Treatment of cancer is an ongoing process during which clinicians make a series of therapeutic decisions over the course of the disease. However, while there is increasing interest in identifying the overall strategy of sequential decisions leading to the most beneficial clinical outcomes, where those decisions may be predicated on complex information on the patient up to that point, current cancer clinical trials evaluate only the therapeutic options available at a single decision point, mostly in a """"""""one-size-fits-all"""""""" manner. Attempts to synthesize information from several isolated trials conducted at different milestones in the disease are problematic, because the best treatment at any one decision point may not be best when placed in the context of the entire decision process owing to possible delayed effects of past treatments on the efficacy of future treatments. Considering cancer treatment strategies as dynamic treatment regimes, which are formal algorithms for sequential decision-making that use accrued information on the patient at each decision point in an evidence-based manner to determine the next step of treatment, along with analytical reinforcement learning methods from computer science that provide a principled framework for identifying the optimal such regime, offers the potential to revolutionize how cancer treatment is viewed and effect a paradigm shift in the design and conduct of cancer clinical trials. The four specific aims of this project seek to catalyze this advance by studying these issues for the first time in the cancer treatment context.
The first aim will evaluate various learning methods to establish the best techniques for use in developing optimal dynamic treatment regimes for cancer, and the second will focus on a specific version of this methodology when clinicians are interested in finding the best regime among a particular set of regimes.
The third aim will develop new methods for making formal statistical inference on regimes developed based on data, which have been heretofore unavailable owing to the theoretical complexity of the problem. In the fourth aim, methods for design of so-called sequentially randomized trials for the specific purpose of developing dynamic treatment regimes, including determination of sample sizes that will ensure identification ofthe best regimes from among those in the trial, will be developed. Coupling trial design with learning methods for analysis, a new model, the clinical reinforcement trial, will be developed and applied to designing studies to identify optimal regimes for non-small cell lung cancer and other cancers. Collectively, these aims will result in high-impact, new methodology that will allow individualization of the therapy to the patient over time.

Public Health Relevance

Although treatment of cancer involves a series of therapeutic decisions over time, cancer clinical trials evaluate treatments only at specific decision points, and hence the best treatment in such a trial may not be best when placed in the context of the overall decision-making process. This research will study cancer treatment formally as an overall, individualized strategy so that the entire series of decisions leading to the best outcomes can be determined, promoting a paradigm shift in the way cancer therapies are evaluated.

National Institute of Health (NIH)
National Cancer Institute (NCI)
Research Program Projects (P01)
Project #
Application #
Study Section
Special Emphasis Panel (ZCA1)
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
University of North Carolina Chapel Hill
Chapel Hill
United States
Zip Code
Wang, Xiaofei; Wang, Xiaoyi; Hodgson, Lydia et al. (2017) Validation of Progression-Free Survival as a Surrogate Endpoint for Overall Survival in Malignant Mesothelioma: Analysis of Cancer and Leukemia Group B and North Central Cancer Treatment Group (Alliance) Trials. Oncologist 22:189-198
Zhang, Danjie; Chen, Ming-Hui; Ibrahim, Joseph G et al. (2017) Bayesian Model Assessment in Joint Modeling of Longitudinal and Survival Data with Applications to Cancer Clinical Trials. J Comput Graph Stat 26:121-133
Kang, Suhyun; Lu, Wenbin; Song, Rui (2017) Subgroup detection and sample size calculation with proportional hazards regression for survival data. Stat Med 36:4646-4659
Zhao, Jingkang; Li, Dongshunyi; Seo, Jungkyun et al. (2017) Quantifying the Impact of Non-coding Variants on Transcription Factor-DNA Binding. Res Comput Mol Biol 10229:336-352
Silva, Grace O; Siegel, Marni B; Mose, Lisle E et al. (2017) SynthEx: a synthetic-normal-based DNA sequencing tool for copy number alteration detection and tumor heterogeneity profiling. Genome Biol 18:66
Stinchcombe, Thomas E; Zhang, Ying; Vokes, Everett E et al. (2017) Pooled Analysis of Individual Patient Data on Concurrent Chemoradiotherapy for Stage III Non-Small-Cell Lung Cancer in Elderly Patients Compared With Younger Patients Who Participated in US National Cancer Institute Cooperative Group Studies. J Clin Oncol 35:2885-2892
Li, Zhiguo (2017) Comparison of adaptive treatment strategies based on longitudinal outcomes in sequential multiple assignment randomized trials. Stat Med 36:403-415
Ding, Jieli; Lu, Tsui-Shan; Cai, Jianwen et al. (2017) Recent progresses in outcome-dependent sampling with failure time data. Lifetime Data Anal 23:57-82
Liang, Baosheng; Tong, Xingwei; Zeng, Donglin et al. (2017) SEMIPARAMETRIC REGRESSION ANALYSIS OF REPEATED CURRENT STATUS DATA. Stat Sin 27:1079-1100
Kong, Dehan; Maity, Arnab; Hsu, Fang-Chi et al. (2017) Rejoinder to ""A note on testing and estimation in marker-set association study using semiparametric quantile regression kernel machine"". Biometrics :

Showing the most recent 10 out of 462 publications