Bayesian Methods for Complex Precision Biotherapy Trials in Oncology

点击翻译按钮获取中文摘要

Project Summary/Abstract Most clinical trial designs use \one-size- ts-all" rules for treatment assignment and evaluation based on models that ignore patient heterogeneity. This is disconnected from medical practice, where physicians use each patient's diagnosis and prognostic variables to make personalized, precision medicine treatment decisions. Modern precision medicine exploits biotechnologies such as proteomics, genomics, gene sequencing, mass spectrometry, or cytometry methods that evaluate multiple cell surface markers. These generate vectors of biomarkers that may be used to re ne existing disease subgroup de nitions, construct new disease classi cations, and formulate clinical trial designs and statistical rules for personalized/precision treatment assignment. In oncology and other disease areas, there is rapidly increasing development of new biotherapies, including cell therapies, immunotherapies, and targeted molecular agents. A biotherapy may be administered once or in multiple cycles; used in combination with conventional treatments such as cytotoxic chemotherapy, radiation, or surgery; and often generates complex outcomes, such as repeatedly evaluated tumor status, multiple biological variables, and occurrence times of both early and late onset toxicities. This complicates the de nitions of \response" and \toxicity," and produces multidimensional treatment e ects that may di er between subgroups. An example is a phase I-II trial to optimize subgroup-speci c doses of donor derived natural killer (NK) cells for treating B-cell hematologic malignancies, where donated NK cells are engineered using chimeric antigen receptors to enhance their cancer killing e ects, then expanded using growth factors to obtain cell doses large enough for therapeutic use. Subgroups may be de ned using disease subtypes and prognostic variables. Co-primary outcomes may include ordinal disease status, including complete or partial remission, stable disease, or disease progression, evaluated either once or at monthly intervals; time to severe NK cell-related toxicity, such as cytokine release syndrome; and a binary indicator of 100-day survival. Considering (biotherapy, dose, administration schedule) a treatment regime, a clinical trial of one or more new biotherapies may include a subgroup-speci c risk-bene t tradeo based dose or schedule optimization for each biotherapy, randomization among regimes restricted to achieve balance within subgroups, and subgroup-speci c group sequential rules to select superior regimes or drop unsafe or ine ective regimes. The proposed research will construct robust Bayesian regression models for regime-outcome e ects that account for patient heterogeneity, including possible regime-subgroup interactions. These will be the basis for sequential decision making and regime assignment, and they may include latent variables to adaptively combine subgroups with similar regime-outcome e ects. Each clinical trial design will be tailored to address a combination of these goals in speci c biotherapy settings. For each design, user-friendly computer software will be provided, including programs for trial simulation to establish design operating characteristics, trial conduct, and use by practicing physicians to choose optimal regimes for their patients. The overarching goal of the proposed research is to develop and identify optimal personalized biotherapy regimes, spanning a variety of di erent diseases and clinical settings, for greater anti-disease e ects, increased safety, and improved survival.