PET Radiomics Features Predictive of Outcomes Following CAR-T in Patients with LBCL

Chimeric antigen receptor T-cell (CAR-T) therapy has shown promising efficacy in patients with relapsed or refractory (R/R) large B-cell lymphoma (LBCL), although responses and survival probability are highly variable. High tumor burden before CAR-T administration has been associated with inferior survival, and recently, radiomic features extracted from PET/CT scans have shown to be predictive of outcomes in patients with LBCL treated with conventional chemotherapy; however, such data in the CAR-T therapy setting remain scarce. Findings from a single-institution study of PET-based radiomic features in relation to clinical outcomes in patients with R/R LBCL receiving CAR-T therapy were presented at the 64th American Society of Hematology 2022 Annual Meeting and Exposition.

Consecutive patients with R/R LBCL treated with commercially approved CAR-T products were included in this study, with PET/CT scans performed within a week of starting lymphodepleting chemotherapy. Lesions were delineated using a semiautomated preselection of the ¹⁸F-FDG–avid lesions. Quantifiable radiomic features from the standard uptake values (SUVs) of PET images were extracted from the largest lesion, along with conventional PET features, such as total metabolic tumor volume (TMTV) and maximum SUV (SUV_max). Data were split into training (80%) and validation (20%) sets (balanced for treatment response, type of construct, and international prognostic index [IPI] score) to identify potential PET-based radiomics variables associated with progression-free survival (PFS). Feature selection was performed using the least absolute shrinkage and selection operator (LASSO) regression. A logistic regression model was trained and tested for predicting clinical benefit defined as PFS >3.2 months. The performance of the novel radiomics score (RADscore) was compared to conventional PET features for predicting clinical benefit using the area under the curve (AUC) from the receiver operating characteristic curve. Univariate and multivariate survival Cox models were evaluated for continuous PFS and overall survival including the selected variables from LASSO-Cox models.

Of 93 patients included in the study, median age was 59 years (interquartile range, 50-68), 68% were men, and 32% had >2 prior lines of therapy. Seventy-three percent of patients had advanced disease (stage III-IV), and 32% had prior stem-cell transplantation. Thirty-three percent of patients received CAR-T with a CD28 costimulatory domain and 67% with a 4-1BB domain. Four features constituted the RADscore; RADscore was predictive of clinical benefit in both the training and validation sets, with AUCs of 0.79 (95% confidence interval [CI], 0.68-0.89) and 0.73 (95% CI, 0.48-0.98), respectively. RADscore outperformed conventional markers SUV_max and TMTV both individually and in combination. In the multivariate analysis, RADscore was associated with higher PFS (hazard ratio, 0.09; 95% CI, 0.03-0.25; P <.001). In the univariate analysis, lactate dehydrogenase (P = .006), IPI score (P = .005), TMTV (P = .003), SUV_max (P = .022), and RADscore (P <.001) were all significantly associated with PFS.

In conclusion, prediction models using ¹⁸F-FDG PET/CT radiomic features at baseline aid in identifying patients with LBCL benefiting from CD19-targeted CAR-T therapy. The newly developed RADscore outperformed conventional PET parameters and showed independent predictive value after adjusting for known clinical factors in patients with LBCL.

Source

Barba P, Ligero M, Carpio C, et al. PET radiomic features are predictive of outcomes after chimeric antigen receptor T-cell therapy in patients with large B-cell lymphoma. Presented at: 64th American Society of Hematology Annual Meeting and Exposition, December 10-13, 2022; New Orleans, LA. Poster presentation 1643.