Anticlotting drugs reduce risk to thrombosis and treat conditions that might lead to stroke, pulmonary embolism, deep vein thrombosis or other blood clotting related disease. The impact and value of anticlotting medication in the U.S. is dramatic. For example, stroke is the third leading cause of death in the U.S. with over 140,000 deaths annually. The majority of stroke incidences are due to ischemia (87%) or transient ischemic attack (TIA, ~5-10%) and are typically managed by the use of anticlotting drugs including anticoagulants (e.g., warfarin and dabigatran) and antiplatelets (e.g., clopidogrel). Whatever the patient's disease or condition leading to a prescription of an anticlotting agent, selecting the bet combination of drug and treatment protocol is complicated by the individual differences in anticlotting drug response due to genetics (e.g. >20-fold difference for warfarin), physiology, and compliance. In practice, providers use a combination of experience, scientific evidence and clinical trial results to develop anticlotting "best practice" treatment plans designed to roughly minimize the patient-to-patient response variability and risks across the provider's patient population. However, the high degree of patient heterogeneity causes variations in individual patient response to these "best practice" drug-protocol approaches. In short, no practical optimal anticlotting treatment plan exists for large heterogeneous patient populations that accounts for individual risk factors;drug and protocol options;and achieves minimal risk to stroke. Access to large comprehensive electronic medical records (EMR) covering diverse patient populations, coupled with novel modeling and computational simulations provides an unprecedented opportunity to conduct in silico identification and validation of optimal anticlottin treatment strategies. We propose a novel computational approach that uses individual patient data and outcome evidence from two large electronic medical record (EMR) databases to conduct side-by-side clinical simulations comparing outcomes for two or more anticlotting drug and dose protocols. The approach first converts EMR data to EMR- based simulated data that reflects the statistical and individual characteristics of the EMR population. We then apply advanced treatment simulation methods to predict outcomes and costs of multiple drug-dosing protocols. Finally, we apply an optimization approach to identify the optimal treatment plans for segments of the population (e.g. the African American segment, white females over 50 segment, ...). Finally, we will conduct in silico tests of the robustness and validation of the predicted optimal anticlotting treatment plan. This approach, promises to provide the first environment in which side-by-side anticlotting clinical simulations and outcome predictions for an entire population based on existing EMR data sets can be calculated, compared and contrasted. Such predictive evidence can then be used to guide clinical trial designs, and suggest improvements to hospital-wide anticlotting treatment plans.

Public Health Relevance

Every hospital and clinic in the US faces increasing complexity of choices of anticlotting medications and evidence supporting various treatment protocols. Failure to optimize the anticlotting treatment plan across a large physiologically and genetically diverse population increases the risk of serious adverse events to inaccurate dosing such as stoke or hemorrhage. Our project will develop and validate a new method and tool to guide the design and test the efficacy of predicted optimal hospital-wide treatment plans that reduce risk, improve outcomes and maintain if not reduce overall costs.

National Institute of Health (NIH)
Research Project (R01)
Project #
Application #
Study Section
Biomedical Library and Informatics Review Committee (BLR)
Program Officer
Sim, Hua-Chuan
Project Start
Project End
Budget Start
Budget End
Support Year
Fiscal Year
Total Cost
Indirect Cost
Harvard Medical School
Schools of Medicine
United States
Zip Code
Gafni, Erik; Luquette, Lovelace J; Lancaster, Alex K et al. (2014) COSMOS: Python library for massively parallel workflows. Bioinformatics 30:2956-8