BRENNEN

Industries / Health / Healthcare / Optometry

Measuring the prescription

Refraction finds the lens power that brings a person's eye into focus.

6

Milestones achieved, from 1864 to 2019.

2

To watch for: announced, not yet confirmed outside.

2

Open deliverables. Nobody has shipped them.

The road

AchievedWatch forOpen

Achieved

  1. 1864

    Refraction becomes a science

    Frans Donders, of Utrecht, published On the anomalies of accommodation and refraction of the eye in London in 1864, after an 1862 Dutch work on astigmatism and cylindrical lenses. The book covered the eye's errors of focus and their correction by lenses. De Jong writes that until it, eye doctors and opticians did not know that short-sighted and long-sighted people could also accommodate, that is, refocus the eye's own lens, which a refraction has to control for. Donders asked his assistant Herman Snellen to design the letter test used to check the result.

    pmc.ncbi.nlm.nih.gov

    Credit

    F. C. Donders

  2. 1922

    Lenses on a dial

    A phoropter holds wheels of plus and minus lenses, prisms and other test lenses in front of the face, so the examiner can switch lenses by turning a dial instead of changing trial lenses in a frame one at a time. Henry DeZeng introduced the Model 570 Phoro-Optometer in 1915, and on 10 January 1922 the US Patent and Trademark Office registered the name Phoroptor; DeZeng released the No. 584 Phoroptor that year, according to Reichert, the brand's present owner. American Optical bought DeZeng Standard in 1925, and its 589 model of 1934 added the Jackson cross cylinder, a lens flipped between two positions to find the axis and power of astigmatism.

    reichert.commuseum.aco.org.au

    Credit

    Henry DeZeng; American Optical

  3. 1970s

    Machines measure the focus

    An autorefractor measures the eye's error of focus without asking the patient anything, from infrared light reflected off the retina. G. Collins described an electronic refractionometer in 1937. The first commercial autorefractor, the Ophthalmetron, developed by Safir and sold by Bausch & Lomb, appeared in the 1970s, followed by the Dioptron (Coherent Radiation) and the 6600 Autorefractor (Acuity Systems), Kabat and colleagues wrote in 2019. An autorefractor reading is now the usual starting point for an examiner's refraction, Bullimore and colleagues wrote in 1998.

    pmc.ncbi.nlm.nih.goveuropepmc.org

    Credit

    G. Collins; Safir and Bausch & Lomb

  4. 1998

    Two refractions rarely match

    At a University of California optometry clinic, two optometrists refracted the same 86 people. Their results for spherical equivalent, the average power of the correction, differed within 95 percent limits of −0.90 to +0.65 D (D is diopters, the unit of lens power), about twice as wide as for repeated autorefractor readings (Bullimore, Fusaro and Adams, 1998). When 40 optometrists in Oxford and Westminster each refracted one man's eye, their results differed by more than 0.78 D no more than once in 20 (MacKenzie, 2008).

    Lessona difference of 0.25 to 0.50 D between two refractions is normal variation; Goss and Grosvenor's 1996 review found about 80 percent agreement within ±0.25 D and 95 percent within ±0.50 D.

    europepmc.orgeuropepmc.orgeuropepmc.org

    Credit

    M. A. Bullimore, R. E. Fusaro and C. W. Adams; G. E. MacKenzie

  5. 2019

    A handheld wavefront refractor

    A wavefront sensor measures how an eye distorts light leaving it; J. Liang, J. F. Bille and colleagues at the University of Heidelberg applied the Hartmann-Shack sensor, an array of tiny lenses, to the human eye in 1994. PlenOptika's handheld QuickSee is a wavefront autorefractor. In a 708-person study at Aravind Eye Care System in India, run with PlenOptika staff, glasses set to its readings gave acuity one letter worse on average than an experienced refractionist's, and more than half of participants had no preference or preferred the device's. An outside comparison of 190 patients at Sankara Eye Hospital, Bangalore, led from Harvard, found QuickSee within −0.05 D of an ophthalmologist's refraction on average (standard deviation 1.04 D), against 0.61 D (2.20 D) for the EyeNetra phone attachment and −0.41 D (2.14 D) for the Nikon Retinomax.

    europepmc.orgeuropepmc.orgeuropepmc.org

    Credit

    N. J. Durr and PlenOptika with Aravind; A. Agarwal, D. E. Bloom and colleagues

  6. 2019

    An online test recalled

    Opternative, later renamed Visibly, of Chicago sold an online refraction test. The American Optometric Association (AOA) complained to the FDA in April 2016. After the FDA warned the company in a letter that it lacked the required clearance, the test was recalled for lack of 510(k) clearance, the FDA's premarket review, and the FDA published the recall notice on 8 August 2019. In August 2022 the FDA cleared a narrower product, the Visibly Digital Acuity Product, which checks acuity only, for adults aged 22 to 40, and whose label says it does not replace an in-person exam. In 2023 the company paid US$500,000 to settle a multistate investigation into claims that its test was as accurate as an in-person exam.

