Yan Leyfman, MD

Honored and grateful to be invited to speak at this outstanding forum hosted by Memorial Sloan Kettering Cancer Center 🌍✨

I’m excited to share two oral presentations exploring how global oncology, AI, and evidence-based innovation can be translated into real-world action to reduce inequities in cancer care worldwide 📊🧬🤝

Immense gratitude to my mentors, Dr. Bonilla & Dr. Park, and collaborators whose guidance, support, and trust continue to shape this work — none of this is possible without you 🙏💙

Looking forward to engaging conversations, learning alongside leaders across disciplines, and advancing our shared mission to improve cancer care for patients everywhere 🌎🔬
@DrArturoAI
@CParkMD

Clinical notes are written for clinicians—but patients are increasingly being invited to read them. The challenge is making sure they can understand what they are reading.

A new study explored whether large language models could help bridge that gap through a patient-facing tool powered by GPT-4. Rather than simply rewriting notes, the tool generated several complementary “augmentations”: simplified text, definitions of medical terminology, frequently asked questions, extracted key information, and actionable to-do items.

In an evaluation involving 200 healthy female participants and eight clinical notes, access to these augmentations was associated with greater self-reported confidence and significantly better comprehension scores compared with the control group.

The investigators also interviewed seven individuals with a history of breast cancer who reviewed their own notes. Participants described the tool as easy to use and felt it improved their understanding, while also raising concerns about relying on LLM-generated explanations.

Importantly, clinician review found that most identified errors were minor, often reflecting assumptions about clinical context.

The broader lesson is that making medical information accessible requires more than simplifying language. Patient-facing AI must be evaluated for accuracy, transparency, and usability—with patients involved before deployment, not after.

The goal should not be to replace the clinician’s explanation, but to help patients become more informed participants in their own care.

nature.com/articles/s4159…

#ArtificialIntelligence #LargeLanguageModels #PatientEngagement #HealthLiteracy #DigitalHealth #Oncology

Can we turn the cerebrospinal fluid (CSF) from an immunologically “cold” environment into one capable of mounting an effective anti-tumor response?

Leptomeningeal disease (LMD) remains one of the most devastating complications of breast cancer, with limited treatment options and poor prognosis. The CSF immune environment is predominantly innate, with little evidence of meaningful adaptive immune activity.

Building on preclinical work showing that intrathecal dendritic cell (DC) administration could eliminate LMD in mouse models, investigators explored this strategy in a first-in-human Phase I trial in patients with HER2-positive or triple-negative breast cancer-associated LMD.

Fifteen patients received weekly intrathecal DCs across four dose levels.

Key findings:
Two dose-limiting toxicities occurred at the highest dose (5 × 10⁷ cells), establishing 1 × 10⁷ cells as the recommended Phase II dose.

Headache was the most common adverse event; two patients experienced grade 3 headaches.

Median overall survival was 16.24 months (95% CI, 11.21–not reached).

CSF profiling demonstrated increased Th1-associated cytokines, including IFN-γ, IL-6, and TNF-α, without a corresponding increase in IL-4.

Transcriptomic analyses showed enrichment of adaptive immune programs, including cytotoxic T-cell activity and antibody-secretion pathways.

B-cell influx into the CSF and preliminary evidence of HER2-dependent cellular cytotoxicity provided additional signs of immune activation.

The significance extends beyond delivering an immune therapy into the CNS. The approach appears to reshape the local immune environment, promoting coordinated T-cell and B-cell responses where adaptive immunity is typically limited.

This remains an early-phase, single-arm study, and the survival findings cannot establish clinical efficacy. Nevertheless, identifying a recommended Phase II dose and demonstrating measurable immune remodeling provide an important foundation for further investigation.

A compelling direction for cellular immunotherapy: not simply targeting tumor cells, but reprogramming the immune ecosystem surrounding them.

#BreastCancer #LeptomeningealDisease #DendriticCells #CellTherapy #CancerImmunotherapy #NeuroOncology #TranslationalResearch

jitc.bmj.com/content/14/Sup…

From Cancer Data to Discovery: Advancing the Future of AI in Oncology

The future of cancer care will be shaped not only by the therapies we develop, but by how effectively we harness data, leverage artificial intelligence, and translate innovation into meaningful opportunities for patients worldwide.

