Industries / Health / Healthcare / Telehealth
Reaching people with poor connections
This work gets care to people whom ordinary video calls cannot reach: ships at sea, the outback, rural villages, and people with only a basic phone or a weak signal.
6
Milestones achieved, from 1929 to 2026.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1929
A radio worked by pedals
For homesteads in the Australian outback, Alfred Traeger built a radio set powered by a floor-mounted generator turned with bicycle pedals, which sent Morse code. On 19 June 1929 Gertrude Rothery, at Augustus Downs Station, about 150 miles from Cloncurry, Queensland, made the first transmission to the Flying Doctor base there. By the 1950s hundreds of sets were in use, letting people across two thirds of the continent contact the service wherever they were.
Credit
Alfred Traeger; John Flynn, founder of what is now the Royal Flying Doctor Service
1935
Doctors by radio for ships
Most merchant ships carry no doctor. In 1935 the physician Guido Guida, with the backing of Guglielmo Marconi, founded the Centro Internazionale Radio Medico (CIRM) in Rome to give medical advice by radio to ships at sea; its first recorded case, on 7 April 1935, was a stoker with a high fever aboard the cargo ship Perla. By CIRM's own count it had assisted 81,016 patients on ships by its 80th year, 2015, and it now also works by satellite, internet and video.
Credit
Guido Guida and Guglielmo Marconi
1973
Microwave links to a reservation
In STARPAHC (Space Technology Applied to Rural Papago Advanced Health Care), NASA, the US Indian Health Service and Lockheed Missiles and Space Company used microwave relays to link patients in mobile units on the Papago (now Tohono O'odham) reservation in southern Arizona with doctors at Indian Health Service hospitals in Sells and Phoenix. Arizona was one of a handful of states that let paraprofessionals such as physician assistants deliver care. NASA took part from 1973 to 1977, and the project went on without it into the 1980s.
Lessona 1996 committee of the US Institute of Medicine found that only one formal telemedicine program started before 1986 survived into the mid-1990s; most of the 1960s and 1970s projects did not outlast their federal grants.
Credit
NASA; Indian Health Service; Lockheed Missiles and Space Company; the Papago (Tohono O'odham) Nation
2014
Pregnancy advice by text
South Africa's National Department of Health launched MomConnect on 24 August 2014. Pregnant women register through USSD, a menu system that works on any mobile phone without mobile data, and then get health text messages twice a week, with a helpdesk for questions and complaints. By August 2017 it had reached more than 1.7 million pregnant women, 63% of those attending a first antenatal visit, in more than 95% of public health facilities, at about US$1 million a year.
Lessonthe cheapest channel has limits: each text is paid for and capped at 160 characters, and many sign-ups were lost when the menu session dropped before it was finished.
Credit
South Africa's National Department of Health; Peter Barron and colleagues
2022
Who used phone, not video
The US Census Bureau's Household Pulse Survey asked adults about their care from April to October 2021. In ASPE's analysis, 23.1% had used telehealth in the previous four weeks. Among those users, 72.5% of 18-to-24-year-olds and 68.8% of people earning at least $100,000 had a video visit, against 43.5% of adults aged 65 and older and 38.1% of people without a high school diploma; the rest were by phone only.
Lessoncounting only video visits misses the people who most often used the phone.
Credit
Madjid Karimi and colleagues, Office of the Assistant Secretary for Planning and Evaluation (ASPE), US Department of Health and Human Services
2026
Half a billion consultations
India's national telemedicine service, eSanjeevani, links more than 142,000 Ayushman Arogya Mandirs, local health centers that serve as "spokes", to 19,473 medical hubs where doctors take the consultation, and also runs 901 online outpatient clinics. By the Ministry of Health and Family Welfare's count it passed 500 million teleconsultations on 15 September 2026, in 15 languages, at 250,000 to 300,000 a day; 57.14% of users are women.
Credit
India's Ministry of Health and Family Welfare
Watch for
AI advice by voice call
PATH and Viamo, a Canadian social enterprise, run CHEWA, a toll-free hotline on which community health extension workers ask a generative AI model questions by voice from a basic phone, with no internet or smartphone. A trial with 100 workers in Kano and Oyo states, Nigeria, began recruiting on 26 June 2025 and runs until about 3,000 patient encounters are logged; a panel of medical experts reviews each transcript within the hour and calls the worker back if an answer was unsafe.
