BRENNEN

Industries / Health / Healthcare / Physical therapy

Assessment

Assessment is the first examination: measuring how far joints move and how strong muscles are, watching how a person walks and moves, checking for signs of serious disease that need a doctor, and judging who is likely to recover and what treatment fits.

7

Milestones achieved, from 1912 to 2023.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1912

    Strength graded against gravity

    Wilhelmine Wright described a way to grade muscle strength by hand during the polio outbreaks: a muscle that can move a limb only with gravity taken out of the way scores lower than one that can lift it against gravity, and higher grades need added resistance. Robert Lovett expanded the method in 1916, Florence and Henry Kendall refined the testing positions in the 1940s, and by Fisher and Harrington's 2015 account manual muscle testing is still the mainstay of strength assessment in physical therapy and medical schools.

    ecommons.udayton.edu

    Credit

    Wilhelmine Wright; Robert W. Lovett; Florence and Henry Otis Kendall

  2. 2005

    Therapists diagnose like surgeons

    At Keller Army Community Hospital, US Army physical therapists examine patients with muscle and joint injuries first and may order MRI scans. A review of 560 patients sent for MRI over 18 months found the clinical diagnosis matched the scan for 74.5% of the therapists' patients, 80.8% of orthopedic surgeons' and 35.4% of other doctors'; the therapists and surgeons did not differ significantly.

    europepmc.org

    Credit

    John H. Moore and colleagues, US Army

  3. 2008

    A prediction rule fails

    A clinical prediction rule is a short checklist of findings meant to pick out the patients a treatment will help. A 2004 US trial of 131 people with low back pain reported that those meeting at least 4 of 5 criteria, such as how long the pain had lasted and how far the hips could rotate, did far better with spinal manipulation; a University of Sydney team then tested the rule in 239 patients from general practice and found it did no better than chance.

    Lessona rule found and confirmed by the same group of clinics needs testing by others, in other patients, before it guides care.

    europepmc.orgpmc.ncbi.nlm.nih.gov

    Credit

    John Childs, Julie Fritz, Timothy Flynn, Anthony Delitto and colleagues (2004); Mark Hancock, Christopher Maher and colleagues (2008)

  4. 2011

    Sorting back pain by risk

    The STarT Back tool is a short questionnaire that sorts people with back pain into low, medium or high risk of lasting disability, each matched to a level of care: minimal treatment for low risk, physiotherapy for medium risk, and physiotherapy with added psychological methods for high risk. In a trial of 851 patients from 10 general practices in England, sorted care reduced disability more than usual best practice at 12 months (by 1.06 points on the Roland Morris disability questionnaire) and cost less.

    Lessonwhen the US TARGET trial sent high-risk patients in 77 primary care clinics to similar therapy, 47% still had chronic pain at 6 months against 51% with usual care, a difference that was not significant, so a tool that works in one health system has to be retested in another.

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

    Credit

    Jonathan Hill, Elaine Hay and colleagues, Keele University

  5. 2013

    Most red flags fail

    Red flags are warning signs in a patient's story or examination that may point to a serious cause, such as a fracture or cancer, behind back pain. A review of 14 studies covering 53 red flags found that many listed in guidelines barely changed the chance of fracture or cancer, or had never been tested; the more useful ones were older age, long use of steroid drugs, severe injury and a bruise or graze for fracture, and a past cancer for spinal cancer (33% chance after the finding).

    Lessonwarning-sign lists in guidelines need their accuracy measured, not passed on by habit.

    pmc.ncbi.nlm.nih.gov

    Credit

    Aron Downie, Christopher Maher and colleagues, University of Sydney

  6. 2019

    Gait lab changes little

    Instrumented gait analysis measures walking in three dimensions with cameras, skin markers and force plates. In a Danish trial, 60 children with cerebral palsy aged 5 to 9 who could walk on their own were randomly given treatment plans built on gait analysis or usual care; after 52 weeks there was no significant or clinically relevant difference in gait, walking or reported function.

    Lessona more detailed measurement only helps if it leads to a different treatment, and here most recommendations were for non-surgical care either way.

    europepmc.org

    Credit

    H. M. Rasmussen, A. Holsgaard-Larsen and colleagues

  7. 2023

    Motion capture by phone

    OpenCap is free, open-source software that estimates 3D joint movement, and the forces behind it, from videos taken on two or more smartphones. Its Stanford authors report that in a 100-person field study a clinician measured movement 25 times faster than with a laboratory system, at under 1% of the cost. Outside groups have since checked it against marker-based labs: with 31 users of above-knee prostheses it matched the lab on step length and timing (agreement scores above 0.99 on a 0-to-1 scale) but its joint angles needed improvement, and a St. Pölten University team that filmed 240 drop jumps said it cannot yet be recommended for judging the risk of re-tearing a knee ligament.

    pmc.ncbi.nlm.nih.goveuropepmc.orgpmc.ncbi.nlm.nih.gov

    Credit

    Scott Uhlrich, Scott Delp and colleagues, Stanford University; W. Niamsang and colleagues; Bernhard Färber, Brian Horsak and Florian Paternoster

Watch for

  1. Digital triage before physio

    Phio, from the company EQL, gives people with joint, muscle and back pain a digital assessment at any hour and sends them to self-management or onward care. After a pilot that began on 18 March 2025, NHS Shetland reported that 512 people used it, 74% were judged suitable for self-management and 75% needed no further musculoskeletal treatment, and it extended the service for two years.

