Hacker News new | ask | show | jobs
by jseliger 766 days ago
For my part, I've been building an AI agent that watches the daily change feed from clinicaltrials.gov and sends out a personalized newsletter that filters for specific trials and answers specific questions about those matched trials: https://zeeq.ai. Hopefully a useful tool for anyone that is interested in participating in or tracking clinical trials.

Thank you for making it. From what I've seen and experienced, the problem has been "garbage in, garbage out"—that is, there isn't sufficient data posted publicly on clinicaltrials.gov to figure out which trials are best and which are actually open and available. My wife wrote "Please be dying, but not too quickly: a clinical trial story. A three-part, very deep dive into the insanity that is the 'modern' clinical trial system" on using the system, and the actual experience of it: https://bessstillman.substack.com/p/please-be-dying-but-not-...

To figure out what's actually going on, we've had to make a lot of appointments and talk to oncologists to understand what is available and what isn't. The AI companies whose systems we tried missed the better treatments (e.g. BCA-101, or petosemtamab / MCLA-158), although we did not try yours, so perhaps it's capturing material others aren't. "Phone calls and appointments" are how I wound up learning about Seagen / Pfizer's antibody drug conjugate (ADC) PDL1V: http://jakeseliger.com/2024/04/22/the-emotional-trial-of-cli... (which appears to be working right now, albeit with side effects).

Right now, keeping a true system up to date would require a lot of phone calls, along the lines of VaccinateCA: https://worksinprogress.co/issue/the-story-of-vaccinateca/, which seems hard.

1 comments

The current use case for Zeeq AI is really focused on monitoring for new trials or updated trials and not so much for finding trials at the moment.

Your best bet in that case is the AACT database which is pretty accessible if you have a basic knowledge of SQL.

The main gap as you identified is that the trial information in CT.gov is quite sparse and not necessarily deep enough so a true system would need to perhaps also crawl first party sources (e.g sponsor websites) or research papers to find more information.