2026 ASH®
Annual Meeting & Exposition December 12-15, 2026 ▪ New Orleans, Louisiana

Paper Details

Abstracts, authors, affiliations, and disclosures.

Sunday, December 13, 2026
09:30 a.m. - 11:00 a.m.
MCCNO - La Nouvelle C
Technology and Procedures, Omics Technologies, Pathology, Serologic Tests, Bioinformatics, Computational Biology, Emerging Technologies, Treatment Considerations, Therapy Sequence, Biological Therapies, Bispecific Antibody Therapy, Chimeric Antigen Receptor, (CAR)-T Cell Therapies, Immunotherapy, Artificial Intelligence (AI), Deep Learning, Machine Learning, Biological Processes, Pathogenesis, Genomics, Immune Mechanism, Multi-Systemic Interactions, Diseases, Lymphoid Malignancies, Plasma Cell Disorders, Measurable Residual Disease, Research, Fundamental Science, Translational Research, Clinical Research, Real-World Evidence, Study Population, Human, Adult
PAPER

547 Multimodal standard-of-care profiling reveals biological drivers of early treatment failure after BCMA CAR-T therapy in multiple myeloma

Myles Kim, PhD, Robert Norberg2, Omkar Bhabad3, Larry Grullon-Polanco, N/A4, Adam Duca, BS5, Dietrich Werner, MD6, Emily C. Merritt7, Daniel De Avila8, Gabriel De Avila9, Omar Alexis Castaneda Puglianini, MD10, Hien Liu, MD11, Melissa Alsina, MD12, Taiga Nishihori, MD13, Rachid Baz, MD14, Brandon Blue, MD15, Ariel Grajales-Cruz, MD16, Doris K. Hansen, MD17, Reginald Atkins, PhD18, Meghan A. Menges19, Luis A. Cuadrado, MSc20, Rolf Petter Tonseth21, Justyn Y Nakashima, MD22, Nicholas Figura, MD23, Praneeth Reddy Sudalagunta, PhD24, Ariosto Siqueira Silva, PhD25, Kenneth H. Shain, MD, PhD26, John L Cleveland, PhD27, Frederick L. Locke, MD28, Ghulam Rasool, PhD29, Ciara Louise L Freeman, MD, PhD30
1 Moffitt Cancer Center, Tampa, Florida, United States
2 Moffitt Cancer Center, Tampa, Florida, United States
3 Moffitt Cancer Center, Tampa, Florida, United States
4 Moffitt Cancer Center, Tampa, Florida, United States
5 Moffitt Cancer Center, Tampa, Florida, United States
6 Moffitt Cancer Center, Tampa, Florida, United States
7 Moffitt Cancer Center, Tampa, Florida, United States
8 Moffitt Cancer Center, Tampa, Florida, United States
9 Moffitt Cancer Center, Tampa, Florida, United States
10 Moffitt Cancer Center, Tampa, Florida, United States
11 Moffitt Cancer Center, Tampa, Florida, United States
12 Moffitt Cancer Center, Tampa, Florida, United States
13 Moffitt Cancer Center, Tampa, Florida, United States
14 Moffitt Cancer Center, Tampa, Florida, United States
15 Moffitt Cancer Center, Tampa, Florida, United States
16 Moffitt Cancer Center, Tampa, Florida, United States
17 Moffitt Cancer Center, Tampa, Florida, United States
18 Moffitt Cancer Center, Tampa, Florida, United States
19 Moffitt Cancer Center, Tampa, Florida, United States
20 Moffitt Cancer Center, Tampa, Florida, United States
21 BC Cancer Agency, Vancouver, British Columbia, Canada
22 Moffitt Cancer Center, Tampa, Florida, United States
23 Moffitt Cancer Center, Tampa, Florida, United States
24 Moffitt Cancer Center, Tampa, Florida, United States
25 Moffitt Cancer Center, Tampa, Florida, United States, Moffitt Cancer Center, Tampa, Florida, United States
26 Moffitt Cancer Center, Tampa, Florida, United States
27 Moffitt Cancer Center, Tampa, Florida, United States
28 Moffitt Cancer Center, Tampa, Florida, United States
29 Moffitt Cancer Center, Tampa, Florida, United States
30 Moffitt Cancer Center, Tampa, Florida, United States

Introduction

Risk assessment before CAR-T can be limited by models incorporating only selected components of routine standard of care (SOC) data and may miss biologically relevant signals.  We hypothesized that integrating multimodal SOC features would improve risk discrimination and identify clinically interpretable drivers of early treatment failure (ETF).

