(249 days)
The CSF-3 is intended to record, store, transfer, and display single-channel electrocardiogram (ECG) rhythms. The ECG signal is for quality checks of the data and for manual interpretation of heart rate. The CSF-3 is also indicated for use in measuring and displaying functional oxygen saturation of arterial hemoglobin (SpO2). The CSF-3 is for adult patients and health-conscious individuals in hospitals, clinics, long-term care facilities, and homes. The CSF-3 is a prescription device and should be used under the care of a physician. CSF-3 does not provide any alarms. It is not intended for pediativ use or use in critical care settings.
The CSF-3 is a non-invasive system consisting of software, hardware and mechanical components that enables the user to measure electrocardiography (ECG) and oxygen saturation of arterial hemoglobin (SpO2). The CSF-3 consists of 3 main components: 1) CS Watch 3 with CSF-3 Watch firmware ("Watch"): The Watch is a wrist worn device embedded with non-invasive sensors. The Watch includes firmware that activates the sensors, synchronizes the data sampled by the sensors, processes the data, stores the processed data in non-volatile memory, and provides the data to the user. The processed data is transferred to the Mobile App via a secured BLE communication channel. In addition, the watch sends realtime raw data signals to the Mobile App. 2) CSF-3 Mobile Application ("Mobile App"): The Mobile App works on both Android OS and iOS. The Mobile App communicates with the Watch via BLE and to the Cloud App via HTTPS, thus acting as the Watch gateway to the Cloud App. The Mobile App caches the processed data from the Watch and transfers it to the Cloud App. It allows the user to conveniently view the measurement results and real time raw data. The Mobile App provides the user with the capability of creating an on-demand report and sharing it using 3rd party sharing applications. 3) CSF-3 Cloud Application ("Cloud App"): The Cloud App securely stores the user and processed data over designated databases. It provides the mechanism of creating and sending periodical reports which are sent to the user's email both automatically and on-demand.
The provided text pertains to a 510(k) premarket notification for a medical device called CSF-3, for which CardiacSense is the manufacturer. The document details the device's intended use, technological characteristics, and various testing performed to demonstrate its substantial equivalence to a predicate device (Withings Scan Monitor).
Here's an analysis of the acceptance criteria and the studies that prove the device meets them, based on the provided text:
Acceptance Criteria and Reported Device Performance
The document describes performance criteria primarily for ECG (QRS detection and Heart Rate accuracy) and SpO2 measurement accuracy.
| Acceptance Criteria Category | Specific Metric | Acceptance Criteria (Implicit) | Reported Device Performance |
|---|---|---|---|
| QRS Detection (Bench) | Sensitivity (MIT-BIH Arrhythmia) | > 98% | > 98% |
| PPV (MIT-BIH Arrhythmia) | > 98% | > 98% | |
| Sensitivity (AHA) | > 98% | > 98% | |
| PPV (AHA) | > 98% | > 98% | |
| Sensitivity (MIT-BIH Noise Stress) | > 93% | > 93% | |
| PPV (MIT-BIH Noise Stress) | > 93% | > 93% | |
| Heart Rate Accuracy (Bench) | HR RMS accuracy (MIT-BIH Arrhythmia) | 1-2% | 1-2% (varies) |
| HR RMS accuracy (AHA) | 1-2% | 1-2% (varies) | |
| HR RMS accuracy (MIT-BIH Noise Stress) | Slightly > 3% | Slightly > 3% | |
| Heart Rate Accuracy (Clinical) | Sensitivity | Not explicitly stated (implied high) | 99.6% |
| False Detection Rate | Not explicitly stated (implied low) | 0.54% | |
| ARMS (Average Root Mean Square) | Not explicitly stated (implied low) | 1.54 BPM | |
| SpO2 Accuracy (Clinical) | Accuracy (range 70% to 100%) | Not explicitly stated (implied within acceptable clinical limits, likely +/- 3-5%) | 2.96% |
Note on "Acceptance Criteria (Implicit)": The document states that the performance "is above XY%" or "varies between AB%," implying that these are the levels deemed acceptable for demonstrating substantial equivalence. The exact numeric acceptance criteria are not explicitly defined as "must be at least X" but are presented as the achieved performance which is satisfactory.
Study Details
The document describes both non-clinical (bench) and clinical testing.
