Prediction Models for Personalized Therapy & Survival Analysis

Developing new statistical techniques for predicting success of stem cell transplants used to treat many blood disorders

Return to Listing

Full Project Name:Ensemble Prediction Models for Personalized Therapy & Survival Analysis using Large DatabasesPrincipal Investigator:Brent Logan, PhD, Institute for Health and EquityCo-Investigator:Bronwen E. Shaw, MBChB, PhD, Medicine; Purushottam W. Laud, PhD, Institute for Health and Equity; Rodney Sparapani, PhD, Institute for Health and EquityAward Amount:$20,000

Award Date

July2015

Project Duration:24 months

Project Description Narrative:

Hematopoietic stem cell transplant is a curative treatment for patients with many blood disorders. The outcomes, however, are highly depending on genetic factors in both the patient and the donor, as well as the interaction between these factors. The use of a donor who is not a perfect genetic match results in graft-versus-host-disease, a devastating immune system complication that reduces quality of life and can be deadly.

The availability of Big Data in biomedical applications has never been greater and continues to grow. Population health databases can link information on large numbers of patient characteristics, including genetic information, with clinical outcomes. These databases can be examined to develop better prediction models for patient outcomes and more effectively tailor treatment to individual patients by predicting how they will respond to different treatments.

Through this award, researchers aim to use data to develop new statistical techniques for predicting the success of stem cell transplants used to treat many disorders of the blood to better predict patient outcomes, improve the selection of donors, reduce the probability of graft-versus-host-disease, and increase patients' post-transplant quality of life.

Outcomes & Lessons Learned:

Developed new statistical methods that better predict patient outcomes. Use of this new method has a likelihood of improving the selection of donors, reducing the probability of graft-versus-host-disease, and increasing patients' post-transplant quality of life.
Demonstrated that a method of generating individualized treatment rules, by focusing on improved predictions and using those to recommend treatments, may improve patient care and outcomes and is superior to leading methods available at this time
Conducted simulation studies, showing that this shift in approach has resulted in better performance in terms of expected patient outcomes
Developed software to implement this methodology so that it is available for a wider use
Applied the statistical methodology to the clinical problem of donor selection for bone marrow transplantation, finding that such an individualized donor selection strategy could reduce the absolute risk of severe graft-versus-host-disease or death within 180 days by about 5% compared to current donor selection strategies for matched unrelated donors
Conducted presentations at national conferences