There are 20 different drugs prescribed to treat depression, anxiety, and PTSD. And this isn't meant to knock clinical medicine, but they are prescribed essentially at random. Clinicians, especially psychiatrists, base their prescriptions off of limited research at best, and in many cases, based on their own (essentially anecdotal) experience.
If we could collect data from tens of thousands of people – we could make much better informed recommendations. Very basic machine learning methods could give a numeric "likelihood of response" to all 20 of those drugs based on your mental health profile. (And this could extend to other things besides antidepressants. ("Here are 6 different types of therapy, and your likelihood of responding to each. Here are 12 different styles of exercise / meditating / changing your diet, and your most likely outcome with each).
Example 2
Are you depressed, or are you anxious? Within the first 100 people who came amd answered the first 20 questions, it was clear that we had grossly underestimated how many people would report struggling with anxiety and depression A LOT.
But how do you differentiate between anxiety and depression? I have both, and I find that my anxiety is 100% a result of my depression, and my depression is 100% a result of my anxiety. (Huh?). Clinical medicine thinks these are two very different things (and I tend to agree), but there is a very complex overlap between then. (And it only gets worse when you factor the presence or absence of trauma). A large detailed dataset like we are trying to build here would help separate the two, and identify more precise "subtypes" of each.
Example 3
Let's say 3 years from now we’ve collected data from 5,000 users. Half of them only did the first 20 questions, but the other half stuck around to answer more. And ~3-500 of those users are bought into this idea, so I buy a few consumer-grade EEG headsets, and start mailing them to people with a basic protocol for “record yourself for 30 min - 10 at rest, 10 watching this video, 10 while doing this basic task.” After a year, we have several hundred individuals with detailed mental health profiles and solid EEG data. Now we can tell you fun, interesting, and hopefully useful things about your brain and it's health. (This is how the future of psychiatry should work, and I believe growing a dataset like this can help lead things in that direction).
Example 4. A real finding from my time working in this field.
Asking about suicidality is really hard. But in a large dataset, where people answered hundreds and hundreds of these kinds of questions, you know what best correlated with suicidal thoughts? These four questions from the social adjustment scale: “Have you been interested in what your children are doing during the last two weeks?'. “Have you been able to talk and listen to your children during the last two weeks?” “How have you been getting along with your children during the last two weeks?”. “How have you felt toward your children these last two weeks?”.
Example 5, Real example 2. Another finding from my time working in this field.
The more time between when someone starts dealing with depression and when they start treatment with an antidepressant, the less likely that antidepressant is going to be to work for them. Why? Maybe because people who are less likely to immediately seek help have less confidence that an antidepressant is going to help them. OR, maybe because the longer depression goes untreated, the deeper it goes and the harder it becomes to treat. Either way, this is not talked about nearly enough – if people understood this fact (based on about 30,000 people – it was well established from data, not a few anecdotal accounts), it might change the way we as a society approach and treat mental illness. It could help normalize antidepressants as ‘hey, try this, it might help a lot,’ instead of ‘here, start taking this thing that you’ll be stuck on for the rest of your life.’