BRENNEN

Industries / Health / Healthcare / Physical therapy

Exercise at home

Exercise at home is the part of treatment a patient does alone between visits, from a printed sheet, a video call with a therapist or an app that sets and checks the exercises.

6

Milestones achieved, from 1997 to 2024.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1997

    Home exercise prevents falls

    In Dunedin, New Zealand, a physiotherapist visited 116 women aged 80 and over at home to set them strength and balance exercises tailored to each woman, later known as the Otago Exercise Programme; 117 others got usual care and social visits. Over a year the exercise group had 88 falls against 152, a rate of 0.87 falls a year against 1.34, and fewer falls with injury.

    pmc.ncbi.nlm.nih.gov

    Credit

    A. J. Campbell, M. C. Robertson and colleagues, University of Otago (BMJ, 1997)

  2. 2010

    Nobody counts adherence well

    Adherence means how closely a patient does what was agreed, here the exercises set for home. A review of 20 high-quality studies in musculoskeletal outpatient physiotherapy found that people who were inactive, depressed, anxious, short of support or in more pain during exercise were less likely to keep at their exercises. It also found that the size of the problem was unclear: one study put non-adherence as high as 70%.

    Lessonan exercise that is prescribed but not done cannot be judged, and the field still lacks a standard way to measure what is done.

    pmc.ncbi.nlm.nih.gov

    Credit

    Kirsten Jack and colleagues, Hull, England (Manual Therapy, 2010)

  3. 2015

    Video calls replace home visits

    Telerehabilitation is rehabilitation given over a video link. In Quebec, 205 people were randomly given either telerehabilitation sessions at home or home visits by a therapist for two months after a total knee replacement, with the same exercises. Four months after discharge their scores on the WOMAC, a standard questionnaire on knee pain, stiffness and function, differed by less than 2%, inside the 9% margin the trial set in advance.

    europepmc.org

    Credit

    Hélène Moffet, M. Tousignant and colleagues, Université Laval and Université de Sherbrooke (Journal of Bone and Joint Surgery, 2015)

  4. 2021

    Balance app, two years on

    StandingTall is a home balance exercise program for older people delivered by a tablet app, without a therapist present. In a trial of 503 people aged 70 and over in Sydney, fall rates did not differ after one year, but over two years the app group had 16% fewer falls and 20% fewer people with an injurious fall. At 12 months 68.1% were still exercising for a median of 114 minutes a week.

    Lessonthe benefit showed only in the second year, so shorter trials would have called it a failure.

    pmc.ncbi.nlm.nih.gov

    Credit

    K. Delbaere and colleagues, Neuroscience Research Australia (BMJ, 2021)

  5. 2022

    An exercise app on prescription

    Under Germany's DiGA system for apps on prescription, ViViRA, a back-pain app that sets four exercises a day and adjusts them after each session, was listed provisionally in October 2020 and permanently in February 2022 by the federal medicines regulator, BfArM. In its trial of 213 people, pain fell by 3.35 points on a 0 to 10 scale over 12 weeks, against 0.91 with standard physiotherapy; several authors worked for the company, and they asked for independent evaluation.

    bundesgesundheitsministerium.depmc.ncbi.nlm.nih.gov

    Credit

    ViViRA Health Lab, Berlin; BfArM

  6. 2024

    Virtual physical therapy assessed

    The Peterson Health Technology Institute, a US nonprofit, reviewed more than 2,000 publications on virtual musculoskeletal programs, in which patients exercise at home with an app and often a remote physical therapist. It found that programs guided by a physical therapist gave gains in pain and function comparable to in-person physical therapy, and did better than programs without one. It estimated savings of $4.4 million per 1 million commercially insured people if a quarter of those using physical therapy for low back pain moved to digital care.

    apta.org

    Credit

    Peterson Health Technology Institute, June 2024, as reported by the American Physical Therapy Association

Watch for

  1. An AI physiotherapy clinic

    Flok Health's app gives back-pain patients same-day treatment from an AI physiotherapist, built from video of a human one; the service is approved by England's Care Quality Commission and certified as a medical device. Over 12 weeks from February 2025, more than 2,500 patients in Cambridgeshire and Peterborough used it; the company and the NHS trust say it cut the back-pain waiting list by 55%, fewer than 2% needed referral to face-to-face care, and 80% rated it equivalent to or better than physiotherapy in person.

    Achieved whenan independent evaluation compares its patients' pain, function and missed serious conditions with those of patients who saw a physiotherapist.

    healthcare-management.uk

    Claimed by

    Flok Health and Cambridgeshire Community Services NHS Trust, July 2025

  2. Fewer operations after app care

    Using a US nationwide insurance claims database, Sword Health's researchers matched 4,190 adults with spine, knee, hip or shoulder problems who started Sword's home program or in-person physical therapy. They report that the app group had a 58% lower relative risk of any musculoskeletal surgery within 12 months and an 82% lower risk of low-value surgery, operations that evidence suggests do little good. The company itself notes the study is retrospective and shows an association, not a cause.

