Risk prediction equations are widely used in cardiovascular medicine for risk stratification and to determine cost-effective and appropriate courses of treatment. Whether new predictors can add clinical utility to established models such as the Framingham risk score is an important question. Previous work by these investigators has introduced new methods for comparing risk prediction models, including risk reclassification, which assesses the ability of new models to more accurately classify individuals into higher or lower risk strata. This proposal offers several novel extensions of these methods, particularly related to calibration, which directly compares the predicted to the observed risk. Such calibration is essential for estimating risk for the individual and computing differences in absolute risk. While measures of improvement in discrimination are available, this application proposes an integrated approach examining improvement in calibration. Also, since pre-specified risk strata are not identified for all applications, a category-free measure of reclassification calibration is proposed. A second area of interest is extending methods for reclassification calibration to other study designs. While methods for survival data are available for discrimination measures, how well measures of calibration, particularly the reclassification chi-square statistic, extend to survival settings is not yet known. In addition, case-cohort and matched case-control studies are widely used in cardiovascular research, especially for biomarker assessment. It is not known how well reclassification measures translate to these designs. Finally, to limit costs, clinicians are often interested in multi-stage screening of diseae. Reclassification of those at intermediate risk is of most interest, and the 'clinical NRI'has been introduced to assess reclassification in this group. How well this new method performs in practice remains to be determined. We propose to conduct a portfolio of research projects regarding the comparison of predictive models in general, and reclassification calibration specifically, that will further the development of these methods and their clinical utility. We wil examine the characteristics of the new measures in data on cardiovascular disease from the Women's Health Study, in simulations, and in existing case-control and case-cohort data on CVD outcomes. Extensions to these novel applications would be highly innovative and directly and immediately applicable to clinical risk prediction for cardiovascular disease.

Public Health Relevance

Risk prediction equations are widely used in cardiovascular medicine for risk stratification and to determine cost-effective and appropriate courses of treatment. This project proposes to develop new methods for comparison of risk prediction models, particularly related to reclassification calibration. It will extend these methods to other study designs and settings other than binary outcomes.

Agency
National Institute of Health (NIH)
Institute
National Heart, Lung, and Blood Institute (NHLBI)
Type
Research Project (R01)
Project #
5R01HL113080-02
Application #
8607472
Study Section
Cardiovascular and Sleep Epidemiology (CASE)
Program Officer
Burwen, Dale R
Project Start
2013-01-25
Project End
2016-12-31
Budget Start
2014-01-01
Budget End
2014-12-31
Support Year
2
Fiscal Year
2014
Total Cost
$278,119
Indirect Cost
$120,619
Name
Brigham and Women's Hospital
Department
Type
DUNS #
030811269
City
Boston
State
MA
Country
United States
Zip Code
02115
Paynter, Nina P; Everett, Brendan M; Cook, Nancy R (2014) Cardiovascular disease risk prediction in women: is there a role for novel biomarkers? Clin Chem 60:88-97
Paynter, Nina P; LaMonte, Michael J; Manson, JoAnn E et al. (2014) Comparison of lifestyle-based and traditional cardiovascular disease prediction in a multiethnic cohort of nonsmoking women. Circulation 130:1466-73