A suicide risk prediction study pipeline written with Claude
A clinical psychologist with no coding background used Claude to generate future narratives and write the Python pipeline for a suicide risk prediction study.
Credits
- Reddit poster AI_Psych_Research Source ↗
Earliest dated source:
The full story
A research clinical psychologist with no coding background used Claude in a study on predicting suicide risk. Using 164 recordings of people talking about their expected future selves, Claude 3.7 Sonnet generated two "future narratives" for each person, one with a crisis and one without, and a local Llama model then scored which was more plausible from the original interview.
Claude also wrote the whole Python pipeline, which the researcher checked as best they could. People whose crisis narrative scored as more likely were significantly more likely to report suicidal ideation at an 18-month follow-up, and the method caught 75% of high-risk people a standard questionnaire missed. The work was submitted for peer review.
How it was done
Tools
Themes
Loading themes…
Sources
From a short post by the maker, so the details are as they tell them.
- Reddit: A vibecoded project worked to predict suicide risk (preprint + code)Reddit poster: AI_Psych_ResearchPosted
The maker's post on r/ClaudeAI.
If it disappears: archived copy of Reddit: A vibecoded project worked to predict suicide risk (preprint + code) (opens in a new tab)