    Achieved whena randomized trial or an analysis by a group with no tie to the company finds a similar fall in surgery.

    swordhealth.com

    Claimed by

    Sword Health, in the Journal of Medical Internet Research, December 2025

Open

  1. 01

    Tested home exercise instructions

    A public test of how well people perform common home rehabilitation exercises from different kinds of instructions: a physiotherapist's illustrated sheet, an AI-written sheet, an AI-made video, and each of these translated by AI into at least three other languages. Healthy adult volunteers do each exercise from one kind of instruction on camera, and two physiotherapists who do not know which instruction was used score each attempt.

    The brief
    Credit · unclaimed
  2. 02

    Exercise class by phone call

    A phone line that any basic mobile phone or landline can call, where an AI voice agent leads a fixed, seated strength and balance session written by a physiotherapist, in at least three languages. The agent keeps to the script, answers only questions the script covers, sends anything else to a person, and logs what the caller says they did.

    The brief
    Credit · unclaimed

Open 01 · the brief

Tested home exercise instructions

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 well people perform common home rehabilitation exercises from different kinds of instructions: a physiotherapist's illustrated sheet, an AI-written sheet, an AI-made video, and each of these translated by AI into at least three other languages. Healthy adult volunteers do each exercise from one kind of instruction on camera, and two physiotherapists who do not know which instruction was used score each attempt.

Why it is still open

The method is tested: in 2016 D. J. Berkoff and colleagues randomly gave 58 healthy adults either a video showing four shoulder exercises and the common errors, or one session with a physical therapist and a handout, and blinded scorers of filmed attempts a week later found no difference. That study used human-made instructions in one language. The closest work with AI, a 2026 trial at a hospital in Ankara, Türkiye, gave 159 patients either videos made with Google's Veo 3 or standard leaflets for four foot stretches; adherence was similar (71% against 73%) and the video group did the more complex stretches more accurately. Studies of AI-written exercise programs rate the text against guidelines instead of watching people follow it, and searches found no public set of filmed, scored attempts across instruction types and languages.

pmc.ncbi.nlm.nih.goveuropepmc.org

Built with what exists

Claude, GPT or Gemini to write and translate sheets; Google Veo 3 or an open video model for demonstration clips; phone cameras; a three-point scoring sheet; Label Studio for blinded scoring.

  1. First step

    Take 10 exercises from a public leaflet set, make the four instruction types for each, and pilot with six volunteers.

  2. Done looks like

    60 volunteers, 20 low-risk exercises (such as heel slides, bridges, clamshells and calf raises) and four instruction types balanced across people, giving at least 2,400 scored attempts; agreement between the two scorers reported; the share done correctly for each instruction type and language with 95% confidence intervals; consented videos and scores released.

Open 02 · the brief

Exercise class by phone call

Credit · unclaimed

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

What to build

A phone line that any basic mobile phone or landline can call, where an AI voice agent leads a fixed, seated strength and balance session written by a physiotherapist, in at least three languages. The agent keeps to the script, answers only questions the script covers, sends anything else to a person, and logs what the caller says they did.

Why it is still open

The WHO says more than half of people in some low- and middle-income countries do not get the rehabilitation they need, and many such countries have fewer than 10 skilled practitioners per million people. Automated phone coaching is not new: in a 2011 trial by David Sparrow and colleagues at the VA Boston Healthcare System, 103 middle-aged and older adults who followed a home strength program through an interactive voice response system, which plays recorded prompts and takes simple answers, gained knee strength and single-leg balance over 12 months against a control group. That system could not hold a conversation. Newer work used smart speakers with screens: a 12-week Australian study ran home exercise for 15 people aged 60 to 89 through Amazon's Echo Show 5, and its participants named voice-recognition errors as a barrier. Searches found no test of an AI agent delivering exercise over an ordinary phone call.

who.intpmc.ncbi.nlm.nih.govpmc.ncbi.nlm.nih.gov

Built with what exists

Twilio Programmable Voice or the open-source Asterisk exchange; speech-to-text and text-to-speech models; an AI model held to the written script; ElevenLabs Scribe v3 to transcribe the recorded calls for scoring (see the Note "Fewer word errors").

  1. First step

    Record a physiotherapist reading the session script in one language, and build a call flow that plays it with pauses and records the caller's replies.

  2. Done looks like

    30 healthy adult volunteers, 10 per language, each doing three filmed calls. By transcript audit the agent gives at least 95% of script steps correctly and answers at least 90% of 20 scripted questions as the physiotherapist's answer key does, declining the rest; a physiotherapist watching the video rates at least 80% of exercises as done correctly; the call log matches the video count of exercises done in at least 90% of calls; code and anonymized transcripts published.