Acumen Assisted Fluid Management (AFM) Software Feature

• K Number: DEN190029
• Device Name: Acumen Assisted Fluid Management (AFM) Software Feature
• Manufacturer: Edwards Lifesciences
• Date Cleared: 2020-11-13 ( 529 days)
• Product Code: QMS , OMS
• Regulation Number: 870.5600
• Type: Direct
• Panel: CV
• Reference & Predicate Devices: K160552 , K152980

Intended Use

The Edwards Lifesciences Acumen Assisted Fluid Management (AFM) software feature provides the clinician with physiological insight into a patient's estimated response to fluid therapy and the associated hemodynamics. The Acumen AFM software feature is intended for use in surgical patients ≥18 years of age, that require advanced hemodynamic monitoring. The Acumen AFM software feature offers suggestions regarding the patient's physiological condition and estimated response to fluid therapy. Acumen AFM fluid administration suggestions are offered to the clinician; the decision to administer a fluid bolus is made by the clinician, based upon review of the patient's hemodynamics. No therapeutic decisions should be made based solely on the Assisted Fluid Management suggestions.

Device Description

The Acumen™ Assisted Fluid Management (AFM) Software Feature ("the device") consists of software running on the Edwards Lifesciences EV1000 Clinical Platform (K160552 cleared on June 1, 2016) coupled with an Acumen 10 sensor (which was called FloTrac IO sensor in K152980 cleared on January 19, 2016) connected to a radial arterial catheter. The goal of AFM is to reduce the barriers slowing the utilization of perioperative goal directed therapy (PGDT) during surgical procedures by easing the implementation of PGDT, recognizing patterns of fluid responsiveness (i.e. hemodynamic data and past responses to fluid), and suggesting when fluid administration may improve the patient's hemodynamic state. The clinician is responsible for reviewing the AFM software suggestion in addition to a patient's current hemodynamic state and, if the clinician agrees, the clinician can deliver fluid in the standard-of-care fashion. Alternatively, if the clinician disagrees with the fluid suggestion, it can be rejected as the clinician chooses to not deliver any fluid.

The AFM algorithm can be used on the EV1000 Clinical Platform to help maintain patient fluid balance throughout a surgery. The AFM algorithm continuously estimates patient fluid responsiveness (percent increase in Stroke Volume, A SV%) using current hemodynamic parameters and past responses to fluid boluses. The Acumen AFM software feature is intended to simplify the implementation of fluid management protocols/perioperative goal directed therapy (PGDT).

When an Acumen IO sensor is connected and the AFM algorithm is initialized. the EV1000 Clinical Platform will provide notifications to the user when fluid is recommended by the AFM algorithm. The AFM algorithm learns from the stroke volume response to each fluid bolus to determine if a patient is in a fluid responsive or pre-load dependent state. The patient's tidal volume must be ≥ 8 mL/kg while using the AFM software feature. Throughout the case. the algorithm tracks and records bolus and patient response information to adapt its suggestions based off of the individual patient. In order for the algorithm to analyze each fluid bolus, the start and stop time of each infusion must be entered in the system, as well as the volume of the fluid bolus. The algorithm uses data from the current patient in order to predict their fluid responsiveness; this data is not used by the algorithm to determine fluid responsiveness in future patients.

Each bolus can be administered with the fluid, rate, and volume at the discretion of the clinician. Additionally, any fluid bolus can be declined or discarded as deemed appropriate by the clinician. The AFM algorithm will analyze fluid boluses within the following range: Volume: 100 - 500 mL: Rate: 1 - 10 L / hr.

AI/ML Overview

Here's a breakdown of the acceptance criteria and the study proving the device meets them, based on the provided text:

Acceptance Criteria and Device Performance

The primary effectiveness endpoint for the Acumen AFM feature was its ability to predict a patient's fluid responsiveness. The acceptance criterion was based on exceeding a historical performance criterion of 30% fluid responsiveness, derived from the OPTIMISE study.