-
Sample Size and Data Provenance:
- Bench Testing (QRS and HR accuracy): The performance of the QRS algorithm was evaluated against three databases:
- MIT-BIH Arrhythmia database
- AHA database
- MIT-BIH Noise Stress database
The number of recordings/samples within these databases is not specified. The provenance of these databases (e.g., country of origin, retrospective/prospective) is also not explicitly mentioned, but these are standard, publicly available, and widely accepted benchmark databases for ECG algorithm testing.
- Clinical Study (ECG HR performance): "The study included 52 subjects and a total of 23,579 samples resulted with sensitivity of 99.6% and false detection rate of 0.54% and ARMS of 1.54 BPM."
- Sample Size: 52 subjects, 23,579 samples.
- Data Provenance: Not explicitly stated (e.g., country, retrospective/prospective), but implied to be prospective clinical data collected for this study, as it involved "comparing the CSF-3 to a Holter."
- Clinical Study (SpO2 Accuracy): "The clinical study with n=234 samples, the SpO2 range was validated to be from 70% to 100% with accuracy of 2.96%."
- Sample Size: 234 samples.
- Data Provenance: Conducted in the Hypoxia Research Laboratory, Department of Anesthesia Perioperative Care, University of California at San Francisco (UCSF) in compliance with the ISO 80601-2-61:2017 standard. This implies a prospective, controlled clinical study.
- Bench Testing (QRS and HR accuracy): The performance of the QRS algorithm was evaluated against three databases:
-
Number of Experts and Qualifications:
- For Bench Testing (QRS and HR): Not applicable for the algorithmic evaluation against standard databases. The ground truth in these databases is established through expert annotation by the creators of these databases, but not by experts specifically for this submission.
- For Clinical Studies (ECG HR and SpO2): The document does not specify the number or qualifications of experts (e.g., cardiologists, anesthesiologists) involved in reviewing or establishing ground truth for the clinical studies. It mentions comparison to a "Holter" for ECG and "A-line as a reference" for SpO2, which are established clinical measurement standards. The implication is that the reference measurements serve as the ground truth, not human expert interpretation of the CSF-3 output.
-
Adjudication Method for Test Set:
- Not specified. Given that the ground truth for clinical studies relies on established reference devices (Holter, A-line), a separate human adjudication method for the device's output against a human-reviewed ground truth is not explicitly described. For the bench tests, the ground truth is pre-established within the databases.
-
Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study:
- No MRMC comparative effectiveness study is described for how human readers improve with AI vs. without AI assistance. The device is for recording and displaying ECG and SpO2 data for manual interpretation or quality checks, not for AI-assisted image interpretation or diagnosis. The ECG signal is stated to be "for quality checks of the data and for manual interpretation of heart rate," implying human readers will interpret the output from the device.
-
Standalone (Algorithm Only) Performance:
- Yes, for QRS detection and HR accuracy on the bench. The document states: "The performance of the QRS algorithm was evaluated against three databases following the requirements stated in the IEC-60601-2-47 standard." This describes the algorithm's standalone performance.
- Not explicitly for SpO2 beyond the stated accuracy. The SpO2 accuracy (2.96%) is a measure of the device's (including its algorithm's) performance against a reference standard.
-
Type of Ground Truth Used:
- Bench Testing (ECG): Established, publicly available, and expertly annotated databases (MIT-BIH Arrhythmia, AHA, MIT-BIH Noise Stress). These database annotations serve as the ground truth.
- Clinical Study (ECG HR): Comparison to a "Holter" device. The Holter recording and its interpretation serve as the ground truth.
- Clinical Study (SpO2): Comparison to "the A-line as a reference" and performance "in compliance with the ISO 80601-2-61:2017 standard." This implies an arterial blood gas analysis or a similar precise clinical measurement as the ground truth reference.
-
Sample Size for Training Set:
- The document does not provide a sample size for a training set. This suggests that the QRS detection and HR algorithms were either developed using proprietary datasets (not explicitly detailed in the document) or were designed using general signal processing principles rather than being deep learning models requiring large, labelled training datasets as typically described for AI/ML devices. For a 510(k), particularly for devices like ECG monitors, comprehensive training data details aren't always required if the algorithms are well-established.
-
How Ground Truth for Training Set Was Established:
- As no training set details are provided, the method for establishing its ground truth is also not mentioned.
§ 870.2340 Electrocardiograph.
(a)
Identification. An electrocardiograph is a device used to process the electrical signal transmitted through two or more electrocardiograph electrodes and to produce a visual display of the electrical signal produced by the heart.(b)
Classification. Class II (performance standards).