Etiometry

A real split between a data platform and the risk indices on it. The indices estimate, continuously, the probability that a patient is in a defined physiologic state, using what the company calls a Bayesian approach that updates on each patient's incoming vital signs, device data and laboratory results; the K254066 summary describes them as produced by a model designed according to physiological principles with parameters drawn from the medical literature, and the company says they were developed using more than 150 million hours of de-identified patient data.

Around them sits a vendor neutral data aggregation, visualization, pathway and quality improvement platform that the company's own FAQ lists first among its capabilities, and that a hospital would still use with the indices switched off. The models are the engine of a core module, and the platform carries other value.

What the system may and may not do is stated, and the triggers belong to the hospital. The company states repeatedly that the platform is not an active patient monitoring or alarm system, provides adjunctive information and informational flags, supplements rather than replaces monitoring, must not be the sole source of patient status information, and leaves clinicians responsible for all assessment and treatment decisions.

Each index is defined against a published threshold (inadequate oxygen delivery as the probability of mixed venous saturation below a configurable 30, 40 or 50 percent; hypercapnia as PaCO2 above 50 or 60 mmHg; hyperlactatemia as lactate above 4 mmol/L; acidemia as pH below 7.25), and clinical pathway automation highlights patients only when hospital-defined criteria are met, for clinician review. Nothing is auto-actioned. The company's FAQ, though, describes the indices across adult cardiac conditions that three of them are not cleared for, so its marketing and its labeling draw different population limits.

What produces the output, what it estimates and which version is running are all public. The indices are physiologic models: the K254066 summary says they are designed according to physiological principles with parameters drawn from the medical literature, and the company describes a Bayesian approach that updates each patient's estimated state as heart rate, blood pressure, SpO2 and, where available, filling pressures, venous oxygen saturation, hemoglobin and blood gases arrive, tolerating missing inputs.

Each estimand is defined against a laboratory threshold a clinician can check by drawing blood. Versions are named: K241479 cleared DAV 5.4 with RAE 9.2, and the current electronic instructions for use cover DAV 5.4 with RAE 10.0, with document revision and date, and each change has gone through a 510(k). Validation cohorts are characterized in the filing (779 adults and 3,281 pediatric patients for K254066) and in peer reviewed validation papers, and a public patents page supports the method. A.

A structurally short chain. The indices are the company's own physiologic models running on servers inside the hospital, with no outside model provider in the data path. The parties that do sit in the path are named: medical device integration platforms such as Capsule, Bedcomm and IBUS carry device data in, typically under the hospital's own licenses. The one flow out is the de-identified research corpus the company holds, governed by each hospital's agreements. No subprocessor list is published.

A substantial peer reviewed record, almost all in pediatric cardiac intensive care, and the headline figures run ahead of it. The home page leads with "Reduce Length of Stay by up to 36%" and readmissions "by up to 41%". The 36 percent traces to a 2017 conference abstract (Salvin, Circulation supplement).

The 41 percent comes from Gaies and colleagues (Circulation: Cardiovascular Quality and Outcomes, 2023), an independent single-site natural experiment at the University of Michigan using registry data from 21 control hospitals, which found relative reductions of 41 percent in unplanned readmissions, 18 percent in postoperative length of stay and 14 percent in cardiac arrest among medical admissions.

The largest trial, Gazit and colleagues (Critical Care Medicine, 2025; three pediatric cardiac ICUs, 775 infants after bypass, before-and-after design, with two Etiometry authors), found 29 percent shorter vasoactive infusions and no discernible change in ICU length of stay, cardiac arrest or weaning failure. Validation studies at 9 and 10 pediatric centers report discrimination for the acidemia index (0.93 on 1,858 patients, 2025) and hyperlactatemia index (0.95 on 3,496 patients, 2023).

The company says plainly that these results are observational, vary by site and do not establish causality. Published research short of the multi site prospective evidence the length of stay headline would need, and adult evidence is thin.

The company is candid about how widely it reuses hospital data. It says the hospital owns its patient data, and that it has accumulated more than 150 million hours of deidentified physiologic data used "to develop and validate its risk algorithms" and for research, governed by each hospital's master services agreement and data use agreement.

It also says every data point ingested is stored indefinitely in its Quality Improvement Application, with full-fidelity waveforms archived separately. No de-identification method, opt-out, retention limit or statement of what happens to a hospital's contribution at termination is published. The platform runs inside the hospital's environment, which limits exposure for identified data. No guardrail or safety event process specific to the indices is published.

A substantive governance page stands behind the stated business associate status. The data governance page says Etiometry signs business associate agreements, that the hospital retains ownership of its data, that de-identified data use is governed by the master services agreement and data use agreement, and that encryption, access management, audit logging, incident response and business associate provisions are documented and provided during onboarding and security review. The agreement itself is not published, though the website privacy policy may add detail.

Recognized certifications are named, and the reports are handed over during security review rather than published. The company states ISO 27001 certification, ISO 13485 for medical device quality, SOC 2 Type II compliance (elsewhere described as compliance with the SOC 2 and HIPAA trust services criteria with third-party operational audits), and that its last two clearances treated the platform as a cyber device under Section 524B, which requires a cybersecurity review by FDA.

Published controls include TLS 1.3, access through the hospital's Active Directory with location-based role restrictions, reauthentication after 15 idle minutes by default, hardened RHEL servers and no inbound internet traffic. Encryption, audit logging, incident response and business associate provisions are documented and "made available during the customer onboarding and security review process"; there is no trust portal, report date or auditor named. B.

