Can AI Predict Cytokine Release Syndrome Early? New Insights from EHA2025 | Yi Lin, MD, PhD | #EHA 2025

Hi, I'm Yi Lin from Mayo Clinic, and I'm here at EHA 2025 in Milan.

I'm very excited to share with you results that my collaborators are presenting at this Congress and how we are using AI and machine learning to analyze the clinical data in patients who are receiving CAR-T cell therapies to be able to predict the onset of cytokine release syndrome days before a patient may experience that first symptom.

Fever is often the most common first symptom that patients may feel and can be clinically detected. Currently, management starts when patients have their first fever with tocilizumab as a frontline treatment for cytokine release syndrome.

This type of side effect really limits the centers that are able to prescribe and manage CAR-T in the first month after infusion.

So we wanted to use AI tools to see if we can be more able to predict that onset to help these CAR-T centers do more individual, patient-directed, risk-adapted management.

This may help the centers be able to use their clinics, hospital beds, and resources much more effectively and efficiently, and help the patient potentially stay outpatient in more settings and not need to come into the hospital until they need that escalated management.

Having a tool like this may also enable us to identify even earlier preventative strategies to study, so that side effects like cytokine release syndrome could be reduced further in the future.

What we're able to find out is that even just with current clinically available data—things like vital signs that you may normally be taking in clinic visits and labs that are standardly done in patients post-infusion—we can predict when the patient may experience that first fever even one day, two days, or up to three days before they have the fever.

This is during times when patients may feel perfectly normal and not necessarily have any of these symptoms.

We're hoping that this is something that can be implemented in real-world practice, as we see that hospital access can be a limitation to the number of patients that we can treat with CAR-T, as well as with bispecific antibodies, where they may experience some of these same side effects.