AI Remote Patient Monitoring Is Now a Billable, Mainstream Program, Not a Pilot

Remote patient monitoring stopped being an experimental add-on a while ago. What changed recently is the amount of AI sitting inside the data pipeline, and the amount of money CMS is now willing to pay for shorter, lighter-touch monitoring periods. If you run a clinic or health system and haven’t looked at RPM since before 2026, the billing rules, the vendor list, and the evidence base have all moved.

This piece walks through what AI actually does inside RPM platforms today, which tools US health systems are actually using, what Medicare will and won’t reimburse as of the 2026 fee schedule, and where the real risks sit, especially alert fatigue, which is the part vendors talk about least.

Quick Reference: AI Remote Patient Monitoring at a Glance

Category	What to know
Core CPT codes (2026)	New code 99470 covers at least 10 minutes of monthly RPM management time, alongside legacy codes 99453, 99454, 99457, 99458
New lower-threshold code	CPT 99445 covers device data when only 2 to 15 days of readings are transmitted in 30 days, down from the old 16-day minimum
Related program	RTM (remote therapeutic monitoring) tracks non-physiological data like adherence and function, expanded with three new 2026 codes
Hospital-at-home status	The Acute Hospital Care at Home waiver was extended through September 30, 2030 under the 2026 Consolidated Appropriations Act
Leading hospital/health-system vendors	Biofourmis, Cadence, Current Health, Masimo, Philips, GE HealthCare
Documented outcome example	UMass Memorial Health–Harrington Hospital reported a 50% cut in 30-day heart failure readmissions using an AI-paired remote care program
Biggest operational risk	Alert fatigue: ICU alarm research found 88.8% of annotated arrhythmia alarms were false positives in one large study

What “AI” Actually Means Inside a Remote Patient Monitoring Platform

Strip away the marketing and RPM AI does three things. It looks for patterns across a stream of vitals rather than single readings, it flags deviations from a patient’s own baseline instead of a fixed population threshold, and it routes the resulting alert to a human. AI-powered monitoring platforms use predictive analytics, biometric sensors, and automated alerts to support early diagnosis and proactive care, according to market research from Fortune Business Insights. The clinical value lives almost entirely in that third step: how well the system decides which alerts actually deserve a nurse’s attention.

Vendors describe this as moving from reactive to predictive care. One recent industry write-up put it plainly: connected devices capture ongoing physiological signals while AI converts those streams into clinically meaningful insights by flagging trends and detecting anomalies early. That’s the pitch. Whether it holds up depends heavily on which vendor, which condition, and how the alert thresholds are tuned for that specific patient population, not just the technology label.

The AI Remote Patient Monitoring Tools Actually Deployed in US Health Systems

Biofourmis: Hospital-at-Home and Post-Discharge Deterioration Detection

Biofourmis is probably the most cited name in acute-care RPM. Biofourmis combines wearables, AI analytics, and clinical services for hospital-at-home, heart failure, and oncology programs, with customers including Brigham and Women’s, Mayo Clinic, and Premier. Its recent work with Lee Health illustrates the model in practice: the partnership achieved a 50% reduction in 30-day readmissions with an average daily census of more than 700 patients, one of the largest RPM programs in the country. In a separate heart failure population, the company has reported even larger effects, a 70% reduction in 30-day readmissions and 38% cost reduction by catching deterioration early, though that figure comes from company-published data rather than an independent peer-reviewed trial, so it’s worth treating as a vendor claim pending outside replication.

One caution worth flagging directly: an independent review of Biofourmis’s FDA record found that its 510(k) clearance for the Everion+ G2 wearable covers monitoring vital signs at rest and does not substantiate a cleared predictive algorithm for heart-failure hospitalizations. That distinction, cleared to measure vitals versus cleared to predict a specific outcome, matters when you’re evaluating any RPM vendor’s clinical claims. In 2025 the company merged with CopilotIQ, and the combined platform now spans pre-surgical optimization through acute, post-acute, and chronic in-home care.

