Sibel Health

Sibel is a sensor engineering company, and its product delivers value before any model runs. The hard problem it solved is physical: soft, flexible, wireless skin sensors (ANNE Chest and ANNE Limb) that capture clinical grade electrocardiography, photoplethysmography, oxygen saturation, respiratory rate, temperature and position on patients from extremely premature neonates to adults and laboring mothers, cleared eight times for measurement, display and alarming.

Vital signs are directly interpretable, and the central hub's alarms and the maternal product's obstetric early warning score apply thresholds and a published clinical scale rather than a learned model. Company material has described AI enabled analytics, smart alarming and derived positioning insights, but none of it describes a learned model, and the diagnostic summary for ANNE Sleep (K220095) produces an apnea index scored manually by a clinician. AI is at most a feature layer on a product whose value stands without it.

The alerting structure is described and its calibration is not. The central hub, ANNE View (K242842, cleared January 2025), displays and alarms across multiple patients, results reach clinicians who decide, and the software takes no clinical action.

The maternal platform integrates an established obstetric early warning system that generates automated signals when prompt evaluation is indicated, so a midwife can see what triggered an alert and check the reasoning against a scale already taught and audited.

What is missing is the operating point: no alert volume per patient per day, false alarm rate, sensitivity for deterioration, time to notification or intended nurse to patient ratio is published, although alert fatigue is the documented failure mode of ward monitoring. European material describes smart alarming as letting nurses cover more patients safely, a supervision claim with no number.

What produces the measurements is described precisely, and the analytic layer is named only. The chest sensor is documented as producing electrocardiography, skin temperature, heart rate, respiratory rate, body position and activity, and the limb sensor photoplethysmography, oxygen saturation, pulse rate and skin temperature; the central hub also shows third party cleared blood pressure and temperature devices, and the sensors communicate over a named open standard.

The maternal early warning is a published clinical scale that any clinician can inspect, and the ANNE Sleep summary (K220095) states that its apnea index is scored manually by a clinician rather than computed by the device. The AI enabled analytics, smart alarming and positioning insights described in company material have no description of what they compute, no model, no operating point and no version or update practice.

The core dependencies are visible and the chain around them is not enumerated. The sensors are Sibel's own, developed from Northwestern University research and cleared in its own name; the communication layer is an open published standard; a patient monitoring manufacturer is disclosed as investor and co development partner; and third party cleared devices feed the central hub.

No cloud or hosting provider, subprocessor register, contract manufacturer or model provider is named, and the only privacy policy published covers the website and names Google's advertising and analytics tools. No statement says what data the AI enabled analytics were developed on, whether clinical trial or research data from the philanthropically funded deployments is used commercially, or which parties receive patient data.

Regulatory performance data and named deployments across varied settings, without an outcome study. The ANNE Sleep 510(k) summary (K220095, July 2022) reports a single arm, open label, multi center US study of 225 adults against polysomnography: 90 percent sensitivity and 98 percent specificity for moderate to severe obstructive sleep apnea using a clinician scored index, and oxygen saturation accuracy of 2.31 percent root mean square error.

Each of the eight US clearances rests on its own performance testing, across neonates, adolescents and adults and maternal fetal monitoring. Deployments are named: the Capital Region of Denmark selected the platform with a monitoring manufacturer partner for Copenhagen hospitals, a Chicago pediatric hospital holds a collaboration, and the maternal platform is deployed in India, Pakistan, Nigeria and Rwanda with a UK university research unit on a $17.5 million philanthropic grant.

The home page reports 5,000 systems deployed, 20 countries and 15,000 patients monitored. No study shows that deploying the platform reduces deterioration, mortality, length of stay or nursing time, which is the commercial argument.

General assurances are published and the AI questions are not answered. Company material says platform data is synced securely to the cloud with full traceability for patient care, an auditable chain from sensor to record that lets a clinician verify when and by which device a reading was taken, and states compliance with HIPAA and EU data protection law for the remote monitoring platform.

The only privacy policy published (last modified 11 November 2025) covers website visitors: it names Google advertising and analytics tools, says data may be stored or processed outside the visitor's country and retained only as long as necessary, and does not reach patient data.

No retention schedule, encryption description, access model, ownership statement or position on whether physiological data trains models is published, for a dataset spanning neonates, laboring mothers (fetal data concerns two people), inpatients and home patients in about 20 countries. The clinical trials platform runs under different consent and retention rules, and nothing describes how the two are separated. No safety engineering is published.

Compliance is claimed and not supported by a document that reaches the service. Company material states that the remote monitoring platform complies with HIPAA and EU data protection law. The only privacy policy published (last modified 11 November 2025) covers website visitors, addresses EU and UK rights and standard contractual clauses for website data, and does not mention HIPAA, business associate status or patient data.

No business associate agreement, execution requirement or subcontractor position is published, and the compliance claim is scoped to home monitoring, with nothing stated for the hospital central hub, the clinical trials platform or the maternal system. Neonatal and maternal monitoring produce data about patients who cannot consent for themselves, in the maternal case two people through one system, and no material addresses consent, record ownership or retention for either.

No information security certification, attestation, trust center, penetration testing statement or vulnerability disclosure policy is published. The website privacy policy (11 November 2025) says the website is scanned regularly for vulnerabilities and malware, which concerns the website rather than the monitoring platform.

The open device communication standard the sensors use supports secure communication between devices, a protocol property rather than an audit of how the company operates security, and says nothing about the cloud platform, staff access or incident handling.

The surface is substantial: body worn wireless sensors, bedside hubs and a cloud platform in hospitals and homes across about 20 countries, with device authentication, firmware updates, pairing and decommissioning of sensors that leave with discharged patients all unaddressed. EU Class IIb certification and a regional hospital deployment imply security review in private; none of it is published.

