BRENNEN

Industries / Health / Healthcare / Telehealth

Video visits

A patient and a clinician see and hear each other over a live video link, from a clinic, a kiosk or the patient's own home, instead of meeting in the same room.

6

Milestones achieved, from 1964 to 2020.

1

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1964

    Psychiatry by microwave link

    The Nebraska Psychiatric Institute in Omaha opened a two-way television link, carried by microwave, to the state mental hospital in Norfolk, 112 miles away, paid for by a National Institutes of Health grant. The historian of medicine Jeremy Greene calls it "the first large-scale test of interactive television as a medical medium". Before that, on the Omaha campus, the team had a neurologist examine a patient and a psychiatrist run group psychotherapy over closed-circuit television.

    neh.govunmc.edu

    Credit

    Cecil Wittson, Reba Benschoter and the technician Van Lear Johnson, Nebraska Psychiatric Institute

  2. 1968

    An airport clinic by TV

    Massachusetts General Hospital ran a medical station at Boston's Logan Airport, staffed around the clock by nurses. From April 1968 a doctor at a console in the hospital could examine patients there through television cameras and receive their X-rays, EKGs and microscope images of blood smears. Kenneth Bird, who ran it, coined the word "telemedicine" in a 1970 grant proposal. By 1974 a Rockefeller Foundation report counted 11 active television care projects, and Greene writes that all were technical successes, yet by the early 1980s every one of these federally funded projects had closed.

    Lessonworking technology did not keep the clinics open; they ended when their federal funding did.

    neh.gov

    Credit

    Kenneth Bird, Massachusetts General Hospital

  3. 2016

    Actors test online doctors

    Researchers at the University of California, San Francisco, sent 67 trained actors, called standardized patients, to the 8 most visited commercial virtual-visit websites, where a patient is connected on demand to a doctor they have never met, by video, phone or web chat. In 599 visits for 6 common complaints, such as a sore throat or a urinary infection, doctors named the correct diagnosis 76.5% of the time and followed guidelines on the key treatment decision 54.3% of the time; guideline-adherent care ranged from 34.4% to 66.1% across the websites.

    europepmc.org

    Credit

    A. J. Schoenfeld, R. A. Dudley and colleagues

  4. 2017

    Convenience adds new visits

    Direct-to-consumer telehealth lets a patient reach an on-demand doctor by phone or video app, outside their usual practice. RAND studied more than 300,000 members of a CalPERS health plan that offered it from 2012 and found that only about 12% of telehealth visits for respiratory infections replaced a visit to a doctor's office or emergency room; 88% were new use. Net yearly spending on these illnesses rose by $45 per telehealth user, although the companies' own patient surveys had suggested only about 10% of visits were new.

    Lessona cheaper visit can raise total spending when it adds visits instead of replacing them.

    rand.org

    Credit

    J. Scott Ashwood and colleagues, RAND Corporation

  5. 2019

    More antibiotics for children

    Researchers matched children's visits for acute respiratory infections in 2015–2016 claims from a large national commercial health plan. Antibiotics were prescribed at 52% of direct-to-consumer telemedicine visits, against 42% at urgent care and 31% at primary care offices, and antibiotic decisions matched guidelines least often at telemedicine visits (59%, against 67% and 78%).

    Lessonquick online visits for minor illness can lower prescribing quality, so they need the same checks as office visits.

    europepmc.org

    Credit

    K. N. Ray, A. Mehrotra and colleagues

  6. 2020

    Telehealth visits rise 63-fold

    In 2020, during the COVID-19 pandemic, Medicare fee-for-service telehealth visits rose from about 840,000 in 2019 to nearly 52.7 million, from under 1% of visits under Part B (the part of Medicare that pays for doctor visits) to 5.3%. Of the beneficiaries who used telehealth, 92% did so from home, and telehealth made up a third of visits to behavioral health specialists.

    aspe.hhs.gov

    Credit

    Office of the Assistant Secretary for Planning and Evaluation (ASPE), US Department of Health and Human Services, which measured it

Watch for

  1. An online hospital for England

    NHS England says NHS Online, which it calls the NHS's online hospital, will let family doctors refer patients to specialists anywhere in the country for video consultations, with triage through the NHS App and monitoring at home. It will start with nine common conditions, including severe menopause symptoms and prostate problems, and is expected to deliver the equivalent of up to 8.5 million virtual appointments and assessments in its first three years.

    Achieved whenNHS England's published activity data show appointments at that scale, and an independent evaluation compares NHS Online's waiting times and outcomes with ordinary hospital outpatient clinics.

    htn.co.uk

    Claimed by

    NHS England, January 2026, ahead of a planned 2027 launch

Open

  1. 01

    Skin seen through video calls

    A public test of how much a video call degrades the look of skin conditions, on light and dark skin. Take consumer photos of skin conditions, send them through the same compression that video calls use at fixed bitrates (the amount of data sent each second) and with packet loss (data lost on the way), and have dermatologists diagnose the original and the degraded versions. A second arm can film the images on a calibrated screen with two phone cameras, to add the camera's own exposure and color handling.

