Michael Gensheimer

HealthScout Next-generation cancer trial matching

I'd welcome any feedback or feature requests, and please share with your colleagues!
Here's the link:
www.healthscout.app

The inspiration for HealthScout came from my patients. One in particular was extra motivated, spending hours researching trials. But they got overwhelmed trying to figure out eligibility & which trials held promise, and eventually gave up. I knew it was possible to do better.

We're launching in early release. It currently works for advanced/metastatic non-small cell lung cancer and site-agnostic solid tumor trials. We're going to add more cancer types soon, and some cool features to make it easier to zero in on the best trials.

matched to trial eligibility. When showing matched trials, we highlight the most promising ones using tags like molecular match, promising drug, and close site.

[also posted to X] Time to release my project out into the world... it's called HealthScout, it's a clinical trial matching web site for oncologists and patients. It uses advanced AI to translate complex patient info (cancer type, molecular profile, treatment history) into a format that can be...

Automated patient selection and care coaches to increase advance care planning for patients with cancer AbstractBackground. Advance care planning and serious illness conversations can help clinicians understand patients’ values and preferences. Data are limit

That's it! For more, see our recent paper on our randomized rollout of the AI model:
academic.oup.com/jnci/advance...
And excellent accompanying editorial:
academic.oup.com/jnci/advance...

The most effective things were to directly ask colleagues for feedback, and also monitor real-world usage since actions speak louder than words...

Get feedback from users, and be accessible. We have struggled a bit with this as the clinicians are busy and some are more invested in advance care planning (what we are using the model to encourage) than others. We sent out a Qualtrics survey to clinicians but got few responses...

Plan ahead for validation to make sure the model is performing well. Need to get data sources set up early. Can take a long time to have hospital IT get you data, they are often overworked and have many active projects. We validated using Epic follow-up/death info and Social Security Death Index...

We tried to plan carefully by involving important stakeholders (doctors, other providers, patients, IT, quality team, etc.). We have a wonderful Patient and Family Advisory Council and we presented to them to get feedback from the most important people, the patients...

First, this kind of thing is new, sensitive, and important to get right. When you are predicting serious health events, there is potential to cause harm and different people will have different attitudes toward it. Don’t embark on this if you are not fully committed or will lose interest...

For my first Bsky thread I thought I’d share some thoughts on deployment of clinical prediction models. Our home-grown AI mortality prediction model for cancer patients has been in production use for 4 years now, so we have been through the highs and lows...

:) Wake me up when they make the civilization AI actually smart instead of making the hard mode give it ridiculous amounts of resources and tech. In other words, if it was a priority to make the game AI smart wouldn’t they have done it already with older tech? Different story with dialogue though.