NHS AI Conference

conference / Feb 03, 2025

A short summary of what I found out about the conference. My initial thoughts on what I wanted to go see were: Synthetic data in medicine, regulation and bureaucracy and any interesting papers on display.

Sythetic Data

The guy talking about synthetic data had helped write a Utah bill and EU regulations on synthetic data. His most interesting/worrying point he presented was any synthetic data is not subject to GDPR regulatory rules. This means a "fake" data can be produced from real protected data but used without regulation.

Positive use cases for the synthetic data could be testing proof of concepts on the synthetic data to check for new avenues of research, so the researchers don't have to wait for long periods of time to get approval for the data. Once the synthetic data shows promise, the real medical data can be requested.

Doctors Bias

When training models on doctors notes with quantitative data (i.e. blood pressure), the researchers found the model would perform better when the notes were removed from training - simplifying the data. Overall the doctors notes would vary so much that just giving the model clear quantitative data would allow it to perform better. The model would be able to spot patterns in the data.