BRENNEN

Industries / Health / Healthcare / Physical therapy

Measuring progress

Measuring progress is repeating the same tests and questionnaires over the course of care, such as timed walks, balance tests and disability scores, to show whether a patient is getting better and by how much.

6

Milestones achieved, from 1989 to 2017.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1989

    How much change matters

    A score on a questionnaire can move without the patient feeling any different. Roman Jaeschke, Joel Singer and Gordon Guyatt compared score changes with patients' own ratings of how much they had changed, and called the smallest change that patients notice the minimal clinically important difference (MCID); on questionnaires about breathlessness, fatigue and mood in heart and lung disease it was about 0.5 points per item on a 7-point scale.

    europepmc.org

    Credit

    Roman Jaeschke, Joel Singer and Gordon Guyatt

  2. 1991

    A three-meter walk, timed

    In the Timed Up and Go test, a patient rises from an armchair, walks 3 meters, turns, walks back and sits down while someone times it. In 60 patients at a geriatric day hospital, mean age 79.5, the time was reliable between and within raters, tracked the Berg Balance Scale (r = -0.81) and the Barthel Index of daily activities (r = -0.78), and needed no special equipment or training.

    europepmc.org

    Credit

    D. Podsiadlo and S. Richardson, timing the Get-Up and Go test of Mathias and colleagues (1986)

  3. 2001

    A shared language for function

    The International Classification of Functioning, Disability and Health (ICF) gives codes for body functions, for the activities a person can do and for their part in daily life, along with the surroundings that help or hinder them. All 191 WHO member states endorsed it at the World Health Assembly on 22 May 2001 (resolution WHA54.21) as the international standard to describe and measure health and disability.

    who.int

    Credit

    World Health Organization

  4. 2009

    Half skip the measures

    A survey of 1,000 randomly chosen members of the American Physical Therapy Association found that 48% of those who answered used standardized outcome measures. Non-users most often blamed the time patients needed to fill them in, the time therapists needed to score them, and patients' difficulty completing them alone.

    Lessona measure that takes too long to give and score goes unused, however well it has been tested.

    europepmc.org

    Credit

    D. U. Jette and colleagues (Physical Therapy)

  5. 2011

    Walking speed predicts survival

    Pooling 9 cohort studies of 34,485 adults aged 65 or older, followed for 6 to 21 years, found that each 0.1 meter per second of extra walking speed went with a lower risk of death (hazard ratio 0.88). Walking speed, a test that needs a stopwatch and a marked floor, was associated with survival in every one of the studies.

    pmc.ncbi.nlm.nih.gov

    Credit

    Stephanie Studenski and colleagues, University of Pittsburgh and the US National Institute on Aging

  6. 2017

    Step counters miss slow steps

    166 patients in an inpatient rehabilitation unit, walking at an average 0.4 meters per second, each wore 16 activity monitors from 7 devices while a physical therapist counted their steps. The StepWatch and an ankle-worn Fitbit One agreed closely with the count (StepWatch mean 98%), while the Fitbit Charge, a hip-worn Fitbit One, Garmin Vivofit, ActiGraph and G-Sensor agreed poorly (agreement scores 0.12 to 0.40 on a 0-to-1 scale).

    Lessona step counter checked on healthy walkers cannot be trusted for slow walkers, so it has to be tested on the people it will measure.

    europepmc.org

    Credit

    D. Treacy, C. Sherrington and colleagues (Physical Therapy)

Watch for

  1. A knee that reports recovery

    Zimmer Biomet's Persona IQ knee replacement carries a sensor in its stem that records stride length, range of motion, step count and walking speed after surgery; it received FDA de novo authorization in 2021. In March 2026 the company and Hospital for Special Surgery researchers reported, from insurance claims, that 1,081 patients with the implant and the company's Mymobility app had fewer revisions at one year than 4,324 patients without them (0.3% against 1%) and fewer physical therapy visits; the company said this shows an association, not that the implant caused it.

    Achieved whena prospective study, repeated by a group independent of the company, shows that acting on the implant's gait data improves recovery; the company's own 200-patient cohort study is due to finish in December 2026.

    medtechdive.comclinicaltrials.gov

    Claimed by

    Zimmer Biomet, March 2026

  2. Scores read from notes

    A large language model run on the hospital's own computers estimated item scores on the Functional Independence Measure, a standard rehabilitation score of how independently a person manages daily tasks, from physical, occupational and speech therapy notes at Saiseikai Moriyama Municipal Hospital in Japan. Where a score from the same day existed, the model could give an estimate 33.5% of the time, and 77.7% of those estimates were within 1 point; overall its error was larger than simply carrying the last recorded score forward (0.825 against 0.648 points).

