Analysis of Devices Authorized by the FDA for Clinical Decision Support in Critical Care

Abstract

This case series study examines the clinical evidence cited for US Food and Drug Administration–approved clinical decision support devices for use in the critical care setting.

The use of predictive clinical decision support (CDS) devices (ie, those that use machine learning [ML] or artificial intelligence [AI]) has the potential to improve outcomes in critical care, but a clear regulatory framework is lacking.¹ Recent guidance from the US Food and Drug Administration (FDA) suggests most CDS tools for critical illness will be regulated because of the time-sensitive nature of the decisions informed by these devices. However, growing concerns about the clinical impact of predictive CDS systems raise questions about whether current device regulatory frameworks, developed before advanced statistical learning methods were widely available, are sufficient to ensure effectiveness and safety.²

On September 22, 2021, the FDA released a public database of authorizations for medical devices that use ML or AI. We sought to identify devices that offer CDS in a critical care setting and characterize the evidence cited in their authorization.

Methods

We extracted data from the AI and ML database as of December 15, 2022, and augmented those data through the OpenFDA interface.^3,4 The 2 most common FDA pathways for CDS device approval are the 510(k) pathway, which requires demonstration of substantial equivalence to a previously authorized device (hereafter, a predicate) and typically does not require submission of clinical data, and the de novo pathway, which indicates that a novel device has low to moderate risk and reasonable assurance of being safe and effective. We identified devices relevant to critical care and searched the main FDA database for additional high-profile devices that were not found in the AI/ML data set.⁴ For each device, we recorded whether the decision summary included clinical evidence, software code, safety evaluation, and consideration of potential performance bias among historically marginalized groups. For each device, we traced back all predicates developed before CDS systems using AI/ML methods were widely used. This case series was exempt from institutional review and informed consent by the Common Rule given its use of publicly available data.

Results

Of 521 authorizations in the FDA AI/ML database, we identified 10 that might inform care for patients with critical illness (Table 1). Of these, only 3 included citations of published data, 4 mentioned a safety assessment, and none mentioned an evaluation of performance bias (Table 2). All but 1 device were authorized through 510(k) clearance, which relies on substantial equivalence to predicates, but only 3 devices included AI/ML predicates. Notably, a high-profile sepsis-focused prediction model that ostensibly would meet criteria as a CDS device was not found in the AI/ML or full FDA databases.⁵ We found no studies examining the clinical impact on care processes or patient outcomes for these device authorizations.

Table 1. Summary of Devices Using AI and Machine Learning for Clinical Decision Support in Critical Illness That Have Been Approved or Cleared by the US Food and Drug Administration.

Device	Company	Brief description	Authorization No.	Final decision date	Approval typea
Analytic for Hemodynamic Instability	Fifth Eye Inc	Analyzes ECG waveform to indicate hemodynamic instability	DEN200022	3/1/2021	De novo
Visensia	OBS Medical	Analyzes vital signs to create an index indicating physiologic status	K081140	7/17/2008	510(k) Premarket notification
WAVE Clinical Platform	Excel Medical Electronics, LLC	Analyzes vital signs to predict future hemodynamic instability	K171056	1/4/2018	510(k) Premarket notification
PeraServer and PeraTrend	PeraHealth, Inc	Platform and display for an index integrating 26 variables derived from electronic health record to identify patients at risk of clinical deterioration	K172959	5/1/2018	510(k) Premarket notification
AI-ECG Platform	Shenzhen Carewell Electronics, Ltd	Measures and interprets ECG	K180432	11/19/2018	510(k) Premarket notification
RhythmAnalytics	Biofourmis Singapore Pte. Ltd	Measures and interprets ECG	K182344	3/7/2019	510(k) Premarket notification
PeraMobile and PeraWatch	PeraHealth, Inc	Mobile app and graphic interface to display an index integrating 26 variables derived from electronic health record to identify patients at risk of clinical deterioration	K183370	9/11/2019	510(k) Premarket notification
Acumen Hypotension Prediction Index	Edwards LifeSciences, LLC	Analyzes vital signs to predict future hypotensive events, expanding from prior approval for surgical patients to nonsurgical patients	K183646	5/21/2019	510(k) Premarket notification
AI-ECG Tracker	Shenzhen Carewell Electronics Co, Ltd	Measures and interprets ECG	K200036	3/20/2020	510(k) Premarket notification
CLEWICU System (ClewICUserver and ClewICUnitor)	Clew Medical Ltd	Integrates broad range of data from electronic health record and clinical data to identify patients at risk of clinical deterioration	K200717	1/9/2021	510(k) Premarket notification

