BRENNEN

Industries / Health / Healthcare / Telehealth

Remote monitoring

Remote monitoring sends readings from a patient's body or home, such as heart rhythms, blood pressure, blood sugar, oxygen levels and data from heart implants, to a care team that watches them between visits.

7

Milestones achieved, from 1906 to 2023.

3

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1906

    Heart signals sent by wire

    Willem Einthoven, who built the electrocardiograph in Leiden, linked the laboratory that housed it to a hospital about 1.5 km away by telephone cable and recorded patients' heart traces at a distance. E. M. Vilela and N. Bettencourt, writing in 2024, call this the first "telecardiogram" and date the work to 1906.

    pmc.ncbi.nlm.nih.gov

    Credit

    Willem Einthoven

  2. 2010

    Daily phone check-ins fail

    The Tele-HF trial enrolled 1,653 patients recently in hospital for heart failure at 33 US cardiology practices. Half were asked to make a free call every day to an interactive voice response system, an automated phone line that asked about symptoms and weight. Readmission or death within 180 days was 52.3% with the system and 51.5% without. 14% of the monitored group never used it, and among those who did, the share calling at least three times a week fell from 90.2% in the first week to 55.1% by week 26.

    Lessoncollecting symptoms and weight every day did not by itself keep people out of hospital.

    pmc.ncbi.nlm.nih.gov

    Credit

    Sarwat Chaudhry, Harlan Krumholz and colleagues; funded by the US National Heart, Lung, and Blood Institute

  3. 2011

    A sensor in the artery

    CardioMEMS is a small wireless pressure sensor implanted in the pulmonary artery, the vessel that carries blood from the heart to the lungs; patients take a reading at home every day and doctors adjust drugs from the numbers. In the CHAMPION trial of 550 heart failure patients, the sensor group had 28% fewer heart failure hospital stays at six months and 37% fewer over a mean of 15 months (158 against 254). The US Food and Drug Administration (FDA) approved it in 2014.

    Lessona larger trial, GUIDE-HF, with 1,000 patients, did not meet its main goal (hazard ratio 0.88, p = 0.16); only an analysis of the period before COVID-19 came out borderline in the sensor's favor.

    pmc.ncbi.nlm.nih.govpmc.ncbi.nlm.nih.gov

    Credit

    W. T. Abraham and the CHAMPION investigators

  4. 2015

    Sugar readings shared with others

    A continuous glucose monitor (CGM) is a worn sensor that tracks blood sugar through the day. On 23 January 2015 the FDA allowed marketing of the first set of mobile apps that let a person with diabetes share CGM readings with chosen "followers" on their own phones: Dexcom's Share system. The FDA also ruled that similar products would no longer need its review before sale, only registration.

    healio.com

    Credit

    Dexcom; the FDA's Office of In Vitro Diagnostics and Radiological Health, led by Alberto Gutierrez

  5. 2018

    Home readings guide the dose

    The TASMINH4 trial randomized 1,182 people with high blood pressure in 142 UK general practices. Doctors adjusted drugs using clinic readings, readings patients took at home, or home readings sent by a free text-message service that flagged very high or low values. After 12 months systolic pressure was 3.5 mmHg lower with home readings and 4.7 mmHg lower with texted readings than with clinic readings; the difference between the two home groups (1.2 mmHg) was not significant.

    Lessonthe gain came from doctors using home readings to adjust drugs; sending the readings electronically added little that could be measured.

    pmc.ncbi.nlm.nih.gov

    Credit

    Richard McManus and the TASMINH4 investigators; funded by the UK National Institute for Health Research, with monitors from Omron Healthcare UK

  6. 2020

    Oximeters miss low oxygen

    A pulse oximeter clips on a finger and estimates blood oxygen from light passed through the skin. A University of Michigan team compared 10,789 oximeter readings with arterial blood tests in its own hospital and 37,308 in intensive care units at 178 hospitals. Among patients whose oximeter read 92% to 96%, the blood test showed dangerously low oxygen (below 88%) in 11.7% of Black patients and 3.6% of White patients at Michigan, and in 17.0% and 6.2% in the larger set. The FDA later issued a safety communication (2021) and, on 6 January 2025, draft guidance on testing oximeters across skin tones.

    Lessona home reading that looks safe may not be safe if the device was tested mostly on light skin.

    pmc.ncbi.nlm.nih.govfda.gov

    Credit

    Michael Sjoding, Thomas Valley and colleagues

  7. 2023

    Implants report home daily

    Pacemakers and implanted defibrillators can send their data from the patient's home to the clinic. A pooled analysis of three trials of Biotronik's Home Monitoring, which transmits daily, covered 2,405 patients and found the one-year risk of death 1.9 percentage points lower with it. In 2023 the heart rhythm societies of the US, Europe, Asia Pacific and Latin America stated that remote monitoring is "the standard of care" for patients with these devices.

    Lessonsuccess moved the work to the clinic: some clinics receive more than 100,000 transmissions a year, and unscheduled ones make up 27% to 40% of the workload.

    pmc.ncbi.nlm.nih.govpmc.ncbi.nlm.nih.gov

    Credit

    Gerhard Hindricks, Niraj Varma and colleagues (analysis, funded by Biotronik); Heart Rhythm Society and partner societies

Watch for

  1. Oximeters tested across skin tones

    Philips says its new pulse oximeter, cleared by the FDA, has an accuracy root mean square (ARMS, the typical error against arterial blood tests) of 1.6% in laboratory studies in which volunteers' oxygen levels are lowered under control, and that its accuracy differs by less than 0.5% between people with light and dark skin. The FDA's January 2025 draft guidance, which would require testing across skin tones and a public list of devices that meet it, was still not final in July 2026.

