High Dimensional Interference for Biomarker Identification: An Application to CAR-T Cell Immunotherapy

Chimeric antigen receptor T cell (CAR-T) immunotherapy is improving the therapeutic landscape for patients with relapsed or refractory B cell malignancies, yet a significant portion of patients will not respond or might relapse after an initial response. Biomarker identification would be useful to 1) better predict outcomes after CAR-T therapy; and 2) provide insights into the underlying biology of the failure or success of these therapies, thereby revealing potential opportunities for further improvements. In CAR-T trial, different technologies have been used to collect biomarker data (factors) and clinical outcomes, such as flow cytometry and quantitative polymerase chain reaction (qPCR), that when combined with the multiple efficacy and toxicity endpoints, rapidly increases the complexity of data analysis in CAR-T trial. As most of the immunotherapy trials are still in the stage of phase I/II, the sample sizes (n) are typically limited but can generate huge amounts of biologic data (p covariates). In this situation, conventional statistical models, such as stepwise or penalized regression, might perform poorly. Thus, this project will use a cutting-edge high dimensional inference (HDI) method to identify biomarkers (high dimensional) associated with clinical outcomes (efficacy and toxicity), and provide statistical inference on all possible covariates (not only the selected ones). HDI can deal with "large p small n" situations, handle multicollinearity, and provide “de-biased” estimates for coefficients. Importantly, HDI allows computation of confidence intervals and p-values without refitting, enabling easier interpretation by researchers. To our knowledge, this will be the first study to evaluate HDI for identifying factors associated with outcomes after CAR-T therapy. Successful completion of our study will likely advance useful new strategies and statistical tools to identify biomarkers associated with clinical outcomes for CAR-T therapies, which may thereby profoundly increase the success of CAR-T therapy overall and make a broad impact across medical research.