(173 days)
The Eko Analysis Software is intended to provide support to the evaluation of patients ' heart sounds and ECG's. The software analyzes simultaneous ECG and heart sounds. The software will detect the presence of suspected murmurs in the heart sounds. The software also detects the presence of atrial fibrillation and normal sinus rhythm from the ECG signal. In addition, it calculates certain cardiac time intervals such as heart rate, QRS duration and EMAT. The software does not distinguish between different kinds of murmurs and does not identify other arrhythmias.
It is not intended as a sole means of diagnosis. The interpretations of heart sounds and ECG offered by the software are only significant when used in conjunction with physician over-read and is for use on adults (> 18 years).
The Eko Analysis Software is a cloud-based software API that allows a user to upload synchronized ECG and heart sound/phonocardiogram (PCG) data for analysis. The software uses several methods to interpret the acquired signals including signal processing and artificial neural networks. The API can be electronically interfaced, and perform analysis with data transferred from multiple mobile or computer based applications.
The EAS software is only intended to be used in conjunction with data acquired using two previously-cleared physiological data acquisition devices (Eko DUO (K170874) and Eko CORE (K151319)). The software is designed to be used with companion mobile apps that are used during data acquisition. After analysis, results are returned through an interface to the mobile apps for display.
The algorithm consists of the following components:
- Rhythm detection algorithm: A neural network model that uses ECG to detect normal sinus rhythm and atrial fibrillation.
- Murmur detection algorithm: A neural network model that uses heart sounds to detect the presence of murmurs.
- Heart rate analysis algorithm: A signal processing algorithm that uses ECG or heart sounds as appropriate to calculate heart rate. It also provides an alert if the measured heart rate is indicative of Bradycardia or Tachycardia.
- QRS duration algorithm: A signal processing algorithm that measures the width of the QRS pulse on a single-channel ECG.
- EMAT Interval algorithm: A signal processing algorithm that uses Q peak detection and S1 envelope detection to measure the Q-S1 interval, defined as electromechanical activation time or EMAT.
Here's an analysis of the acceptance criteria and study details for the Eko Analysis Software, based on the provided FDA 510(k) summary:
1. Table of Acceptance Criteria and Reported Device Performance
The FDA 510(k) summary does not explicitly state pre-defined "acceptance criteria" in terms of specific thresholds for sensitivity, specificity, or error rates that the device had to meet for clearance. However, it presents the performance results of the device's algorithms, implying that these results were considered acceptable for demonstrating substantial equivalence.
| Feature | Performance Metric(s) | Reported Device Performance | Implicit Acceptance (Interpretation) |
|---|---|---|---|
| Rhythm Detection | Sensitivity (Normal/AFib) | 100% (95% CI: 93.8 - 100.0) | Excellent sensitivity for detected rhythms. |
| Specificity (Normal/AFib) | 96.2% (95% CI: 93.8 - 97.7) | Very good specificity for detected rhythms, showing low false positives among classified ECGs. | |
| % Classifiable ECG Recordings | 74.3% (544/732) | A significant portion of recordings are classifiable, indicating functional utility. | |
| Murmur Detection | Sensitivity | 87.6% (95% CI: 84.2 – 90.5) | Good sensitivity for detecting murmurs. |
| Specificity | 87.8% (95% CI: 85.3 – 89.9) | Good specificity for murmur detection, showing a balance between true positives and true negatives. | |
| Heart Rate Calculation | Heart Rate Error (MIT-BIH dataset) | 1.14% (95% CI: 0.95 - 1.34) | Very low error rate for heart rate calculation. |
| Bradycardia Detection Sensitivity | 94.7% (95% CI: 89.8 - 97.3) | High sensitivity for identifying bradycardia. | |
| Bradycardia Detection Specificity | 99.7% (95% CI: 99.4 - 99.8) | Excellent specificity for identifying bradycardia, suggesting very few false alarms. | |
| Tachycardia Detection Sensitivity | 93.6% (95% CI: 90.9 - 95.6) | High sensitivity for identifying tachycardia. | |
| Tachycardia Detection Specificity | 99.0% (95% CI: 98.7 - 99.3) | Excellent specificity for identifying tachycardia. | |
| QRS Duration Calculation | Absolute Mean Error (ms) | 9.25 (95% CI: 7.93 - 10.58) | The absolute mean error is quantified, providing a measure of accuracy. The acceptability implicitly relies on clinical relevance. |
| EMAT Calculation | Absolute Error (Physionet 2016 dataset) | 1.68% (95% CI: 1.06 - 2.30) | The absolute error is quantified, providing a measure of accuracy. The acceptability implicitly relies on clinical relevance. |
2. Sample Size Used for the Test Set and Data Provenance
-
Rhythm Detection Test Set:
- Proprietary EKO ECG dataset: 732 ECG recordings from 139 patients.