Table 1: Acceptance Criteria and Reported Device Performance

Criterion/Metric	Acceptance Criterion (Historical Control from OPTIMISE study)	Reported Device Performance (AFM IDE Study)	Notes
Primary Effectiveness Endpoint:
Percentage of time an AFM recommendation (followed by a clinician-accepted and delivered bolus) resulted in an increase in stroke volume meeting the selected fluid strategy.	≥ 30%	66.1% [62.1%, 69.7%] (for AFM Recommendations)	This statistically superior performance against the 30% historical target was based on instances where clinicians followed AFM recommendations. If every declined AFM recommendation was considered a negative response, the rate could be as low as 37%, as fluid was not delivered in those cases, and the response is unknown.
Secondary Effectiveness Endpoint (Descriptive):
Percentage of time a bolus administered after an AFM Test suggestion resulted in an increase in stroke volume meeting the selected fluid strategy.	Not a primary acceptance criterion, but reported descriptively.	60.5% [57.8, 63.2] (for AFM Test suggestions)

Other relevant performance data:

• User Boluses (Clinician-initiated boluses outside AFM recommendations): 40.9% [37.4, 44.1] of the time, user-administered boluses resulted in an increase in stroke volume. However, the study explicitly states that "it is not appropriate to compare AFM boluses against user boluses," as the study was not designed for this comparison.

Study Proving Device Meets Acceptance Criteria

The study used to prove the device meets acceptance criteria is the Assisted Fluid Management IDE study (AFM IDE study), identified by ClinicalTrials.gov identifier NCT03469570.

1. Sample Size and Data Provenance:

• Test Set Sample Size:
  • 330 subjects were initially enrolled.
  • 307 subjects were assigned to the per-protocol pivotal cohort and included in the effectiveness evaluation for the primary endpoint.
  • The primary effectiveness endpoint was based on the 54% (165/307) of subjects who received and followed AFM Recommended suggestions.
• Data Provenance: Retrospective and prospective. The AFM IDE study was a prospective, multi-center, single-arm clinical study. Data for comparison (historical control) was from a retrospective sub-analysis of the OPTIMISE trial. The AFM IDE study was conducted at study sites in the United States (US).

2. Number of Experts and Qualifications for Ground Truth (Test Set):

• The document does not explicitly state the number of experts used to establish ground truth or their specific qualifications for the test set.
• The ground truth for effectiveness (fluid responsiveness) was determined by measuring the percent increase in stroke volume (SV%) following a bolus and comparing it to the selected fluid strategy threshold (e.g., 15% increase for a 15% strategy). This is a physiological measurement, not directly an expert interpretation of an image or signal. Clinical decisions were made by the clinicians in charge during the study, and their actions (administering fluid after an AFM recommendation) were then assessed for outcomes.

3. Adjudication Method for the Test Set:

• For safety events, a Clinical Events Committee (CEC) reviewed and adjudicated events for anticipation, severity, and relatedness to fluid management.
• For effectiveness, the assessment was based on whether the measured physiological response (stroke volume increase) met the predefined fluid strategy threshold. There is no explicit mention of an adjudication method (like 2+1 or 3+1) for the primary effectiveness endpoint, as it relies on objective physiological measurements monitored by the device.

4. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study:

• No, an MRMC comparative effectiveness study was not done in the conventional sense of human readers interpreting data with and without AI assistance to assess diagnostic improvement.
• This device is an "adjunctive open loop fluid therapy recommender," meaning it provides suggestions to clinicians who then make the final decision and administer fluid. The study evaluated the outcome of following the device's recommendations (i.e., did the patient become fluid responsive as predicted?).
• The comparison was against a historical performance criterion (30% fluid responsiveness) rather than a direct comparison of human performance with and without AI assistance in real-time decision-making scenarios where human performance itself is being measured and improved. The text states: "The AFM IDE study was not designed to compare against manually administered fluid management protocols."

5. Standalone Performance (Algorithm Only without Human-in-the-Loop Performance):

• The primary effectiveness endpoint was not purely standalone. It evaluated the performance of the device's recommendations followed by clinician action. The outcome measured was the percentage of times an AFM recommendation that was followed by a clinician-accepted and clinician-delivered bolus resulted in the desired physiological change.
• The algorithm generates the recommendations (standalone function), but the ultimate "performance" (i.e., whether the patient responded as predicted by the recommendation) is assessed in the context of it being a decision support tool where the human makes the final decision. The study notes that a "major study limitation" was that decline rates were high for AFM recommendations (~50%), and the outcome for these declined interventions is unknown.

