robertrmorris

Good questions. And your experience makes a lot of sense. Ideally, we would see positive changes persist over many time points (and so we aren't measuring immediately before self-harming and immediately after). I'd be really grateful if you tried it provided some feedback for us: https://join.kokocares.org/koko-referral-lifelines?source=hn

Scroll to the bottom to try the "managing self-harm" mini course. It only takes 7-8min and there is a spot for feedback at the end.

Thanks a lot, that's really nice to hear. We've had a somewhat meandering path, which you can read about here: https://www.kokocares.org/origin-story

That's great to hear. Kudos to your wife! Koko works well outside of the USA (but is English only). One of our earliest users, from London, is now a member of our core team.

We regularly consult clinical advisors (listed here: https://www.kokocares.org/our-team)

I made some comments on outcome measures in another comment you can find here, but you can also see some of our papers here: https://www.kokocares.org/research

Frankly, I agree with pretty much all of this. We hear similar things from our users. This is why we try to provide a suite of options, including things like peer support and other interventions they can engage with immediately — as compliments to lifelines. We’re still learning about what works best, but the status quo is abysmal. Here’s an example: I can go on Google and search for “flight to Miami” and I’ll be led through an incredible UX that’s designed to get me to a purchase as quickly as possible. But if I search for “depression”, I get a one-box that provides a list of clinical definitions of depression, bipolar, and its various subtypes — better suited for a diagnostic manual than for anyone who might actually be struggling. Other platforms provide tips on how to take a deep breath, reach out to friends, or walk around the block (the digital equivalent to a health brochure you might find in a waiting room). The shortcomings of these approaches have been studied before, and yet they still persist. Why don’t we measure and track these things with the same rigor we do for all online experiences?

Thank you. We can release bindings for new languages very quickly. Just fill out our signup form and we'll prioritize Elixir. https://r.kokocares.org/api_signup/

Great questions, thank you. Looking only at completers creates selection bias, for the reasons you articulate. In published studies, we compare interventions to control conditions (ideally an “active” control, or something that has some purported therapeutic benefit). We love the Trevor project and work hard to get candidates on our platform to that resource. We have done some comparisons with other life lines and the issue is some have incredibly long wait times and drop-offs. Ideally, we can offer both. For the suicide prevention lifeline, we’re a resource that’s listed that people can access while they wait.

It is very true that self-injury is not the same as suicidal ideation, though they can certainly overlap. A common thought is that asking about suicide or presenting resources could be harmful or ‘trigger’ more ideation. The evidence to date suggests, on the contrary, that asking about suicide can actually reduce risks. https://pubmed.ncbi.nlm.nih.gov/24998511/ https://www.cambridge.org/core/journals/the-british-journal-...

Hi, thanks for your interest. The resources appear differently for different platforms we integrate with. In some cases they can be accessed via DM. Some will differ based on the search term as well. The most generic case is here: https://join.kokocares.org/koko-referral-lifelines?source=hn" We are actively updating these based on user feedback.

We’re a non-profit and our service will always be available for free. There is no paywall. We support ourselves through donations. This should be more clear on our landing page, thanks for noting this.

Yea this is an interesting problem. The whole question of whether, when, or how to intercept someone who might be in trouble is really challenging and we’ve thought about this for many years (and had some missteps along the way and learned a lot on what works and what doesn’t).

Our system gently recommends our service to users right when they search and so the cost of a false positive is low (they can just ignore it or it might just seem like an unrelated PSA). Search is also great because we can vary the intensity of the keywords. For one of our partners, we’re now surfacing resources (in subtle ways) for lower risk searches like “depression.” It is super important to us to think about how we might help people upstream, before they reach a state of crisis.

For users flagged, we work well as a layer on top of CTL as our ux works for people across the entire spectrum of severity.

Great questions! Our library runs completely on the server side, caching a list of regexes which are used to match against. So no data is ever sent to our API.

As for publishing the lists, it's definitely something we're thinking of. for now, it's easy to get them if you sign up with us. we don't charge for use

Thanks for the question. I’m going to interpret this broadly and try to go from there, but let me know if you had something more specific in mind.

TL;DR At a high level, for people who complete our interventions, we see 71% feel more hopeful, 42% feel better about their bodies, and 67% feel less self-hatred. Completion rates range from 25-55%. Outcomes would most likely be lower for those who dropout prematurely.

More specifically:

We track multiple outcomes, depending on what the user may be presenting. If they are experiencing suicidal thoughts, we track conversion to crisis lines.

See here: https://psycnet.apa.org/record/2019-14424-004

We follow-up 5hrs later and ask general questions about their experience with the life line.

If they are experiencing self harm, in addition to crisis lines, we offer them a single-session online intervention on managing sh. For that, we see significant improvements pre vs post on measures like “self-hatred”, and “desire to stop selfharm”, with medium effect sizes (.4-.8 cohen’s d). Very hard to show enduring effects for this, however. This research, as well as our work on disordered eating, is still in prep.

For peer support, we have previously published data here: https://pubmed.ncbi.nlm.nih.gov/25835472/

And here: https://pubmed.ncbi.nlm.nih.gov/28903637/

For our interventions on mood and stress regulation, we’ve adapted single session interventions, alongside some wonderful collaborators at Stony Brook. They have published their work here: https://www.nature.com/articles/s41562-021-01235-0

"oops" new landing page. I updated the links, but you can also find them on our main page: "www.kokocares.org". Note also that for the kit, we have our own licensing agreement that's available if you sign up, but is not on this webpage.

Thanks for reporting this. We use an embedded typeform for our signup page which looks to be ok but you can also go directly to the form here (https://koko-ai.typeform.com/to/xB0X2Grc). For the Python docs, our language bindings are open sourced and we’ve been maintaining their documentation directly on the repo to ensure it is up to date without having to copy data across. Was the documentation confusing on the github readme?