    Achieved whenan evaluation not run by the company or the health board compares Phio's decisions with physiotherapists' decisions on the same patients and reports how many serious conditions it missed.

    nhsshetland.scot

    Claimed by

    NHS Shetland and the Shetland Health and Care Partnership, April 2026

  2. A handheld phone gait lab

    The Portable Biomechanics Laboratory, from Shirley Ryan AbilityLab and Northwestern University, fits a body model to video from a single handheld, moving phone. Its authors report median joint-angle errors under 3 degrees against marker-based motion capture across more than 15 hours of data from people with neurological injuries, prosthesis users, children in hospital and controls, and consistent gait measures (agreement scores above 0.9 on a 0-to-1 scale) across 1,021 clinic videos; they released the method for others to use.

    Achieved whena team outside the authors' group repeats the comparison with its own motion capture lab and reports errors in the same range.

    arxiv.orgpmc.ncbi.nlm.nih.gov

    Claimed by

    J. D. Peiffer, R. James Cotton and colleagues, July 2025 (preprint), published in npj Digital Medicine in 2026

Open

  1. 01

    Gait capture with walking aids

    A controlled test of how canes, crutches and wheeled walkers change the accuracy of gait measurement from phone video. Healthy adult volunteers walk with no aid and then with each aid while phones film them and a marker-based lab or pressure-sensing walkway records the reference, and the errors in walking speed, step length and hip and knee angles are reported for each aid and each method.

    The brief
    Credit · unclaimed
  2. 02

    A living prediction-rule register

    An open, searchable register of every clinical prediction rule published for physical therapy, stating for each one how far it has got: derived only, checked by its own authors, tested by an outside group, or failed an outside test, with links to each study. An AI agent searches new papers each month and proposes changes that a person approves.

    The brief
    Credit · unclaimed

Open 01 · the brief

Gait capture with walking aids

Credit · unclaimed

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

What to build

A controlled test of how canes, crutches and wheeled walkers change the accuracy of gait measurement from phone video. Healthy adult volunteers walk with no aid and then with each aid while phones film them and a marker-based lab or pressure-sensing walkway records the reference, and the errors in walking speed, step length and hip and knee angles are reported for each aid and each method.

Why it is still open

The Portable Biomechanics Laboratory authors attribute their slightly higher errors in patients to assistive devices such as canes, walkers, ankle braces and gait belts, but their patients each used whatever aid they normally used, so the effect of each aid was not measured on its own. The outside OpenCap tests under Achieved studied people walking without aids. A Europe PMC search in November 2026 found no study that changed the aid while keeping the same walkers.

pmc.ncbi.nlm.nih.gov

Built with what exists

OpenCap on two iPhones; the Portable Biomechanics Laboratory's released code for a single handheld phone; Google's MediaPipe Pose as a simple single-camera baseline; a university gait lab with marker-based motion capture or a GAITRite or Zeno pressure walkway; a standard cane, forearm crutches, a front-wheeled walker and a four-wheeled rollator.

  1. First step

    Borrow one cane and one walker, film five volunteers walking a taped 10-meter course with OpenCap and with a single phone, and compare each method's walking speed with a stopwatch.

  2. Done looks like

    At least 20 volunteers, 5 conditions and 6 walks each, with mean error and limits of agreement for each method and aid on walking speed, step length and peak hip and knee bending; a stated result for whether any aid adds more than 2 degrees or 2 centimeters of error over walking without one; consented videos, reference data and code published.

Open 02 · the brief

A living prediction-rule register

Credit · unclaimed

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

What to build

An open, searchable register of every clinical prediction rule published for physical therapy, stating for each one how far it has got: derived only, checked by its own authors, tested by an outside group, or failed an outside test, with links to each study. An AI agent searches new papers each month and proposes changes that a person approves.

Why it is still open

The evidence sits in one-off reviews that go out of date. Walsh and colleagues (2021) found 9 outside validation studies of 7 prescriptive rules for musculoskeletal conditions including back, neck and shoulder pain and carpal tunnel syndrome; 7 of the 9 found no link between the rule and the outcome, and the authors concluded that none should be used to choose treatment. New rules are still being derived, for knee osteoarthritis in 2024 and neck pain in 2026. The nearest register, built by Claire Keogh, Tom Fahey and colleagues at the Royal College of Surgeons in Ireland, logged 434 rules for primary care published up to 2009 by stage of testing (54.8% validated, 2.8% tested for their effect on care) and planned a web version; it covers no rules after 2009 and is not kept up to date. A Europe PMC search found no living, regularly updated register for physical therapy.

pmc.ncbi.nlm.nih.goveuropepmc.orgpmc.ncbi.nlm.nih.govpmc.ncbi.nlm.nih.govpmc.ncbi.nlm.nih.gov

Built with what exists

the Europe PMC and Crossref search APIs; an AI model to screen new abstracts, such as the open-weight Mistral Large 4 (see the Note "A trillion open parameters"); PROBAST, a published checklist for judging bias in prediction-model studies; a static website on GitHub Pages with the data on Zenodo as a CSV file.

  1. First step

    Enter the 7 rules and 9 validation studies from Walsh's review in a spreadsheet, one row per study, with the derivation paper for each rule.

  2. Done looks like

    Every rule in four published reviews (May and Rosedale 2009, Haskins 2012, Kelly 2017, Walsh 2021) entered with its studies; two people check 50 random entries and agree with the register at least 95% of the time; over three monthly updates, the agent's proposed papers are compared with a person's own search of the same months, with recall and precision published.