Current Health

The models work at two points in the service. On the clinical side, algorithms read the wearable's continuous vital signs to flag early cytokine release syndrome, sepsis and general deterioration, looking for sustained change across several signals at once rather than one reading crossing a line, and an alarm can trigger an automated chatbot check in before a nurse calls. On the operational side, conversational AI answers routine and technical calls for the command center, and AI tools plan nurse routes and schedule home services.

The rest of a program is people and equipment: the wearable and peripherals, the 24/7 nurses, kit logistics and consulting. In the Mount Sinai CRS study, the best performing detector was a temperature rule measured against each patient's own baseline.

Every alarm reaches a nurse before anyone acts on it. Command center RNs and medical assistants watch trends, triage alarms against each program's pathway, round proactively on patients, run ICE score neurotoxicity assessments by video with handwriting capture, and escalate any ICANS event above the mildest level to the treating physician at once. The company reports a median alarm response of 2.3 minutes, says 29 percent of patients who trigger a clinical alarm are managed at home, and holds daily huddles with partner clinicians.

Alarm parameters and escalation paths are set per program and per patient from more than 50 DRG specific pathways, and an alarm such as low oxygen saturation can prompt a chatbot symptom check or a video call before escalation. Default threshold values are not shown publicly. One bispecific cohort is presented as 100 percent of CRS events captured, while the Mount Sinai CAR T study the company also cites detected 18 of 20.

The wearable streams respiratory rate, heart rate and its variability, skin temperature, oxygen saturation and movement, an axillary patch adds core temperature, and the alarm logic looks for sustained change across several of those signals rather than a single breach. The company says its models train on more than 1,200 CAR T episodes and billions of data points, and in the Mount Sinai study the best performing CRS approach compared axillary temperature against each patient's baseline and a fixed threshold, scored over 60 minute windows.

Interferon gamma is being studied as a blood marker to pair with the wearable, and the company developed a published ontology of digital CRS biomarkers with drug makers and academic groups. Release notes sit in the customer help center, and the model version running in the shipped product is not stated publicly.

Two logistics partners are named. Cardinal Health's Velocare delivers, installs, troubleshoots and collects kits and brings supplies, equipment, oxygen and meals to patients' homes within two hours, and Impilo powers device delivery, retrieval and reprocessing nationwide. Community programs lean on mobile apheresis providers, blood banks and infusion partners, and kits carry third party peripherals such as pulse oximeters, glucometers and spirometers.

On the data side, the privacy statement refers to third party cloud hosting and analytics providers and an outside alert monitoring firm, and no hosting, analytics or conversational AI provider is named.

Peer reviewed studies from partner cancer centers carry the clinical case, with Current Health staff among the authors. At Mount Sinai, a prospective single center study of 25 evaluable myeloma patients on CAR T therapy found the best wearable model caught 18 of 20 cytokine release syndrome episodes a median of seven hours before nurses recognized them (JCI Insight, 2026). Across four Sarah Cannon sites, 209 outpatient CAR T patients monitored on the platform had 97 to 98 percent survival at 30 days and severe CRS in 4 to 5 percent (JCO Oncology Practice, 2025).

Program results come from customers. Sarah Cannon treats about 90 percent of its CAR T patients as outpatients, saved 1,200 bed days in its first 100 and shortened inpatient stays to a median of 4 days from 16. Mass General Brigham's Home Hospital tripled its throughput in 15 months on the platform and reports 25,000 bed days saved, a 6.7 percent escalation rate, a 10.5 percent readmission rate and no unexpected deaths, and 89 percent of Memorial Sloan Kettering patients who began bispecific step up dosing as outpatients reached full dose. Earlier conference abstracts from the Mount Sinai work put the lead time at 1.7 to 3 hours, and no controlled trial of program outcomes is published.

Patient data trains the models. The patient privacy statement lets Current Health use identifiable monitoring data to develop prediction of worsening, biomarker monitoring, disease detection and post market surveillance, and the terms of use add patient data to the database behind the platform's AI for refining diagnostic algorithms and other commercial purposes. De identified and aggregated data can go into publications and to third parties for their own research and development, and drug makers receive de identified safety data for pharmacovigilance.

Retention follows each provider's contract, in most cases until six years after the provider stops using the platform, and patients can revoke consent through their provider or directly. Stored data is encrypted, and retrieved kits go back through the logistics partner for reprocessing before reuse. Both documents date from November 2022, and the company describes no safety engineering for the models themselves.

Current Health acts as the business associate of each provider that enrolls patients, under a business associate agreement with that customer, and as a data processor under UK GDPR. The patient privacy statement says so, ties uses of patient data to that agreement and lists what is collected, from continuous vital signs to blood pressure, spirometry, weight and glucose from third party devices, and product pages describe the service as HIPAA compliant.

The agreement is signed per customer rather than posted, so its tier and its terms for subcontractors, including the logistics partners whose staff set up and collect kits in patients' homes, are settled in contracting. The statement dates from November 2022, before the 2025 change of ownership.

SOC 2 Type 2 and ISO 13485 appear on every product page, and the life sciences page adds ISO 27001. The patient privacy statement describes encryption of stored data, multifactor authentication, firewalls and staff security training, and retrieved device kits are reprocessed through the logistics partner before they go to another home.

