Search Results

Loss of Pulse Detection is a software-only mobile medical application that is intended to be used with compatible consumer wrist-worn products to analyze pulse data to identify loss of pulse events and provide audio, visual, and haptic alerts to the user.

If the user remains unresponsive to these alerts, Loss of Pulse Detection will attempt to prompt a call to emergency services through the user's connected compatible hardware, such as a smartphone or smartwatch.

Loss of Pulse Detection is intended for over-the-counter (OTC) use. It is not intended to provide a notification on every loss of pulse event and the absence of an alert is not intended to indicate that no such event has occurred; rather the Loss of Pulse Detection is intended to opportunistically surface an alert of possible loss of pulsatility when sufficient data are available for analysis.

These data are only captured when the user is still. Loss of Pulse Detection is not intended to replace traditional methods of diagnosis, treatment, or monitoring.

Loss of Pulse Detection has not been tested for and is not intended for use in people under 22 years of age. It is also not intended for use in individuals previously diagnosed with a high risk for sudden cardiac death such as those with coronary artery disease, cardiomyopathy and/or unexplained syncope/fainting.

The Loss of Pulse detection SaMD is intended to be used for the detection of loss of pulse using photoplethysmography (PPG) and accelerometer sensors present in a wrist worn consumer wearable device. Upon detection of loss of pulse, when the smartwatch is worn, the feature will prompt haptic and audio alerts and notifications to the user and prompt the user's compatible hardware to call emergency services if the user is unresponsive to the notifications and alerts. This is an opt-in feature and will be off by default.

The Loss of Pulse detection SaMD comprises a software component that resides on the compatible consumer wearable device (from here on referenced as a "smartwatch") and a user facing mobile application that resides on general purpose compatible consumer mobile devices such as a smartphone (from here on referenced as "Smartphone"). The software component on the smartwatch analyzes pulse data collected by photoplethysmography (PPG) and accelerometry sensors from qualified smartwatch, using an algorithm employing digital signal processing (DSP) and features-based machine learning based on a convolutional neural network (CNN) to detect possible loss of pulse events.

The document provides the following information regarding the acceptance criteria and the study that proves the device meets the acceptance criteria for the Loss of Pulse Detection (LPD) software:

1. Table of Acceptance Criteria and Reported Device Performance

While explicit "acceptance criteria" are not presented in a formal table with specific thresholds, the document details the performance metrics achieved in the clinical validation study. The performance is primarily evaluated based on sensitivity and specificity.

2. Sample Sizes and Data Provenance

• Test Set (Clinical Validation Studies):

  • Primary Study: 135 participants for sensitivity evaluation (pulseless and pulsatile data).
  • Secondary Study: 21 participants for sensitivity evaluation (pulseless data).
  • Specificity Evaluation: 131 evaluable participants from the first study.
  • Data Provenance: The document states participants were from a "racially and ethnically diverse population, including adults of diverse sex and age." It does not specify the country of origin, nor explicitly state if it was retrospective or prospective, though clinical validation studies are generally prospective. The phrase "simulated loss of pulse events induced by an arterial occlusion model" suggests a controlled, prospective clinical study.

• Training and Validation Sets (Algorithm Development):

  • Training/Validation/Development Datasets: Over one hundred thousand hours of free-living data (from hundreds of participants who experienced no loss of pulse events) and data from 99 participants who experienced simulated loss of pulse events induced by an arterial occlusion model.
  • Data Provenance: The document states "The training, validation, and test splits used for development included data from diverse participants with varied age, sex, BMI, and skin tone." Again, country of origin is not specified. The data comprised both "free-living data" (suggesting retrospective real-world data) and "simulated loss of pulse events" (suggesting prospective controlled data).

3. Number of Experts and Qualifications for Ground Truth

The document does not explicitly state the number of experts used to establish ground truth or their specific qualifications (e.g., "radiologist with 10 years of experience"). For the clinical validation study where simulated loss of pulse was induced, it implies that the ground truth for "loss of pulse" was established directly through the experimental protocol (arterial occlusion model) rather than by expert review of physiological signals post-hoc. For "no loss of pulse" data, the ground truth is inherently the absence of such an event in free-living data of healthy individuals.

4. Adjudication Method for the Test Set

The document does not specify an adjudication method (e.g., 2+1, 3+1) for the test set. Given that the ground truth for simulated loss of pulse was established by an induced event, and for pulsatile data by its natural occurrence, a separate adjudication method using human experts on the collected data is not described as part of the primary ground truth establishment.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study

An MRMC comparative effectiveness study was not explicitly mentioned or detailed. The study described focuses on the standalone performance of the algorithm and its ability to trigger an emergency call workflow, rather than comparing human reader performance with and without AI assistance. The "Real World Evidence Summary" indicates that human users interact with the device's alerts, but it does not quantify human improvement due to AI assistance in a comparative MRMC framework.

6. Standalone (Algorithm-Only) Performance

Yes, standalone performance was done. The reported sensitivity and specificity values (69.3% and 99.965%) represent the performance of the Loss of Pulse Detection algorithm. The clinical validation study was designed to evaluate the software's ability to detect loss of pulse events.

7. Type of Ground Truth Used

• For Loss of Pulse Events: Ground truth was established through simulated loss of pulse events induced by an arterial occlusion model. This is a physiological, objective method.
• For Absence of Loss of Pulse: Ground truth for pulsatile data was established by the absence of induced events in control participants and "free-living data across hundreds of participants who experienced no loss of pulse events." This relies on the natural state of healthy individuals.

8. Sample Size for the Training Set

The training set was part of a larger dataset used for algorithm development, which consisted of:

• "Over one hundred thousand hours of free-living data across hundreds of participants who experienced no loss of pulse events."
• Data from "99 participants who experienced simulated loss of pulse events induced by an arterial occlusion model."
  The document states this dataset was split at the participant level into training, validation, and held-out test sets. The exact number of participants or hours specifically in the training set is not separately quantified, but it is a substantial portion of the entire development dataset.

9. How Ground Truth for the Training Set was Established

The ground truth for the training set was established in the same manner as for the test set:

• For loss of pulse events, it was based on simulated events induced by an arterial occlusion model.
• For the absence of loss of pulse, it was based on free-living data from participants who did not experience such events, implying a ground truth established by the absence of a known medical condition/event.

Ask a specific question about this device