The regulatory position is unambiguous and identifiable product by product. FDA's database lists eight 510(k) clearances for Sibel: ANNE One (K211305, September 2021; K223711, August 2023; product code DRG, 21 CFR 870.2910), ANNE Sleep (K220095, July 2022; MNR, 868.2375, aid to evaluating sleep related breathing disorders in adults), ANNE Pediatric (K221530, November 2022; DRG), ANNE Limb (K240305, May 2024; DQA oximeter, 870.2700), ANNE Chest (K240251, June 2024; DRG), ANNE View central hub (K242842, January 2025; MSX, 870.2300) and ANNE Maternal (K253021, 26 February 2026; HGM, 884.2740, maternal and fetal monitoring).

Company material reports EU MDR Class IIb certification in June 2026 and acceptance of the Aria cough sensor into FDA's drug development tool qualification program as a clinical outcome assessment. Each clearance covers measurement, display or alarming for its stated population; the AI enabled analytics described in some company material carry no separate clearance, and no claim is made that they do.

Sibel addresses the best documented measurement bias in monitoring by assertion, without the numbers. Pulse oximetry can overestimate oxygen saturation in patients with darker skin, a discrepancy serious enough to prompt FDA review and to delay recognition of hypoxia.

The company states it has validated its pulse oximeter's accuracy across a wide range of skin tones, and that work sits alongside philanthropically funded maternal monitoring deployed in India, Pakistan, Nigeria and Rwanda with a UK university research unit. No accuracy figures by skin tone category, sample sizes or study are published, so the statement is a robustness claim with no groups or counts behind it, and validating across a range does not show performance was equal across it.

The ANNE Sleep 510(k) summary (K220095) reports the racial makeup of its 225 subjects (73 percent white, 12 percent Black, 9 percent Asian) but no results by group. No fairness framework, other subgroup analysis or external audit is published, and signal quality across body habitus and across the neonatal and maternal populations is unaddressed.

Nothing is published on what happens when the monitoring is wrong. No service level agreement, accuracy warranty, uptime commitment, indemnity or remediation position appears on the company's pages, and no detection or false alarm performance is published for any alerting function, including the central hub's alarms (ANNE View, K242842).

The one performance figure read, ANNE Sleep's 90 percent sensitivity and 98 percent specificity against polysomnography (K220095, 225 subjects), is for an apnea index scored manually by a clinician, so it is not an error rate for any automated function.

Substitution raises the stakes: the platform is sold on freeing nurses from manual spot checks, so a ward that reduces an existing observation practice relies on the system, and signal continuity and alarm delivery become clinical properties. Neonates and laboring mothers, both served, are where undetected deterioration escalates fastest. Where a monitoring manufacturer partner resells or co develops, the party a hospital contracts with may not be Sibel.

Device interoperability rests on an open standard with named deployments behind it, and no record system is named. The company describes its sensors as operating under an open medical device communication standard designed to replace proprietary monitoring networks, announced its June 2026 EU certification as the first wireless wearable monitoring platform certified to key interoperability standards, and names a patient monitoring manufacturer, also an investor and co development partner, that cites the standard as the reason for the collaboration; the two were selected by the Capital Region of Denmark for Copenhagen hospitals.

The central hub, ANNE View (K242842), displays data from third party cleared blood pressure and temperature devices alongside Sibel's sensors, and a development kit lets partners build on the cleared sensors. What is missing is the chart: no electronic health record is named, no HL7 or FHIR interface is described, and nothing states whether continuous vital signs post to the record as discrete observations.

The local topology is clear and the hosted layer is not. Sensors sit on the patient, a central hub provides bedside and multi patient display and alarming within the facility, and data syncs to a cloud platform, which keeps alarming close to the patient rather than dependent on a round trip.

No cloud or hosting provider, region, residency commitment, tenancy model or continuity position is published; the website privacy policy says only that data may be stored or processed outside the visitor's country. Operation in about 20 countries, including EU deployment and sites in India, Pakistan, Nigeria and Rwanda with their own localization rules, makes regional processing a real question. A ward that has replaced manual observation rounds with wearable monitoring depends on the system, and whether the hub keeps working through connectivity loss is unstated.

Nothing a buyer can use to size the cost is published: no pricing page, unit of charge, range, tier, implementation fee or minimum commitment, and no route to buy beyond contacting the company. The cost structure is complex and unaddressed. Adhesives are consumables that scale with census, the rechargeable chest and limb sensors are a device fleet to size and replace, and the central hub and cloud software are platform costs, and nothing says how any of the three is charged.

Neonatal, adult, maternal, home, sleep and clinical trial buyers differ in volume and procurement, and where a monitoring manufacturer partner resells or bundles the platform, the partner may set the price. The staffing argument, nurses freed from manual spot checks, is calculable and not calculated.

Where the platform is validated to operate is named product by product, backed by separate clearances with their own performance testing. ANNE Pediatric (K221530) covers neonates including those born extremely premature with a sensor designed for fragile skin; ANNE One (K211305, K223711) covers adolescents from 12 and adults in hospital and at home; ANNE Sleep (K220095) covers adults suspected of sleep apnea, validated against polysomnography in 225 subjects at multiple US sites; ANNE Maternal (K253021) covers maternal and fetal monitoring through labor, delivery and postpartum; and the central hub (K242842) covers multi patient display and alarming.

Care settings named are general wards, step down units, hospital at home, home monitoring, sleep diagnostics and clinical trials, and the platform is deployed in Copenhagen hospitals and in maternal programs in India, Pakistan, Nigeria and Rwanda. The installed base by setting and country is not published beyond 5,000 systems and 20 countries in total.