How CancerBot Leverages AI Now and in the Near Future

We started CancerBot to address the grave difficulty of cancer patients finding clinicaltrials that they are eligible for. The “bot” in the title refers to using Large Language Models (LLMs) to extract the unstructured prose participation criteria created by researchers into structured form (a database of course). We have performed some quite innovative prompt engineering to tackle this difficult problem, without which it would not have been solved. But the rapid emergence of the power of LLMs is what enables the innovation. We are now focusing on applying the power of LLM-powered bot to several adjacent problems in supporting cancer patients that will described here.

To begin with, right now instead of hiring a trained clinician (your specialist doctor) to find you trials (although this option is not even available to most patients) you can just enter a minimal amount of your information to find trials that are you are eligible for. You should still stay engaged with your specialist, and let him or her know the trials that you are interested in. Many (but notably not all) of these trials require doctor’s referrals anyway.

The “bot” moniker will have additional meanings as CancerBot progresses. First of all the the “successive data for the most trials” that CancerBot enables (telling you which data fields to fill in to increase your eligibility the most) is something we will be enabling with a chatbot interface. The field by field entry and search will still always be available.

This chatbot will also ask if you are interested in standard of care recommendations. For many diseases, such as multiple myeloma which we support today, the best standard of care option is not always clear. But with our growing patient data we can use machine learning predictive models to make good suggestions for you. Probably more than a clinician would be willing to do. That said, as with trials, you should always discuss your proposed treatment with your oncologist.

Finally, one of the growing areas in cancer research is the recognition that cancer is a metabolic disease. It is a revival of Otto Warburg’s early insights into how cancer survives and grows from the 1920’s: the Warburg Effect. Books such as How to Starve Cancer and The Metabolic Approach to Cancer provide a roadmap on how to disrupt the cancer metabolism to limit progression and increase chances of survival. But the mostly nutrition-based approach is breathtakingly complex. CancerBot will dialog with patients to find out the current status of their diet and then recommend small actionable changes to fight cancer optimally, usually alongside more conventional treatments, increasing the efficacy of those treatments (as many studies have already shown). And the relationship will continue with CancerBot suggest more and more diet optimizations over time.

In general CancerBot will use AI and LLMs to stay connected in a rich and ongoing dialog with the patient, always finding the latest relevant science to guide patients to better outcomes.