Statistical Methods for Integration of Multiple Data Sources toward Precision Cancer Medicine

点击翻译按钮获取中文摘要

Project Summary: The primary objective of this research is to develop novel statistical and computational tools to evaluate new and existing cancer therapies for precision cancer medicine, with a principal focus on integrating multiple data sources including randomized controlled trials (RCT) and real world data (RWD). All of the aims are motivated by multidisciplinary collaboration. Evidence-based clinical decision making involves synthesizing available research evidence from multiple resources, including RCT and RWD. Pivotal RCTs are the primary evidence that established the oncologic equivalence or efficacy of local and systemic treatments. However, a recent systematic review found little agreement between population-based RWD and RCTs when comparing the same oncologic treatment regimens. This difference is thought to stem from the highly selective criteria used for trial enrollment coupled with the rapidly changing nature of multidisciplinary cancer care. Moreover, heterogeneous treatment effects by disease biologic tumor subtype on survival outcomes has not been examined sufficiently in early RCTs. We will develop statistical tools and software to evaluate the agreement of findings from RCTs and the real-world patient population, reassessing standard treatment guidelines on local- regional therapies for early-stage breast cancer by patients’ clinical and tumor subtypes. While the proposed methodology is agnostic to disease type, we will use breast cancer patients as proof of principle for the approaches proposed. The specific aims are: (1) to estimate and assess the agreement of treatment efficacy on survival outcomes across multiple studies (e.g., RCT and RWD) using nonparametric calibration weights to adjust for treatment selection bias and heterogeneity between studies; (2) to test the existence of a subgroup of patients with enhanced treatment effect and predict subgroup membership of a treatment using a semi-parametric isotonic- Cox model, and to develop a concordance-assisted learning tool for threshold identification to guide patient treatment selection; (3) to infer the treatment effects on breast cancer-specific survival when the cause of death is unknown in RWD by integrating data from RCT and RWD; (4) to estimate treatment effect for rare subtypes of breast cancer by combining external aggregate data with individual-level data to improve inference efficiency; and (5) to develop and disseminate publicly available, user-friendly software and facilitate the reproducibility and applications of our methods to multiple existing databases, including large-population-level data and RCT data for breast cancer research. The proposed research will advance general methodologic development in comparative effectiveness and precision medicine research by efficiently integrating multiple data sources. More importantly, the study findings could improve evidence-based treatment recommendations, better informing clinicians to select optimal treatments according to patients’ tumor subtypes and other characteristics, thus furthering clinical care via better integration of clinical science.