In making the most important decisions for cancer care, clinicians naturally consider their patients' evolving characteristics and risks factors and tailor the treatments or diagnostic tests accordingly. In other words, they consider dynamic treatment strategies that depend on the evolution of one or more time-dependent covariates (e.g. functional status, frailty, new comorbidities) as opposed to static treatment strategies which do not. Despite the clinical relevance of dynamic treatment strategies, most research continues to compare static strategies. This is because, in part, randomized clinical trials comparing dynamic strategies can be particularly expensive, time-consuming, and inefficient. Additionally, observational studies of dynamic strategies require specific methods because even in the absence of unmeasured confounding, conventional estimates fail to have a causal interpretation when (i) there exists a measured time-varying prognostic factor that also predicts subsequent treatment, and (ii) past treatment predicts subsequent prognostic factor level. This treatment- confounder feedback is pervasive in biomedical research. This application's broad objective is to create an innovative research program that integrates data from diverse sources such as electronic medical records, administrative claims, epidemiologic cohorts (and in the future wearable health technology and patient portals) with advanced epidemiological methods to study dynamic strategies of health care delivery and create publicly available analytic tools to accelerate discovery and support a more strategic delivery of cancer care.
The specific aims i nvolve training in open-source programming, data science and advanced epidemiological methods. These skills will be used to develop three projects that will use different sources of data (a traditional epidemiological cohort, a linkage of tumor registries with claims and electronic medical records) to address questions on prevention, surveillance and treatment of cancer: ? What is the best dynamic strategy for colorectal cancer screening? ? What is the best dynamic strategy for bladder cancer surveillance? ? What is the best dynamic strategy of treatment for prostate cancer patients with a PSA-only relapse? These projects will serve as a backbone to develop and fine tune the specific methods required to study dynamic strategies of health care delivery: the g-methods. Specifically, this application will focus on the application and dissemination of Marginal Structural Models and the parametric g-formula for clinical cancer research.

Public Health Relevance

The proposal addresses the systematic application and dissemination of cutting-edge methods (g- methods) for causal inference to clinical cancer research using complex sources of data. G-methods are particularly useful for answering clinical questions where treatments, patient characteristics and outcomes change over time (which is virtually always the case). By tailoring these methods to cancer research and developing open-source analytic platforms, we plan to spread their use across the scientific community and ultimately benefit patient care.

Agency
National Institute of Health (NIH)
Institute
National Cancer Institute (NCI)
Type
Career Transition Award (K99)
Project #
1K99CA207730-01A1
Application #
9314154
Study Section
Subcommittee I - Transition to Independence (NCI-I)
Program Officer
Radaev, Sergei
Project Start
2017-03-07
Project End
2019-02-28
Budget Start
2017-03-07
Budget End
2018-02-28
Support Year
1
Fiscal Year
2017
Total Cost
$137,137
Indirect Cost
$10,158
Name
Harvard University
Department
Public Health & Prev Medicine
Type
Schools of Public Health
DUNS #
149617367
City
Boston
State
MA
Country
United States
Zip Code
02115
Emilsson, Louise; García-Albéniz, Xabier; Logan, Roger W et al. (2018) Examining Bias in Studies of Statin Treatment and Survival in Patients With Cancer. JAMA Oncol 4:63-70
García-Albéniz, Xabier; Hsu, John; Hernán, Miguel A (2017) The value of explicitly emulating a target trial when using real world evidence: an application to colorectal cancer screening. Eur J Epidemiol 32:495-500