The system spans an upper arm wearable, home hubs on cellular networks, tablets that move between households and a cloud platform. Reports, certificate dates and device level controls such as firmware updates are not posted.

The upper arm wearable and the monitoring platform hold five FDA 510(k) clearances as a Class II physiological monitoring system (product code MSX; K190073 through K231506, 2019 to 2023), cleared for post acute and remote patient monitoring, along with an EU CE mark as Class IIa and ISO 13485 certification for device quality; an Australian TGA listing has been applied for. Product pages describe the monitoring as FDA cleared and the CRS algorithms as proven by research.

The CRS model is on a device route of its own. Software as a medical device for clinical decision support is listed in development, and in October 2026 FDA selected the model for READI-Home, a CDRH program that gives home use devices early regulatory engagement, with the company noting that selection is not a marketing authorization. The patient terms of use, last updated in 2022, still describe the wearable as an FDA approved Class I device.

Alarm behavior is studied in the field and reported by cohort. Published work maps alarm patterns across a multicenter lisocabtagene maraleucel cohort, where 14 percent of patients never triggered an alarm, compares alarm volume between heart failure and COPD hospital at home programs, measures day and night vital sign variability to tune alarm settings, and found that 10 percent of patients generated half of all alarms. The platform tracks signal quality on every patient and flags data gaps, and post market surveillance is a named use of patient data.

Access work covers who can use the service: multilingual clinical and technical support, interfaces built for low literacy users, cellular kits for homes without internet, and studies showing patients over 75 keep up with monitoring. Detection accuracy has not been compared across patient groups, and the wearable reads oxygen saturation optically, which tends to overestimate it in people with darker skin.

The CRS detection has been measured. The company cites the Mount Sinai study (Rajeeve and colleagues, JCI Insight, 2026, led by Mount Sinai with Current Health coauthors), in which the best wearable model detected 18 of 20 cytokine release syndrome episodes in 25 evaluable patients, with a sensitivity of 0.72, a specificity of 0.80 and a median lead of seven hours, in hospitalized patients with device outputs blinded. Every alarm is reviewed by a nurse, and adverse events are time stamped with the intervention that followed, giving each episode an audit trail.

The patient terms of use cap Current Health's liability at $1,000 and offer discontinuing the service as the only remedy. Health system contracts, including any service level for alarm response, are negotiated per program.

Integration runs both ways with Epic, Cerner and Meditech: vital signs and alerts post to the chart, and orders, medications and clinical notes feed the monitoring protocols, through direct HL7 or FHIR interfaces or integration partners, with EHR specific technical packages and templates for implementation. With Cardinal Health's Velocare, programs can enroll a patient and order the monitoring kit from inside Epic, and patient reported symptoms flow into clinical workflows.

Best Buy Health announced the platform's listing in Epic's Toolbox for remote patient monitoring with devices in March 2025, before the divestiture. Virtua uses the dashboard for its post discharge program across five hospitals.

US patient data is stored in US cloud infrastructure and UK patient data in Ireland, encrypted at rest. The home hub connects over cellular networks, so patients need no home internet, and the tablet carries video visits, symptom surveys and medication reminders; device engineering and data science sit in Edinburgh while clinical operations run from Boston.

The privacy statement also says UK data never leaves the UK, which does not square with storage in Ireland. The cloud provider and how monitoring keeps running through a network or platform outage are not described.

Programs are priced per care episode through risk based or bundled arrangements, and the consulting team builds the economics with the customer: actuarial episode costing, business case and ROI modeling, and negotiation of bundles or case rates with commercial payers. Service levels are configured per episode type to meet payer requirements, and the health system pitch weighs outpatient CAR T against inpatient days costing $3,000 or more and against the 340B pricing that outpatient delivery preserves.

UCSF's director of nursing for infusion and apheresis is quoted saying the savings from keeping patients outpatient made the cost of Current Health's support a wash. Hospital at home revenue depends on the CMS waiver, which Current Health says now runs through 2030. No list or unit prices are given.

Four kinds of program run on the platform: outpatient CAR T and bispecific delivery for health systems and academic centers, the same service for community oncology practices, hospital at home under the CMS Acute Hospital Care at Home program alongside post discharge recovery, and safety monitoring in drug trials from Phase I to IV. Supported therapies include Yescarta, Kymriah, Breyanzi, Tecartus, Abecma and Carvykti and the bispecifics Tecvayli, Talvey, Elrexfio, Lunsumio, Columvi and Epkinly, with tumor infiltrating lymphocyte therapy in development, and post discharge pathways cover cardiac, orthopedic, transplant, wound care and chronic exacerbation patients.

Named programs sit behind most of this: Sarah Cannon's four site CAR T network, Memorial Sloan Kettering's bispecific step up dosing, Mass General Brigham's Home Hospital, UMass Memorial, Virtua's post discharge program across five hospitals, and UK virtual wards. Kit logistics, scheduling and home services make it a home care operations product too, and the inpatient beds it frees make it a hospital operations purchase. Community practice programs have no published results yet.