Cadence: Chronic Disease Management for Hypertension, Diabetes, and COPD

Cadence takes a different angle, built around continuous chronic-condition management rather than acute hospital substitution. Its AI-powered Proactive Care Engine analyzes data in real time, alerts clinical teams to emerging health risks between visits, and supports patients in receiving medication adjustments and lifestyle guidance. It pairs the software with continuous RPM with employed clinicians for hypertension, diabetes, heart failure, and COPD, which is a meaningfully different staffing model than platforms that only sell software to a health system’s existing nurses.

Current Health, Masimo, and GE HealthCare: Vital-Sign Surveillance at Scale

For general hospital and post-acute surveillance, the bigger device makers remain heavily used. Masimo’s Patient SafetyNet and Radius VSM platforms have been deployed by health systems for centralized remote surveillance of general-care patients, alongside Philips, GE HealthCare, Baxter/Hillrom, Medtronic, Current Health, and the two vendors above. This is where a lot of hospital-at-home command centers actually run.

Why AI Remote Patient Monitoring Reimbursement Changed Meaningfully in 2026

The single biggest operational shift this year is CMS’s willingness to pay for shorter monitoring periods. The old rule required 16 days of transmitted data in a 30-day window to bill the core device code. If a patient gets between 2 to 15 readings in a calendar month, providers can now bill code 99445, while 16 or more readings still bill under 99454. That change matters because a large share of real-world patients, especially frail or elderly ones, don’t reliably transmit data every single day, and the old threshold effectively locked clinics out of billing for legitimately monitored patients.

On the time side, new CPT code 99470 covers the first 10 minutes of monthly RPM management time requiring interactive communication with the patient or caregiver, a lower bar than the previous 20-minute minimum. Legal analysts describe this as one of the most consequential rulemaking cycles for remote monitoring: the Final CY 2026 Medicare Physician Fee Schedule marks one of the most consequential rulemaking cycles for remote monitoring since the creation of the original RPM codes. The same fee schedule also broadened accepted communication modalities, with billing guidance noting broader acceptance of interactive communication modalities including audio-only calls, secure messaging, asynchronous chat, and AI-driven check-ins.

On the RTM side, the parallel program for non-vital-sign data got its own expansion. The 2026 Physician Fee Schedule delivered the most significant RTM expansion since the program launched in 2022, adding three new CPT codes that lower billing thresholds and recognize shorter monitoring durations. Practices running physical therapy adherence tracking or musculoskeletal recovery programs should check whether their current billing setup already reflects these codes, since a lot of billing software and staff training lags behind fee schedule updates by months.

Hospital-at-Home Just Got a Longer Runway, Which Changes the AI Remote Patient Monitoring Calculus

Separate from the RPM/RTM codes, the Acute Hospital Care at Home waiver, the program that lets hospitals bill inpatient rates for care delivered in a patient’s home, was on a series of short-term extensions that created real uncertainty for health systems deciding whether to invest. That changed this year. The five-year extension through September 30, 2030 under the Consolidated Appropriations Act, 2026 gives health systems regulatory certainty they did not have when making 12-month pilot decisions. As of the most recent count, 366 programs across 139 health systems were operating under the waiver.

That certainty matters for AI RPM specifically because hospital-at-home is one of the few care models where continuous AI-driven vital sign surveillance is a structural requirement, not a nice-to-have. The operational bar for participating hospitals remains unchanged, though: 24/7 RN call coverage, two daily visits, EMS coordination, clinical AI for selection and documentation, and ongoing CMS reporting all remain requirements. Not every health system has the staffing model to meet that bar even with a longer runway.

What the Outcomes Evidence Actually Shows, Including the Mixed Studies

It’s tempting to cite only the best headline numbers, but the underlying research on remote monitoring for heart failure, the most studied use case, has been genuinely mixed for over a decade. The landmark BEAT-HF randomized controlled trial tested telemonitoring plus nurse coaching against usual care across six academic health systems in California. That trial’s published results, in JAMA Internal Medicine, are frequently cited precisely because RCT-level evidence for RPM’s readmission benefit has been harder to produce than industry case studies suggest. Earlier reviews of the same literature acknowledged this directly: results of telemonitoring interventions designed to improve outcomes and reduce readmissions have been inconclusive, and patient adherence to these interventions is often low.