    The brief
    Credit · unclaimed
  2. 02

    Score live AI interpreters

    A public test of live speech translation in staged video visits. Actors play patients who speak six or more languages, including some that clinics rarely staff interpreters for, and say scripted facts that matter clinically: a dose, an allergy, when a symptom began, a warning sign. A clinician speaks English. Each script runs through several live translation tools and through a certified human interpreter, and certified medical interpreters grade the transcripts blind, counting omissions, additions, substitutions and editorializing, and judging which errors could change care.

    The brief
    Credit · unclaimed

Open 01 · the brief

Skin seen through video calls

Credit · unclaimed

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

What to build

A public test of how much a video call degrades the look of skin conditions, on light and dark skin. Take consumer photos of skin conditions, send them through the same compression that video calls use at fixed bitrates (the amount of data sent each second) and with packet loss (data lost on the way), and have dermatologists diagnose the original and the degraded versions. A second arm can film the images on a calibrated screen with two phone cameras, to add the camera's own exposure and color handling.

Why it is still open

A 2017 study compared dermatology exams by uncompressed video at almost 1.5 gigabits per second with H.264-compressed video at about 2 megabits per second, and its authors wrote that "the amount of video compression that can be applied without noticeable image degradation is a question for future research". A search found no study that measures diagnostic agreement against video bitrate, or that breaks it down by skin tone.

pmc.ncbi.nlm.nih.gov

Built with what exists

Google's SCIN dataset, more than 10,000 skin photos donated by US internet users, with dermatologist labels and estimated Fitzpatrick and Monk skin tones, under a license that allows adapted copies with attribution and forbids re-identification; ffmpeg with the VP8, H.264 and AV1 codecs used in browser video calls; Linux netem to add packet loss and rate limits; three or more board-certified dermatologists as raters; a vision-language model such as Gemini or Claude as an extra rater.

github.comman7.org

  1. First step

    Download SCIN, choose 60 cases spread across skin tones, encode them at 4 bitrates with ffmpeg, and have one dermatologist rate them to size the effect before recruiting more raters.

  2. Done looks like

    300 SCIN cases balanced across skin-tone groups, each shown as the original and at 6 bitrates from 150 kilobits to 4 megabits per second, with and without 2% packet loss; each dermatologist's diagnosis compared with their own diagnosis from the original; the bitrate at which agreement falls by 5 percentage points, reported for each skin-tone group with 95% confidence intervals; code, settings and degraded clips published under the SCIN license.

Open 02 · the brief

Score live AI interpreters

Credit · unclaimed

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

What to build

A public test of live speech translation in staged video visits. Actors play patients who speak six or more languages, including some that clinics rarely staff interpreters for, and say scripted facts that matter clinically: a dose, an allergy, when a symptom began, a warning sign. A clinician speaks English. Each script runs through several live translation tools and through a certified human interpreter, and certified medical interpreters grade the transcripts blind, counting omissions, additions, substitutions and editorializing, and judging which errors could change care.

Why it is still open

Published tests are small, cover one tool at a time, and mostly cover English and Spanish. A 2025 study ran 10 simulated urgent care visits in text only, with ChatGPT 4.0 playing both patient and interpreter; it found 23 errors in 3,597 words, 3 of them with possible clinical consequences, and its authors called for tests with speech. A March 2026 preprint by Iris Feinberg and colleagues ran 14 doctors' statements from recorded visits through one AI voice app and through certified interpreters into seven languages: clinical errors were 33.3% with the app against 4.8% with interpreters, and higher in less common languages. It tested one app and only the doctor's side, outside a video visit. A 2026 study of simulated English–Spanish visits found that AI scribes copied interpreter errors into the clinical note. A benchmark by the interpreting company Boostlingo covers English and Spanish and was run by the company itself. No test of several live translation tools, in both directions, in staged video visits was found.

europepmc.orgeuropepmc.orgeuropepmc.orgboostlingo.com

Built with what exists

Google Meet's speech translation, which since 22 October 2026 translates speech between any of 63 languages in near real time on eligible paid plans; a speech-to-text model such as ElevenLabs Scribe v3 to produce transcripts for grading (see the Note "Fewer word errors"); volunteer actors who are native speakers; two certified medical interpreters per language as graders.

workspaceupdates.googleblog.com

  1. First step

    Write 5 English visit scripts with 10 critical facts each, have a certified interpreter translate them into two languages, record one actor per language, and run them through Google Meet's speech translation.

  2. Done looks like

    6 languages, 20 scripted visits per language, each run through at least 3 tools and one human interpreter; errors per 100 words and the share of the scripted critical facts translated wrongly, each with 95% confidence intervals and the two graders' agreement; scripts, audio and grades published.