Achieved whenthe trial's results are published, including the share of AI answers the expert panel judged unsafe.
Claimed by
PATH, July 2025
AI help inside eSanjeevani
India's health ministry says eSanjeevani now includes an AI-powered clinical decision support system, software that suggests diagnoses and next steps, which assists providers during consultations; the system won the Gold Award under India's National Awards for e-Governance 2026. No accuracy figures were given.
Achieved whenan independent evaluation publishes how often its suggestions match expert judgment, or whether they change care.
Claimed by
Ministry of Health and Family Welfare, October 2026
Open
01
Telehealth calls on weak links
A public bench test of telehealth video services under the network conditions rural patients actually have. Run scripted visits between two devices through a network emulator set to profiles taken from open speed data (for example, median rural 4G, a congested 3G cell, and a high-latency satellite link), across services such as Zoom, Doxy.me, Microsoft Teams, WhatsApp and national services such as eSanjeevani where test access is given. Actors read standard sentences; record whether the call connects, how long video freezes, whether it falls back to audio, and how much of the speech survives.
The briefCredit · unclaimed02
Which clinics can video call
An open table that gives, for every mapped health facility in a set of countries, the measured download speed, upload speed and latency of the nearest public speed-test data, and labels each facility as able to hold a video visit, audio only, or no data. Refresh it every quarter when new data comes out.
The briefCredit · unclaimed
Open 01 · the brief
Telehealth calls on weak links
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public bench test of telehealth video services under the network conditions rural patients actually have. Run scripted visits between two devices through a network emulator set to profiles taken from open speed data (for example, median rural 4G, a congested 3G cell, and a high-latency satellite link), across services such as Zoom, Doxy.me, Microsoft Teams, WhatsApp and national services such as eSanjeevani where test access is given. Actors read standard sentences; record whether the call connects, how long video freezes, whether it falls back to audio, and how much of the speech survives.
Why it is still open
Measurement studies of video calling under constrained links cover general apps, not telehealth services: one tested Zoom, Google Meet and Microsoft Teams and found some take as long as 50 seconds to recover from a temporary drop in capacity. The only telehealth comparison found is observational: one US free clinic counted 15 connection losses on Doxy.me and 2 on Zoom across 72 visits. Neither measured whether the clinician could still understand the patient.
Built with what exists
Linux tc/netem on a small router or laptop to shape bandwidth, delay and loss; Ookla's open performance tiles to set the profiles; two phones and two laptops; screen recording; a speech-to-text model such as ElevenLabs Scribe v3 to score how many spoken words come through intact (see the Note "Fewer word errors").
First step
Build two profiles from the latest Ookla mobile tiles for one rural district, and run 10 calls each on two services.
Done looks like
At least 5 services, 6 network profiles and 10 scripted calls each, with call setup success, seconds of frozen video per minute and word error rate of the received speech published in a table, with the profiles and scripts so anyone can rerun them.
Open 02 · the brief
Which clinics can video call
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
An open table that gives, for every mapped health facility in a set of countries, the measured download speed, upload speed and latency of the nearest public speed-test data, and labels each facility as able to hold a video visit, audio only, or no data. Refresh it every quarter when new data comes out.
Why it is still open
The World Bank has done such checks country by country with Ookla's licensed Speedtest Intelligence data: in El Salvador only about 22% of 738 facilities had speeds above 10 Mbps and 89 had no test within 400 m; a study across several least developed countries found video calls limited mostly to urban areas. Neither published a facility-level table anyone can reuse. UNICEF's Global Connectivity Map began showing health facilities in September 2026, starting with The Gambia, the first country to share its facility locations; it marks each one connected or not, with no speeds given.
Built with what exists
Ookla's open fixed and mobile tiles (about 610 m square, quarterly, free for non-commercial use); facility locations from healthsites.io and OpenStreetMap; GeoPandas or DuckDB for the spatial join; published bandwidth needs of common video services for the thresholds.
First step
Join one country's healthsites.io facility points with the latest Ookla mobile tiles and publish the first table with its missing-data count.
Done looks like
At least 10 countries, every facility with a label and the test count behind it, a check against at least 50 on-site speed tests collected by volunteers, and the table and code published under an open license.