    Achieved whenan independent laboratory reproduces the figures, as UCSF's Hypoxia Research Laboratory did for 34 other oximeters in 2025 (it found 1 of 34 passed the FDA's proposed criteria), or the FDA finalizes the guidance and lists the device.

    medtechdive.compmc.ncbi.nlm.nih.gov

    Claimed by

    Philips, July 2026

  2. Watches that flag hypertension

    Apple's hypertension notification feature reads the watch's sensor data over 30 days and alerts adults with no diagnosis who may have high blood pressure. The FDA cleared it on 11 September 2025 on Apple's own study: of 2,229 people enrolled, 1,863 were analyzed against a home blood pressure cuff, and the feature caught 41.2% of those with hypertension (53.7% of the more severe stage 2 cases) while correctly clearing 92.3% of those without.

    Achieved whena study run by people outside Apple, in an ordinary population, reports how many cases the alerts catch and miss.

    accessdata.fda.gov

    Claimed by

    Apple, September 2025

  3. An AI agent sorts readings

    AnsibleHealth says its AI agent Sentinel, which uses 21 software tools to gather each patient's context, sorted remote monitoring readings into four levels of urgency with 95.8% sensitivity for emergencies and 85.7% specificity, against the majority vote of clinicians, on 467 readings, at a median cost of $0.34 a reading. The paper does not say whether the readings are public.

    Achieved whena group with no stake in the agent tests it on readings it did not choose, with results published.

    arxiv.org

    Claimed by

    Seunghwan Kim, Ming Jack Po and colleagues, from AnsibleHealth, Stanford School of Medicine and George Washington University, on arXiv, March 2026

Open

  1. 01

    Readings through a dead zone

    A bench test of at least six remote monitoring kits sold to clinics, such as cellular blood pressure cuffs, cellular scales and Bluetooth cuffs that send through a phone or hub. Feed each a known input, cut its connection for set periods, and compare what reaches the clinician portal with what was measured: readings lost, duplicated, delayed or stamped with the wrong time.

    The brief
    Credit · unclaimed
  2. 02

    A home-readings triage key

    A public set of at least 1,000 synthetic remote monitoring readings, each with its context: an invented patient's conditions and medicines, and 30 days of home blood pressure, weight, pulse, oxygen and blood sugar, with deteriorations planted on purpose, such as a heart failure patient gaining 2 kg in three days. Three clinicians who run monitoring programs each rate every reading on a four-level urgency scale (no action, routine, same day, emergency); then AI agents and ordinary threshold alarms are scored against them.

    The brief
    Credit · unclaimed

Open 01 · the brief

Readings through a dead zone

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A bench test of at least six remote monitoring kits sold to clinics, such as cellular blood pressure cuffs, cellular scales and Bluetooth cuffs that send through a phone or hub. Feed each a known input, cut its connection for set periods, and compare what reaches the clinician portal with what was measured: readings lost, duplicated, delayed or stamped with the wrong time.

Why it is still open

Makers describe store-and-forward behavior but publish no loss figures. Smart Meter's brochure says its cellular cuff saves up to 90 readings "only when transmission fails" and sends them with the next successful one, and claims, without a source, that "70% of people have trouble sending health data with Bluetooth devices". A search found vendor material and opinion pieces, but no independent test of how many readings arrive.

smartmeterrpm.com

Built with what exists

a non-invasive blood pressure simulator such as Fluke Biomedical's ProSim 8 to give each cuff the same "patient"; certified test weights for scales; a shielded RF enclosure or Faraday bag to cut cellular signal; Linux tc/netem on a test router for Wi-Fi and hub kits; each maker's clinician portal or API; a short script to log every reading at both ends.

  1. First step

    Buy one cellular and one Bluetooth cuff, borrow a blood pressure simulator from a hospital biomedical engineering department, and run 50 readings with a 24-hour outage in a Faraday bag.

  2. Done looks like

    For each kit, 200 scripted readings across five outage patterns (none, 1 hour, 24 hours, 72 hours, and longer than the device's stated memory), with the share delivered intact, median and 95th-percentile delay, and counts of duplicates and wrong time stamps, published with the scripts and logs.

Open 02 · the brief

A home-readings triage key

Credit · unclaimed

The first to publish is credited on this line, and the deliverable moves from Open to Achieved.

What to build

A public set of at least 1,000 synthetic remote monitoring readings, each with its context: an invented patient's conditions and medicines, and 30 days of home blood pressure, weight, pulse, oxygen and blood sugar, with deteriorations planted on purpose, such as a heart failure patient gaining 2 kg in three days. Three clinicians who run monitoring programs each rate every reading on a four-level urgency scale (no action, routine, same day, emergency); then AI agents and ordinary threshold alarms are scored against them.

Why it is still open

The best-reported AI agent for this work, AnsibleHealth's Sentinel, was tested by its own team on 467 readings that its paper does not say are public (see Watch for). A search for open monitoring data found synthetic sets for intensive care units and for falls at home, but none of home readings labeled for urgency.

arxiv.org

Built with what exists

Synthea, an open-source generator of synthetic patient records, for the invented patients; a short script that writes daily readings and plants events; three monitoring clinicians as raters; language models such as Claude, Gemini or Mistral Large 4 as agents under test, each made to answer with one of the four levels (see the Note "A thousand questions, one pass"; for the open-weight model, the Note "A trillion open parameters").

  1. First step

    Write 50 synthetic heart failure patients with 30 days of readings each, plant 10 deteriorations, and have one monitoring nurse rate them.

  2. Done looks like

    1,000 readings, 200 of them with planted deteriorations; the raters' agreement with each other published; each tool's sensitivity for emergencies and for all readings needing action, and its false escalations per 100 readings, with 95% confidence intervals; data and scripts released under an open license.