6. Type of Ground Truth Used:

• The ground truth for the effectiveness endpoint was based on physiological outcomes data: specifically, the percent increase in stroke volume (SV%) after a fluid bolus, compared against a pre-selected fluid strategy threshold (e.g., 10%, 15%, 20%). This is an objective, measured physiological response.
• The historical control for comparison was also derived from clinical study data (OPTIMISE trial).

7. Sample Size for the Training Set:

• The document does not explicitly state the sample size for the training set used to develop the AFM algorithm.
• It mentions that "Algorithm unit testing was performed using privately collected patient data."
• "The AFM algorithm learns from the stroke volume response to each fluid bolus to determine if a patient is in a fluid responsive or pre-load dependent state. The algorithm uses data from the current patient in order to predict their fluid responsiveness; this data is not used by the algorithm to determine fluid responsiveness in future patients." This suggests a patient-specific learning component rather than a large, fixed, pre-trained model for all patients.

8. How Ground Truth for Training Set was Established:

• The document describes the algorithm's learning process: "The AFM algorithm learns from the stroke volume response to each fluid bolus to determine if a patient is in a fluid responsive or pre-load dependent state." This implies that the ground truth for training (or rather, for its adaptive learning) is the actual measured physiological response (stroke volume change) of a patient to administered fluid boluses.
• The animal study also provided "non-clinical justification for the basic validity of the AFM algorithm" by showing more fluid suggestions in hypovolemic states compared to hypervolemic states. This could be considered a form of "validation data" or coarse "ground truth" for the algorithm's underlying physiological model, but it's not described as the primary training data.

Regulation Number and Section

§ 870.5600 Adjunctive open loop fluid therapy recommender.

(a)
Identification. The adjunctive open loop fluid therapy recommender is a prescription device that uses software algorithms to analyze cardiovascular vital signs and predict a patient's estimated response to fluid therapy. The device is intended for adjunctive use with other physical vital sign parameters and patient information and is not intended to independently direct therapy.(b)
Classification. Class II (special controls). The special controls for this device are:(1) Clinical performance testing under anticipated conditions of use must fulfill the following:
(i) A summary of the clinical performance testing must include the relevant patient demographics, and any statistical techniques used for analyzing the data;
(ii) Subjects must be representative of the intended use population for the device. Any selection criteria or sample limitations must be fully described and justified;
(iii) Testing must demonstrate the recommendation consistency using the expected range of data sources and data quality encountered in the intended patients, users, and environments; and
(iv) Testing must evaluate the relationship between algorithm recommendations, therapeutic actions, and predicted physiological event or status.
(2) A software description and the results of verification and validation testing based on a comprehensive hazard analysis and risk assessment must be provided, including:
(i) A full characterization of the software technical parameters, including algorithms;
(ii) A description of the expected recommendation, accounting for differences in patient condition and environment;
(iii) A description of all mitigations for user error or failure of any subsystem components (including signal detection, signal analysis, data display, and storage) that affect the device's recommendations;
(iv) A characterization of algorithm sensitivity to variations in user inputs;
(v) A characterization of sensor accuracy and performance;
(vi) A description of sensor data quality control measures; and
(vii) Safeguards to reduce the possibility of fluid overload.
(3) A scientific justification for the validity of the algorithm(s) must be provided. This justification must include non-clinical verification and validation of the algorithm calculations and clinical validation using an independent data set.
(4) A human factors and usability engineering assessment must be provided.
(5) Labeling must include:
(i) A description of what the device measures, how the device decides to issue recommendations, and the expected range of frequency of recommendations, while accounting for differences in patient condition and environment;
(ii) Detailed information regarding limitations of the device's algorithm, and key assumptions made when the device issues a recommendation;
(iii) Warnings identifying sensor acquisition factors that may impact measurement results;
(iv) Warnings identifying user errors that affect the device's recommendations;
(v) Detailed information regarding the expected impact of user input errors on the device recommendations;
(vi) Guidance for interpretation of the device's recommendations, including a description that the recommendation is adjunctive to other physical vital sign parameters and patient information;
(vii) Description of the impact of the compatible sensor(s) on the device's performance;
(viii) The expected performance of the device for all intended patients, users, and environments;
(ix) Relevant characteristics of the patients studied in the clinical validation (such as age, gender, race or ethnicity, and patient condition) and a summary of validation results; and
(x) Description of the software safeguards that are in place to prevent fluid overload, and description of any limitation of the software safeguards.