That context matters when a vendor cites a single-site case study. The UMass Memorial Health–Harrington result cited above is genuinely impressive, but even the outlet reporting it noted a caveat worth repeating: no accompanying peer-reviewed study was made available to verify the findings at the time of publication. Ask any vendor for their methodology, sample size, and control group before treating a readmission percentage as generalizable to your patient population.

The Risk Every Vendor Undersells: Alert Fatigue in AI Remote Patient Monitoring

This is the section clinicians should read most carefully. Continuous monitoring generates continuous alerts, and a large share of them are not clinically actionable. This isn’t a new problem invented by remote care, it’s a documented ICU problem that remote monitoring inherits and can amplify. A comprehensive observational study of ICU alarms found that 88.8% of the 12,671 annotated arrhythmia alarms were false positives, out of over 2.5 million total unique alarms recorded across a 31-day period in five adult ICUs. That same research traces back to a well-documented case where CMS investigators found that nurses not recalling hearing low heart rate alarms was indicative of alarm fatigue which contributed to a patient’s death.

In the home-monitoring context specifically, a real-world study of the Current Health platform found similarly wide variability: total alarms across 76 home-monitored patients ranged from 65 to 3,113 depending on alarm ruleset and observation frequency, and the same paper warned that pairing alarms with continuous monitoring may help clinicians recognize deterioration, but the tradeoff is an increased likelihood of false alerting and potential alarm fatigue, especially if alarm settings are not selected with the new context in mind. Separate research on implantable cardiac device monitoring found false positive rates around 59.7% in some cohorts, and broader clinical decision support research has found override rates exceeding 90% in some settings once alert volume gets high enough.

The practical takeaway isn’t to avoid AI RPM, it’s to budget real staff time for alarm threshold customization per patient rather than trusting default vendor settings, and to ask any vendor directly what their measured non-actionable alert rate is before rollout. This overlaps with the broader pattern covered in our roundup of real risks of AI in healthcare: the technology doesn’t fail loudly, it fails by generating noise that erodes trust until someone stops listening.

Where AI Remote Patient Monitoring Fits Alongside Other Clinic AI Tools

RPM rarely operates in isolation anymore. Health systems running these programs are typically layering it on top of other AI-driven workflow tools, patient outreach systems that handle enrollment and reminders (covered in our piece on AI patient engagement tools), and documentation tools that turn monitoring encounters into billable notes without adding clinician typing time (see our breakdown of AI medical scribes). Clinics evaluating whether an RPM investment will actually reduce staff burden, rather than just shifting it, should also look at the broader operational efficiency question we cover in operational efficiency in healthcare AI, since RPM alert triage is one of the clearest examples of a task that looks automated but still requires meaningful human oversight capacity.

FAQ

Does Medicare pay for AI remote patient monitoring?

Medicare pays for the RPM service itself through CPT codes, not for an “AI” feature specifically. The 2026 fee schedule finalized new CPT codes supporting shorter-duration monitoring, which makes AI-assisted, lighter-touch monitoring more billable than under the old thresholds.

What’s the difference between RPM and RTM?

RPM tracks physiological vitals like blood pressure and glucose, while RTM tracks therapy adherence, therapeutic response, and musculoskeletal or respiratory status, non-physiological data RPM codes were never built to cover.

Is hospital-at-home with remote monitoring still allowed under Medicare?

Yes. The Acute Hospital Care at Home waiver was extended through September 30, 2030 under the 2026 Consolidated Appropriations Act, giving hospitals a multi-year runway instead of repeated short-term extensions.

Does AI actually cut down false alarms in remote monitoring?

It can, but the underlying problem is stubborn. Research on ICU and remote alarm streams has found non-actionable rates as high as 88.8% for annotated arrhythmia alarms in one large study, so per-patient threshold tuning still matters more than the AI label on the box.