It was a tremendous honor to be invited as a Featured Speaker at the National Cancer Institute's 4th Annual Office of Data Sharing Symposium, where I had the opportunity to present our work exploring the utility of AI-driven, mCODE-aligned frameworks to facilitate clinical trial matching and advance oncology data sharing.

My sincere gratitude to Dr. Ying Huang for the gracious invitation, and to Drs. Huang and Mousumi Ghosh for their exceptional leadership in organizing such an outstanding forum and bringing together a remarkable group of speakers at the forefront of cancer research, AI, and data science.

I am especially grateful to Dr. Bonilla for his continued mentorship, guidance, and belief in the potential of our work, and to the incredible MassiveBio team for their tireless efforts to bridge the gap between scientific innovation and patient access to clinical trials.

One of the most rewarding aspects of the symposium was the opportunity to connect with pioneering leaders in AI, including Dr. Caroline Chung, as well as Drs. Kelli Rasmussen, Satish Viswanath, Xueli Liu, and Jaime Guidry Auvil as we exchanged perspectives on where the intersection of AI, oncology, and data science is heading. These conversations reinforced an important conviction: the true measure of innovation is not simply what technology can accomplish, but how many lives it can meaningfully impact.

As we look toward the next generation of cancer care, I believe our greatest opportunity lies in building a future where AI-driven discoveries transcend institutional, geographic, and socioeconomic boundaries, bringing precision oncology and clinical trials within reach of patients everywhere.

I leave this symposium inspired by the extraordinary work being done across the field, grateful for the opportunity to contribute to these conversations, and even more committed to helping translate these advances into equitable, real-world impact.

The future of oncology will not be built in isolation. It will be built through collaboration, shared knowledge, and a collective commitment to ensuring that innovation reaches every patient who needs it.

#NCI #ArtificialIntelligence #Oncology #ClinicalTrials #PrecisionMedicine #CancerResearch #HealthEquity #DataScience #CancerInnovation

Can bispecific antibodies do more than control disease before CAR-T? Could they help set the stage for a deeper and more durable response?

For patients with relapsed/refractory large B-cell lymphoma (LBCL), sequencing CD19-directed CAR-T and CD20×CD3 bispecific antibodies remains an important clinical question. More than 80% of patients undergoing CAR-T require bridging therapy, yet options for patients whose disease progresses despite chemotherapy-based bridging remain limited.

A recent two-center French study examined CD20×CD3 bispecific antibodies—glofitamab or epcoritamab—as bridging therapy before CAR-T in 23 patients with R/R LBCL.

Notably, 78% had already failed an initial chemotherapy-based bridging strategy, and 87% were refractory to their most recent systemic treatment.

The findings were encouraging:

82.6% responded to bispecific antibody bridging, including 30.4% complete responses (CR).
20 of 23 patients proceeded to CAR-T infusion.
85% achieved CR after CAR-T.
Median follow-up was 12.2 months, with median PFS and OS not reached.

One observation stood out: all patients who achieved CR before CAR-T maintained their CR during follow-up. In contrast, all three patients who relapsed after achieving post-CAR-T CR had entered infusion in partial response.

The safety profile was also notable. No grade ≥3 CRS occurred after CAR-T, although grade 3–4 ICANS occurred in 15% of patients, all following axi-cel.

Importantly, the median vein-to-vein time was 2.9 months—longer than typically reported for CAR-T—yet outcomes remained encouraging.

These findings raise an intriguing possibility: bispecific antibody bridging may not simply buy time while CAR-T is manufactured. By achieving deeper disease control before infusion, it may help optimize subsequent CAR-T outcomes.

However, this was a small, retrospective cohort, and the association between preinfusion CR and durable remission does not establish that bispecific bridging improves CAR-T efficacy. Prospective validation is needed.

The next question is not only which therapy should come first, but whether the depth of response before CAR-T can help determine the durability of response afterward. In addition, does using this approach contribute to increased frequency of antigen loss?

#CAR #BispecificAntibodies #LBCL #DLBCL #Lymphoma #CellTherapy #Hematology

ashpublications.org/bloodict/artic…