing the product.

The same monitors are used by hospitals and home programs, which makes the company a business associate of those providers, and the company has also sold a chest monitor to individuals for self set up, a channel with no covered entity where the federal rule does not reach. One device under two legal regimes, with nothing published separating them.

The product pages show no security attestation. The in hospital monitoring page calls the cloud platform secure; the company's MDSAP certification and CE marking are quality and conformity programs, not information security attestations. The Platform-2 clearance (December 2022) predates FDA's premarket cybersecurity documentation requirement and its summary mentions no cybersecurity testing, so clearance says nothing about security here. The deployment transmits continuously: body worn devices streaming through gateways into a cloud platform from hospitals, long term care and homes across several regions.

The measurements are cleared, and the AI early warning score and alerts the product is marketed for sit outside every clearance, with no regulatory position stated for them. FDA cleared the BB-613WP patch (K190792, August 2019), the Biobeat Platform with the patch (K212153, March 2022), Biobeat Platform-2 adding stroke volume and cardiac output (K222010, December 2022) and the BB-613-BPM blood pressure monitor (K241066, January 2025).

The Platform-2 indication is spot checking and tracking changes in adults, not for critical care patients, in hospitals, clinics, long term care and at home, and its summary states that the device does not contain any alarms. The in hospital monitoring page markets a platform that provides medical staff with real time data and alerts and an automated early warning score using AI based algorithms to identify deterioration. The page does not call the score cleared; it says nothing about its regulatory basis.

Nothing published governs or tests a learned model, because none is described, and the measurement layer's exposure to skin pigmentation is not addressed. The deterioration score is a rule set with expert defined thresholds (Eisenkraft and colleagues, Frontiers in Physiology, 2023), evaluated in 361 hospitalized high risk patients and 160 healthy controls against NEWS; the paper reports alert patterns by diagnosis (heart failure, stroke, COPD, kidney disease, arrhythmias) for the rules, which do not show how a model behaves. The Platform-2 clearance summary reports a study of 90 adults with heart failure (80, 77 and 82 in the analyses) comparing the patch against invasive reference methods, with no results by skin tone, sex or age.

Every parameter comes from one reflective optical sensor, and pulse oximetry is documented to overestimate oxygen saturation in patients with darker skin, a bias FDA has examined; here a pigmentation dependent error would reach blood pressure, oxygen saturation, respiration, stroke volume and cardiac output together, and flow into the warning score built on them. No governance framework, subgroup evaluation or statement on how the score's thresholds are reviewed is published.

Staff see the vital signs behind every alert, and nothing stands behind the score. The platform gives medical staff real time data and alerts, so an early warning alert can be checked against the readings it came from before anyone escalates or stands down. No sensitivity, alert burden or lead time is published for the score, and no warranty, indemnity or remedy covers a missed or false deterioration alert.

The measurement error the clearance reports (oxygen saturation within 2 percent, blood pressure within 5 mmHg after cuff calibration, cardiac output agreement within 30 percent in 80 patients) is an agreement bound on the readings, not an error rate for the score clinicians act on.

Record integration is claimed and no system is named. The in hospital monitoring page says patient data syncs directly into the hospital EMR system; no record system, interface standard, FHIR capability or marketplace listing is named.

Because the device generates its own data rather than consuming existing feeds, it needs little integration to work, and correspondingly produces a parallel stream that must be reconciled with what the hospital already charts, including when patch readings and manual observations disagree.

A single cloud deployment with its own sensors, and no location committed. The clearance summaries describe the path: the patch sends data by Bluetooth to a gateway, then to the cloud for display in a web application, so a ward can be monitored without an existing monitoring estate, which lowers the barrier for general wards, long term care and home programs. The company describes the platform as HIPAA and GDPR compliant, which names regimes rather than a location. No hosting provider, region, residency option or tenancy model is published, for continuous telemetry from an Israeli company selling across the United States, Europe and Latin America.

No pricing is published, and the consumable model is discoverable. The chest patch is single use and runs for up to 120 hours per the clearance summary, so cost recurs with every monitored patient and scales with how widely a ward deploys it, alongside the wrist device and the cloud platform; the company also offers ambulatory blood pressure monitoring and home sleep services. No per patch price, device cost, platform subscription or implementation fee is published.

Settings are named broadly and the clearance validates part of them. The company offers a hospital solution for short term monitoring, a telehealth solution for long term monitoring, and at home services including ambulatory blood pressure monitoring and a home sleep lab, and has described use across wards, step down and post surgical units, long term care and hospital at home.

The Platform-2 indication covers adults in hospitals, clinics, long term care and at home for spot checking and trends, and excludes critical care patients, so intensive care and continuous alarm use sit outside what was cleared.