Eleven clearances are real, and the company's own FAQ claims cleared status for patients the risk indices are not cleared for. FDA's database lists eleven Etiometry 510(k)s from K142732 (March 2015) to K254066 (3 April 2026), the last two for the Etiometry Platform (K241479 names DAV 5.4 and RAE 9.2) under product code PPW, identified as a cyber device under Section 524B.

The K254066 indications are narrow and specific: IDO2 for post-surgical patients 0 to 12 years and 2 kg or more in intensive care, and adults 18 and over in intensive care who are not on mechanical circulatory support; IVCO2 for invasively ventilated patients 0 to 12 years; ACD for invasively ventilated patients 0 to 12 years and 2 kg or more; HLA for post-surgical patients 0 to 12 years and 2 kg or more. So only IDO2 is cleared for adults.

The risk analytics page states limits close to these. But the FAQ says "These indices are cleared for pediatric patients above 2 kg and for adult patients", says IVCO2 is "cleared across all pediatric and neonatal populations", and describes "FDA-cleared risk indices" tracking hypoxia, hypercapnia, hyperlactatemia and acidemia across adult cardiac conditions including STEMI, heart failure, CABG and valve surgery. The same FAQ dates the first IDO2 clearance to 2014, while FDA's database shows no Etiometry clearance before March 2015. The cleared label attached to uses the clearances do not cover.

A monitoring commitment without method, and no results by group. The FAQ says the company "continuously monitors its own algorithm performance in production against each ICU's real-world patient population, so any drift in accuracy is identified proactively", but publishes no method, threshold or report. No performance by age, sex, race or skin pigmentation is published for any index, and the K254066 filing reports only that adult and pediatric validation sets met the acceptance criteria.

One exposure is specific: IDO2 takes SpO2 as an input, and pulse oximetry is known to overestimate oxygen saturation in patients with darker skin, so an oxygen delivery index built on it could understate risk in exactly those patients unless calibrated for it. The cleared device and its quality system are counted under regulatory status. C.

The indices can be checked against a blood test, and the company cites a validation with its denominator. Each index estimates a measurable quantity (lactate above 4 mmol/L, pH below 7.25, mixed venous saturation below a set threshold, PaCO2 above a set level), so a clinician who doubts a value can draw blood and see.

On its research page the company cites Asfari and colleagues (Critical Care Explorations, 2023), a retrospective validation at ten pediatric cardiac ICUs in which the hyperlactatemia index discriminated lactate above 4 mmol/L with an area under the curve of 0.95 across 58,168 lactate measurements from 3,496 children. Nothing stands behind the output beyond that: no warranty, indemnity or remediation commitment is published, and sensitivity at an operating point is not given.

Named record systems with detailed read paths, and an outbound path not tied to the record. The FAQ describes Epic and Oracle Health (Cerner) sending admissions and laboratory results over standard HL7 interfaces, with medications, procedures and diagnoses captured through Etiometry's FHIR application and SMART on FHIR single sign-on, and Meditech over HL7.

On the device side it names bedside monitors from GE Healthcare, Philips, Draeger, Mindray, Nihon Kohden and Spacelabs, ventilators from Medtronic, Draeger, Hamilton Medical and Getinge, ECMO, VAD and CRRT systems, cerebral oximetry and Edwards HemoSphere, through integration platforms such as Capsule, Bedcomm and IBUS, at numerics every five seconds and waveforms up to 500 Hz.

Derived indices can be sent out over HL7 and REST to "other systems", but writing into the EHR is not described, no record system marketplace listing could be confirmed, and one FAQ answer says integration needs no HL7 interfaces, contradicting the others. Named systems with read access and a general outbound interface.

Deployment is documented in detail and the de-identified data path is not. The platform runs on virtual servers the hospital installs, on hardened RHEL with no inbound internet traffic required, with access through the hospital's own Active Directory, remote access through the hospital's VPN, and a named Technical Readiness gate (servers, software, device and EHR integrations, authentication and access points) before clinical rollout, which the company says can take as little as two months.

Patient data therefore rests in the hospital's environment. What leaves it is not described: the company holds more than 150 million hours of de-identified physiologic data used to develop its algorithms, governed by each hospital's services and data use agreements, but where that corpus is stored, how it is extracted and whether any of it crosses a border is not stated. Options and residency stated, with the processing path for the research corpus left open. The ICU count is also given as both 175 and roughly 150 on the same FAQ page.

No price, and how the product is bought can be established. The platform is licensed to hospitals and installed on hospital virtual servers, with a described implementation (technical readiness, super user and unit training, ongoing clinical meetings) that can take about two months; device data may arrive through integration platforms the hospital licenses separately. No unit of charge, band or implementation fee is published, and clinical outcome highlights are offered through a request form. C.

Coverage is named with validation behind the pediatric intensive care part. The cleared indications (K254066) confine three of the four indices to children 0 to 12 years in intensive care and allow IDO2 in adults only in intensive care and off mechanical circulatory support; the data aggregation module is cleared for adult, pediatric and neonatal patients. The published validation and outcome studies are overwhelmingly pediatric cardiac intensive care at academic children's hospitals.

The company describes wider coverage: ICU, CICU, PICU and NICU plus step-down units, adult cardiac conditions from STEMI to heart failure and cardiac surgery, and a strategic aim of extending to pre-hospital and home monitoring, without adult or step-down validation of the indices. B.