- Provenance: Retrospective. Collected using Eko DUO (732 recordings from 139 patients) and Eko CORE (1445 recordings from 236 patients) devices. Geographic origin not explicitly stated, but proprietary datasets from "individual volunteers" suggest it could be a local or multi-center collection by Eko Devices Inc.
- Publicly available databases: MIT-BIH Arrhythmia Database, MIT-BIH Arrhythmia Noise Stress Database, AHA Database, NST Database, Physionet QT Database, PhysioNet 2016 Database.
- Provenance: Retrospective. International, well-established public reference datasets.
- Proprietary EKO ECG dataset: 732 ECG recordings from 139 patients.
-
Murmur Detection Test Set:
- Eko Heart Sound Database: Data collected using both Eko CORE and Eko DUO devices.
- Provenance: Retrospective. Similar to the EKO ECG dataset, proprietary data collected from "individual volunteers." The combined total number of patients/recordings from both devices is 139 + 236 = 375 patients and 732 + 1445 = 2177 recordings for the proprietary dataset. It's unclear if the "Eko Heart Sound Database" is precisely the same as the "EKO ECG dataset" or a subset/superset, but the description points to the same underlying proprietary data collection.
- Eko Heart Sound Database: Data collected using both Eko CORE and Eko DUO devices.
-
Heart Rate Calculation Test Set:
- Publicly available datasets: Same as Rhythm Detection (MIT-BIH, etc.).
- Proprietary EKO ECG dataset: Same as Rhythm Detection (732 ECG recordings from 139 patients).
-
QRS Duration Calculation Test Set:
- Publicly available PhysioNet QT database.
-
EMAT Calculation Test Set:
- Publicly available Physionet 2016 database.
- Proprietary Eko ECG dataset: Same as Rhythm Detection (732 ECG recordings from 139 patients).
3. Number of Experts Used to Establish the Ground Truth for the Test Set and Qualifications
The document does not explicitly state the number of experts or their specific qualifications (e.g., "radiologist with 10 years of experience") used to establish the ground truth for the test sets.
- For publicly available datasets (like MIT-BIH, PhysioNet), the ground truth is typically established by multiple cardiologists or electrophysiologists based on extensive review and annotation, often with published consensus guidelines. The qualifications of these annotators are generally high, representing expert cardiac clinicians/researchers.
- For the proprietary Eko datasets, the document does not specify how the ground truth was established, who established it, or their qualifications. It mentions "retrospective analysis," which usually implies that an expert (or panel of experts) reviewed the recordings and clinical data to determine the presence of conditions (e.g., AFib, murmur) for ground truth labeling.
4. Adjudication Method for the Test Set
The document does not describe a specific adjudication method (e.g., 2+1, 3+1) for establishing the ground truth for any of the test sets, either for public or proprietary data. For public databases, consensus annotations are the typical method. For proprietary data, this information is not provided.
5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study
No mention of a Multi-Reader Multi-Case (MRMC) comparative effectiveness study comparing human readers with and without AI assistance is present in the document. The study focuses purely on the standalone performance of the algorithms.
6. Standalone (Algorithm Only) Performance Study
Yes, the document exclusively describes standalone performance studies. It reports the performance metrics (sensitivity, specificity, error rates) of the Eko Analysis Software's algorithms directly on the test datasets, independent of human interaction or a human-in-the-loop workflow.
7. Type of Ground Truth Used
The ground truth used appears to be expert consensus or expert-annotated data.
- For publicly available databases, ground truth is typically derived from expert annotations and established clinical criteria.
- For proprietary datasets, the nature of the metrics (e.g., sensitivity/specificity for rhythm and murmur detection) strongly implies that a human expert (or panel) reviewed the ECG and heart sound recordings to classify them as having or not having AFib, normal sinus rhythm, or a murmur, which served as the reference standard.
8. Sample Size for the Training Set
The document does not provide a specific sample size for the training set. It mentions that the algorithms use "artificial neural networks" and that testing was carried out using "retrospective analysis on a combination of publicly available (...) and proprietary datasets." While these datasets were used for validation, the size and composition of the training datasets are not described.
9. How the Ground Truth for the Training Set Was Established
The document does not describe how the ground truth for the training set was established. Given that neural networks were used, the training data would also have required labeled ground truth. It is reasonable to infer that if external/public datasets were used for validation, they likely also formed part of or informed the training process, and proprietary data collected by Eko would also have been used for training, with ground truth established similarly to the validation data (i.e., expert review/consensus, though not explicitly stated for training).
§ 870.2300 Cardiac monitor (including cardiotachometer and rate alarm).
(a)
Identification. A cardiac monitor (including cardiotachometer and rate alarm) is a device used to measure the heart rate from an analog signal produced by an electrocardiograph, vectorcardiograph, or blood pressure monitor. This device may sound an alarm when the heart rate falls outside preset upper and lower limits.(b)
Classification. Class II (performance standards).