This talk is aimed at healthcare practitioners and healthcare data teams who are developing registries in the field of cancer. I will share experiences of my team evaluating the relative performance of large language models against previous NLP methods for cancer disease response classification (https://doi.org/10.1093/jamia/ocad133) and sites of metastases (manuscript under review JCO CCI). These will include the libraries, tools we used such as John Snow Labs’ Healthcare NLP & LLM libraries, the effects of data augmentation techniques and prompt-based fine-tuning, as well as challenges we faced. This will be useful to healthcare data registry teams which are lean and low-resource, looking for low-cost and sustainable solutions.

The talk will also cover some of our ongoing work with generative AI models for clinical trial matching and a discussion on some healthcare use cases that we think are of interest to the clinical and digital healthcare community. I will also briefly discuss the general potential and challenges around foundation models for generalist medical AI.