Abbreviations: AI, artificial intelligence; ECG, electrocardiography.

^a US Food and Drug Administration authorization was given through the 510(k) pathway, a premarket submission made to demonstrate that the device to be marketed is as safe and effective as another equivalent, legally marketed device or through the de novo pathway, which can be used for a novel device that has a low to moderate risk and reasonable assurance of being safe and effective but for which there is no predicate device.

Table 2. Characteristics of US Food and Drug Administration Decision Summary Documents for Approved Devices Using AI and ML for Clinical Decision Support in Critical Illness.

Device	Any AI/ML predicate?	Year of earliest identifiable predicate	Studies referenced	Peer-reviewed?	Sample size	No. of study sites	Software code available?	Safety assessment?	Performance bias assessment?
Analytic for Hemodynamic Instability	NA	NA	1 Prospective observational study, 1 retrospective study	Yes	222 Participants in prospective study; 597 participants in retrospective study	1 Medical center	No	Yes, with risks to health identified and mitigation measures	No
Visensia	No	1992	0	NA	NA	NA	No	No	No
WAVE Clinical Platform	No	2009	0	NA	NA	NA	No	Discussed, but no assessment in documentation	No
PeraServer and PeraTrend	No	1991	3 Retrospective cohort studies	Yes	Not specified	Not specified	No	No	No
AI-ECG Platform	No	1982	0	NA	NA	NA	No	No	No
RhythmAnalytics	Yes	2000	0	NA	NA	NA	No	Discussed, but no assessment in documentation	No
PeraMobile and PeraWatch	Yes	1991	0	NA	NA	NA	No	No	No
Acumen Hypotension Prediction Index	No	2018	0	NA	NA	NA	No	No	No
AI-ECG tracker	No	2000	0	NA	NA	NA	No	No	No
CLEWICU system (ClewICUserver and ClewICUnitor)	Yes	1982	1 Retrospective cohort study	No	Not specified	7 Intensive care units in 2 hospitals	No	Discussed, but no assessment in documentation	No

Abbreviations: AI, artificial intelligence; ECG, electrocardiographic; ML, machine learning; NA, not applicable.

Discussion

While many prediction models might offer CDS for patients with critical illness,⁶ our review of the database revealed that only 10 AI/ML CDS devices have received FDA authorization. The clinical evidence for these devices ranged from completely absent to peer-reviewed assessment of model performance, and most of the devices authorized through the 510(k) pathway relied on equivalence to non-AI/ML predicates. Furthermore, at least 1 high-profile and widely implemented model⁵ did not appear to have received FDA authorization. While this study was limited to critical care, these findings highlight the need to update regulatory requirements to align with current knowledge about using AI/ML systems across many clinical practice settings.

Although the release of the curated FDA database⁴ permits easier identification of FDA-authorized devices that rely on AI/ML methods, users must look elsewhere to obtain essential information about the clinical effectiveness, safety, and performance biases of a given CDS system. The criteria for establishing equivalence for 510(k) clearances could be better adapted to both current AI/ML methods and the clinical environment of high-risk decisions for patients with critical illness.

Limitations of this study include not accounting for FDA approvals made through the recently ended precertification pilot pathway or for other devices in widespread use not present in the FDA database.

Supplement.

Data Sharing Statement