undefined | Better HN

0 pointsschoen5y ago0 comments

Cool! I can see there's a lot of creativity that goes into replicating in-person vision tests with a reasonably valid unsupervised version on people's home equipment. I hope you have success replicating as many of them as possible.

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.

0 comments

2 comments · 1 top-level

Killakwinn5y ago· 1 in thread

Yeah, in the absence of in-person supervision, there are a lot of safety measures that need to be implemented. For example, making sure that people are at the correct testing distance (via a number of potential feedback mechanisms that we're testing) and making sure that people can't zoom in on the optotypes to cheat.

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.

100kristine5y ago

Adding a few things from the engineering side of things! :)

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

j / k navigate · click thread line to collapse