Oncology Intelligence Monitor

The Backstory

A couple of years ago I deleted my Twitter account because I didn't agree with its political direction and no longer wanted to support it. But I did miss being able to follow what colleagues were posting — the conference live-tweets, the hot takes on new data, the clinical pearls that only show up on social media.

The idea actually came from a little gadget I bought while skiing at Sun Peaks: a Raspberry Pi that scrapes snow conditions from ski resorts so you know who is getting the most snow. I figured if I could scrape snow data, why not scrape oncology Twitter and PubMed? So I built a system that monitors it all using a third-party API service — no money goes to the Twitter overlords. And since I had it running anyway, I figured why not share it.

How This Works

What goes in

• Papers. PubMed is searched twice a week for new trials, practice-changing results, and guideline updates. Each paper is scored 1–5 on its design and what it actually shows. Papers in low-ranked journals can’t score above 2, and only papers scoring 3 or higher can make the digest.
• Discussion. A curated list of oncologists who reliably post strong clinical content is followed on X and Bluesky, along with disease-specific hashtags. When several of them start using the same new hashtag — as happens at ASCO, ESMO, or WCLC — the system picks it up automatically.

What comes out

• Themes. The system finds the week’s real events — a trial readout, a conference presentation, a regulatory decision — and the conversation around each. An event becomes a theme only if at least two people actually discussed it, not just reposted it. Each theme gets a short summary with the trial name, the key numbers, and the debate, crediting the people who drove it. The best original posts are shown in full.
• Papers. Each selected paper gets a short summary of the design and results, with the real numbers, and a one-line bottom line for practice.

Keeping the AI honest

• The AI writes only from the actual posts and abstracts, not from memory.
• Every claim must quote its source word for word.
• An automated check traces every number and trial name back to a source. A draft that fails is held back.

Nothing goes out until I’ve reviewed it

Every Sunday the system prepares the week’s themes and candidate papers, then stops. I go through each one — drop what doesn’t belong, fix what’s mislabeled, and decide for every paper: email, website only, or out. The digest is written only after I approve it. If I don’t review a week, no digest goes out that week.

Delivery

Digests are split by disease group, so you only read your area, and they arrive Monday morning.

It isn’t infallible — see the disclaimer below — but a made-up number would have to get past the source requirement, the automated check, and me.

Also on this site: a clinical trials search refreshed twice a week from ClinicalTrials.gov, a search of everything collected, and conference digests for major meetings.

Disease Groups Covered

Where You Can Help

• Feedback — this is a work in progress and your feedback is essential. Let us know if the digest is useful, what's missing, or if the AI is including things that shouldn't be there. That input directly shapes how this evolves.
• Share it — if you know someone who might find this useful, forward the digest or send them the subscribe link below.
• Buy me a coffee — if this saves you time, consider buying me a coffee. It helps keep this running.

Contact

Questions, feedback, or a correction? Send a note and it comes straight to me — I'll reply by email. Your message isn't shown publicly.

Leave this field empty

Send message →