1) Currently a lot of other services ask people to take "x" number of footsteps away from the screen to approximate "y" feet from the monitor. Using this context, I'd argue that the variation in arm length isn't as dramatic as variation in foot size. Ultimately though, when we're using near vision as a proxy for distance vision, the natural variation in arm length isn't crucial. But! Once we roll out our distance vision check, we won't be relying on arm length.
2) Will leave this to Kristine.
3) Interesting. Hadn't thought about this one. My guess is no because the most important ratio is optotype size:testing distance. (Optotype = the numbers/letters on the screen that a patient is reading)
4) It's possible and we'll need to pressure test this against gold standard in-person maneuvers.
5) Same as #4. Also this is a particularly interesting point because a similar problem exists in person. As an extreme example, I've had patients come in who've memorized the letters in the 20/20 line because they were very motivated by one thing or another (e.g getting their driver's licenses renewed.)
6) Is this the "which is better, 1-or-2" question? All I'll say is that there are a number of interesting ways we could try to simulate these.
Hope this answers some of the q's! Thank you for all the thought that went into them.
It's interesting to think of the adversarial element in #5 where vision test results are used to qualify for something. In this case a completely unsupervised test is really easy to cheat on -- people can just lean in close to the monitor! If you're not giving people something that they can use to receive a benefit like a job or a license, that incentive to cheat seems weaker, but maybe people will present their fresh prescriptions (!) as purported proof that they have very acute vision.
I was thinking more about psychological aspects where people might not want to admit that they have certain vision problems, so they might feel an incentive to convince themselves that they saw the correct thing. The order and context of presentation might affect how easy it is for people to convince themselves of that. I know I've taken similar tests in person at the optometrist (like looking at a grid to see if any portions appear distorted), but I don't remember exactly how the optometrist asked me to confirm what I'd seen.
This may be an underappreciated soft skill on the part of medical professionals -- getting people to tell the truth about their perceptions in diagnostic tests, or noticing when people may be dishonest or simply uncertain. So that may be pervasively tricky for you to address, at least with a small percentage of patients: if they want to think of themselves as having good vision, they may consciously or unconsciously fudge the results a bit so the assessment comes back better.
The psychology of how people relate to their vision -- especially the independence that good vision affords -- is very complex and certainly something I wish that our training spent more time emphasizing. There are patients who come into clinic with relatively minor and non-vision threatening problems who are afraid of imminently going blind, and there are patients on the other end of the spectrum who are imminently going to go blind but are in denial about it (or are not terribly bothered by the possibility.) Handling these scenarios and all the gray spaces inbetween is one of the more challenging parts of delivering eyecare (and healthcare in general.)
Ultimately, we're aiming for clinical accuracy and scalability first, with an understanding that there are lots of underlying incentives and potential roadblocks that we will tackle head on when the time is right.
1. We're currently exploring doing a distance check via the webcam using triangulation and face detection to tell folks how far from the camera they are.
2. The average monitor won't add much noise - if we're going by in person eye exams as the gold standard - there's actually a lot of variance between various doctor's offices (lighting, use of a projector/mirror to simulate distance vs placing the chart on the wall).
There's an interesting angle to doing eye exams digitally where we may be able to be more accurate than in in person exam (based on institutional research studies - not something we've yet personally explored).
5. Really interesting and valid point - making these types of questions more interactive is definitely on our roadmap.
6. Clinical trials will help vet any methods we build to show how they compare to existing practices
A few questions after trying the prototype:
(1) Is there enough variation in people's arm lengths that it would be beneficial to give another way to describe the distance to the monitor?
(2) Is there any impact from the monitor's dot pitch or contrast settings on which lines people will be able to read, even at a constant size and distance? Are some monitors effectively just better quality in a way that could lead to a different result?
(3) Is there any impact from the monitor's physical aspect ratio because some people are looking more straight ahead and some are looking more to one side (when using a wide-screen monitor in landscape mode)?
(4) How well can people who have double vision or retina problems correctly match the test images they see to one of the intentionally-distorted samples? Could their vision problems also distort the samples so that they end up picking the wrong one?
(5) Could having the intentionally-distorted samples visible at the same time as the test image bias people's perceptions, since they can guess which one is most normal (least distorted) and then optimistically claim to have seen that one?
(6) Will you be able to realistically simulate the effects of a proposed prescription change, even without having physical lenses for the user to look through? (I didn't upload my prescription to the prototype, so I don't know exactly what happens after that step.)