Week 3: Accessibility
Hello everyone,
To diagnose OSA without a professionally conducted sleep test, I have to focus on non-polysomnographic variables. This means that the tool will not be as accurate as I’d like, but will still achieve the purpose of telling someone whether there is a decent chance that they have OSA and if they should go through a professional sleep test. I have access to clinical data of people with OSA but don’t have any details of physical examinations that were conducted on those people. Although, that’s completely fine since clinical data will be enough for this specific project.
I’ve been reading some interesting publications online at NCBI and have come across some studies that have used non-polysomnographic features. However, they have only been able to classify OSA patients with a maximum true positive rate of 0.71 and a minimum false positive rate of 0.15. This has led me toward changing the vision of this project from accuracy to accessibility. While some studies were able to estimate the probability of having OSA as no, mild, moderate, or severe, my tool will only display either “high probability”, “medium probability”, or “low probability” for simplicity. Those that receive “high probability” will be highly recommended to go in for a real test while those that receive “medium probability” can decide for themselves.
Fun fact: A sleep study can cost 1k to 2k per night.
Thanks!
Savinay
Seems interesting! What are some of the specific types of clinical data you will be using, and how are they different from polysomnographic data?
Have you investigated when these articles were published? There’s a lot of exciting research happening, and maybe you could look at papers in other fields but tackling the same type of data to see if they can achieve better accuracy.
Ultimately, I think your idea to create accessibility is extremely important. Maybe, you could even scale your model to a web app or a platform that can be used easily.