Methods

We assembled a cohort of BCMA CAR-T recipients (n=688; train 481, test 207) and integrated routine pre-infusion SOC data at two timepoints: apheresis (APH) and lymphodepletion (LD). The dataset included tumor burden, cytogenetics, peripheral T-cell enumeration, inflammatory markers, and longitudinal laboratory dynamics between APH and LD. Active extramedullary disease (EMD), metabolic tumor volume (MTV), SOC plasma cell (PC) flow and soluble BCMA (sBCMA, n=313) were also included. ETF was defined as progression or death within 365 days after CAR-T infusion, representing failure to achieve durable disease control. CatBoost survival models estimated time to ETF using 40 APH and 55 LD features. Performance was benchmarked against SCOPE, CAR-HEMATOTOX, MyCARe, and InflaMix using bootstrap C-index, 365-day net reclassification improvement (NRI), calibration, and decision-curve analyses. SHAP was used for model explainability. Models were then applied to ide-cel treated CIBMTR (n=788) and cilta-cel treated CARTITUDE-1 (n=95) cohorts.

Results

Multimodal models achieved median test C-indices of 0.739 (95% CI 0.688-0.788) at LD and 0.707 (95% CI 0.646-0.764) at APH, outperforming all benchmarks. Discrimination improved over the best comparator at each timepoint, from 0.648 to 0.707 at APH versus SCOPE (ΔC=+0.066, p<0.001) and from 0.707 to 0.739 at LD versus MyCARe-LD (ΔC=+0.027, p<0.001). NRI improved over all benchmarks by at least 0.123, driven primarily by correct up-classification of patients that experienced ETF but assigned lower risk by comparator scores. Decision curve analysis also demonstrated superior net benefit across clinically relevant thresholds with preserved calibration. Despite differences in available features, models retained discrimination in CIBMTR and CARTITUDE-1 with C-indices 0.636 and 0.757, respectively. SHAP, a method that estimates each feature’s contribution to prediction, identified product, MTV, EMD, gain/amp(1q), platelet count, hemoglobin, red-cell distribution width and CD4+ count at APH as dominant ETF predictors.

Further biologic interrogation identified three orthogonal drivers. First, gain/amp(1q) was a surrogate of chromosomal complexity rather than an isolated cytogenetic lesion: 1q-positive tumors harbored substantially greater co-lesion burden (1.46 vs 0.36; p<0.001), nearly 10-fold more likely to exhibit multi-hit disease ( ≥2 co-occurring high-risk FISH lesions beyond 1q; OR 8.6), and demonstrated progressively increased ETF risk with escalating lesion burden. Second, tumor burden was a dominant driver. Among patients with sBCMA, bone marrow PC and MTV, higher sBCMA predicted ETF (OR 1.57 per SD, p=0.032), but lost significance after adjusting for PC percentage (OR 1.35, p=0.123). MTV was the strongest single discriminator (AUC 0.70 vs 0.62) but requires PET, thus sBCMA offers a practical, low-cost surrogate of total tumor burden when other measures are unavailable. Third, PC differentiation independently influenced outcomes. CD200 loss on SOC flow was associated with ETF after adjustment for disease features, EMD, and CAR-T product, identifying a high-risk differentiation state. Conversely, CD117 expression identified a subgroup with more favorable outcomes, even among patients with gain/amp(1q) disease (median PFS 831 vs 434 days). These findings were not materially modified by the type of CAR-T product, supporting plasma-cell differentiation as a biologically distinct axis of treatment response.

Conclusions

In one of the largest deeply characterized BCMA CAR-T treated cohorts reported to date, a multimodal model improved discrimination over established risk scores and retained performance in external validation cohorts. Beyond prediction, the analysis identifies three interpretable axes of failure: chromosomal complexity, total disease burden and plasma-cell differentiation state. These findings support biology-informed risk assessment before CAR-T and provide a rationale for prospective studies testing risk-adapted bridging and monitoring strategies using routinely available clinical data.

Ciara Louise L Freeman, MD, PhD: AstraZeneca - Research Funding; Bristol Myers Squibb - Research Funding