    Achieved whentested at other hospitals, the method estimates more items and beats carrying the last score forward.

    pmc.ncbi.nlm.nih.gov

    Claimed by

    T. Nishiyama, E. Aramaki and colleagues, October 2026

Open

  1. 01

    An ICF linking benchmark

    A public test of whether AI models can link the items of outcome questionnaires to ICF categories under the published linking rules. Gather the item-by-item linkings that human teams have published for many questionnaires, run several models on the same items, and score how often each model agrees with the published answer.

    The brief
    Credit · unclaimed
  2. 02

    Phone gait metrics on crutches

    A laboratory test of the walking speed, step length, double support time and walking asymmetry that iPhones record in Apple Health, when the person walks with crutches, a cane or a walker as people do after a leg fracture or joint replacement. Healthy volunteers carry phones in a pocket or on the hip while walking across a pressure-sensing walkway with each aid and at several speeds.

    The brief
    Credit · unclaimed

Open 01 · the brief

An ICF linking benchmark

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 whether AI models can link the items of outcome questionnaires to ICF categories under the published linking rules. Gather the item-by-item linkings that human teams have published for many questionnaires, run several models on the same items, and score how often each model agrees with the published answer.

Why it is still open

The nearest work, by H. Gül, M. A. Çınar and K. Bayramlar at Akdeniz University (Burns, 2025), tested ChatGPT-4o on one burn questionnaire against two physiotherapists, whose own agreement was only fair (kappa 0.41, a 0-to-1 agreement score that allows for chance). Two other tasks are nearby but different: a 2025 trial in 185 inpatients in Astana, Kazakhstan, found that the MedQuest app's AI turned five questionnaires' answers into ICF qualifiers, the 0-to-4 severity grades, in substantial agreement with clinicians (kappa 0.842), and a 2021 US National Institutes of Health team trained language-processing models to code descriptions of mobility, self-care and domestic life in disability-benefit records. Neither scores the linking of many questionnaires' items to categories. A 2026 scoping review of 42 studies of AI language models in rehabilitation found only 2 on assessment and functional classification. WHO states that the ICF may be used for the lawful training of AI systems.

avesis.akdeniz.edu.trfrontiersin.orgeprints.whiterose.ac.ukpmc.ncbi.nlm.nih.govwho.int

Built with what exists

WHO's ICF classification files; published linking studies in open-access journals; open-weight models such as Mistral Large 4 (see the Note "A trillion open parameters") and commercial models through their APIs; kappa scoring in Python.

  1. First step

    Find 10 open-access ICF linking papers that print their full item tables and copy the tables into one spreadsheet.

  2. Done looks like

    At least 30 questionnaires and 500 items with published human linkings; at least 3 models run with fixed, published prompts; each model's agreement with the published linking (kappa and share of exact matches) by model and by ICF chapter, set beside the agreement between the human linkers where the papers report it; items, prompts and outputs released.

Open 02 · the brief

Phone gait metrics on crutches

Credit · unclaimed

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

What to build

A laboratory test of the walking speed, step length, double support time and walking asymmetry that iPhones record in Apple Health, when the person walks with crutches, a cane or a walker as people do after a leg fracture or joint replacement. Healthy volunteers carry phones in a pocket or on the hip while walking across a pressure-sensing walkway with each aid and at several speeds.

Why it is still open

Apple's 2022 validation used a Zeno pressure walkway, slow walking "as if recovering from an injury" and a locked knee brace, and does not mention walking aids. An outside 2023 test in 83 children, adults and seniors used no aids either. A 2026 test with crutches, rolling walkers and oxygen trolleys found an Apple Watch undercounted steps by 36.4% with a rolling walker, but it checked step counts and distance from a watch and a waist-worn pedometer, not the phone measures. Orthopedic trauma researchers already use the Apple Health figures to follow 107 patients through recovery from leg fractures.

apple.compmc.ncbi.nlm.nih.goveuropepmc.orgpmc.ncbi.nlm.nih.gov

Built with what exists

current iPhones with Apple Health's Mobility metrics; a ProtoKinetics Zeno or GAITRite walkway in a university lab; axillary and forearm crutches, a cane and a front-wheeled walker; a locked knee brace to copy a stiff knee; Python for the comparison.

  1. First step

    Have five volunteers walk 10 bouts each with and without crutches along a measured 20-meter corridor timed by stopwatch, then check which bouts Apple Health recorded and how its walking speed compares.

  2. Done looks like

    At least 30 volunteers, 4 aid conditions and 3 speeds, with mean error and limits of agreement for each measure and condition against the walkway, and the share of walks for which the phone records any value; data and code published.