    Lessonwhat returned to market was an acuity check for a limited age range, not the online refraction first sold.

    aoa.orgaccessdata.fda.govillinoisattorneygeneral.gov

    Credit

    Opternative (Visibly); US Food and Drug Administration; Illinois Attorney General

Watch for

  1. Kiosk prescriptions in shops

    Eyebot, a Boston company, says its kiosk measures refraction, the power of current glasses and acuity in about 90 seconds, after which a licensed eye doctor reviews the results remotely and writes the prescription. It told the trade site Glance that two studies with nearly 800 participants found its prescriptions gave 20/20 acuity at the same rate as prescriptions from an in-person exam; no published paper was cited. Kiosks are in malls, universities and some Walmart Vision Centers and Sam's Club optical centers, and in March 2026 the AOA asked the FDA to review the kiosk as a combination of three Class I devices used without a trained professional.

    Achieved whenthe studies appear in a peer-reviewed journal, or an outside group repeats them.

    glance.eyesoneyecare.comaoa.org

    Claimed by

    Eyebot (123 SEE, Inc.), February 2026

  2. A wearable digital refractor

    Forus Health and Sankara Eye Hospital report that the 3nethra Specto, a wearable refractor with liquid lenses, matched conventional subjective refraction in 135 adults (mean sphere −1.02 D against −0.99 D) and repeated well, though mean cylinder differed (−0.49 D against −0.83 D). Four of the nine authors work in Forus Health's research department.

    Achieved whena study without the maker's staff reproduces the agreement.

    europepmc.org

    Claimed by

    Forus Health and Sankara Academy of Vision, June 2026

Open

  1. 01

    A simulated patient for refraction

    Open software that plays the patient in a subjective refraction. Each simulated eye has a known sphere, cylinder and axis and answers "first or second lens" with realistic noise. Any method, from the standard phoropter routine to an automated phoropter's algorithm or an AI model asking the questions, can then be scored on how close it ends to the true prescription and how many questions it took.

    The brief
    Credit · unclaimed
  2. 02

    Refraction spread, pooled openly

    An open dataset of every published study of how much refractions differ when repeated, by one examiner, between examiners, and between devices and examiners, with each study's 95 percent limits for sphere and the two astigmatism terms (J0 and J45, which turn cylinder and axis into numbers that can be averaged) in one format.

    The brief
    Credit · unclaimed

Open 01 · the brief

A simulated patient for refraction

Credit · unclaimed

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

What to build

Open software that plays the patient in a subjective refraction. Each simulated eye has a known sphere, cylinder and axis and answers "first or second lens" with realistic noise. Any method, from the standard phoropter routine to an automated phoropter's algorithm or an AI model asking the questions, can then be scored on how close it ends to the true prescription and how many questions it took.

Why it is still open

Perches, Collados and Ares (PLoS One, 2016) built software that shows the retinal image an examiner would produce through a virtual phoropter, for teaching, and note that "the decision dilemma during Jackson Cross Cylinder test" is not modeled in such tools. Automated refraction methods are tested on people, one at a time: Otero, Aldaba and Pujol (Journal of Optometry, 2018) ran a bisection search, which halves the range of candidate lenses at each answer, on a motorized phoropter with 50 adults, so no two methods have been scored on the same eyes. A GitHub search in November 2026 found a rule-based chatbot for subjective refraction (juhoward/chatbot-v1, 2023) and an app-controlled phoropter prototype (MoeHussein/Basira, 2026), but no shared benchmark.

pmc.ncbi.nlm.nih.govpmc.ncbi.nlm.nih.govgithub.comgithub.com

Built with what exists

Python; the reduced eye model in Perches and colleagues; the spread between real examiners from Bullimore (1998) and MacKenzie (2008) to set the answer noise; open-weight or hosted AI models to test as examiners.

  1. First step

    Write a simulated observer for sphere only that answers with noise matched to Bullimore's limits, and score a simple up-and-down staircase on it.

  2. Done looks like

    A public benchmark of at least 1,000 simulated eyes with known prescriptions, scored for at least three methods, reporting the share that end within 0.25 D and 0.50 D of the truth and the median number of questions, with code anyone can run.

Open 02 · the brief

Refraction spread, pooled openly

Credit · unclaimed

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

What to build

An open dataset of every published study of how much refractions differ when repeated, by one examiner, between examiners, and between devices and examiners, with each study's 95 percent limits for sphere and the two astigmatism terms (J0 and J45, which turn cylinder and axis into numbers that can be averaged) in one format.

Why it is still open

Goss and Grosvenor's literature review (1996) predates the power-vector reporting now used and the handheld, phone and kiosk devices. Later studies report agreement in different forms. Carpena-Torres and colleagues at Complutense University of Madrid (Photonics, 2024) measured three optometrists refracting 86 people and proposed ±0.75 D as a realistic yardstick for testing new devices, but from their own data, not a pooled set. A Europe PMC search and a web search (November 2026) found no pooled dataset or meta-analysis of refraction repeatability.

europepmc.orgdoi.org

Built with what exists

The Europe PMC and Crossref APIs to find papers; an AI model to pull values from tables, with every value checked by a person; a CSV on Zenodo; R or Python for a forest plot.

  1. First step

    Extract the studies cited by Goss and Grosvenor and by Bullimore and colleagues into a shared spreadsheet with a fixed list of columns.

  2. Done looks like

    At least 60 studies extracted twice by independent extractors, disagreements resolved and logged, the limits for each method on one chart, and a public CSV.