Abstract 16298: Importance of Event vs. Nonevent Weighting and Baseline Risk Assessment in Risk Reclassification for Heart Failure: The Cardiovascular Health Study
Background: Net reclassification improvement (NRI), considers any correct upward movement of events and downward movement of nonevents (in terms of assigned risk) as model improvement. In the original definition, the impact of events and nonevents on NRI is weighted by event incidence. However, ruling-in may be more relevant than ruling-out disease risk in certain clinical scenarios, and vice versa in other situations. Also, reclassification benefit of a test may not be evenly distributed across risk categories. We tested the effect of NT-proBNP, echocardiographic variables, or both, for 5-year heart failure (HF) risk reclassification under a variety of event vs. nonevent relative importance scenarios and across baseline risk categories.
Methods: We classified 3752 Cardiovascular Health Study participants (age, 72.6±5.4 years; 40.8% men; 86.5% white), without HF at baseline, into <5%, 5-10%, 10-20%, and >20% 5-year HF risk based on the clinical Health ABC HF Risk Score and examined the effect of NT-proBNP, echocardiographic variables, or both, for risk reclassification. We calculated weighted NRI (wNRI) overall and in each risk category, assigning event to nonevent weights ranging from 1:1 (equal importance) to 16:1 (event prediction 16x more important) (Figure 1).
Results: Overall, wNRI with echocardiography ranged from 5.3% (1:1 weights) to 2.4% (16:1 weights), whereas wNRI with NT-proBNP was not materially affected by weights (5.4% to 5.6%). Addition of echocardiography after risk refinement with NT-proBNP yielded wNRI 4.0% to 4.5%. However, in risk category-specific wNRI analysis, reclassification was confined to intermediate risk (5%-20% 5-year HF risk) individuals (Figure 1), who derived substantial benefit (>20% wNRI) regardless of weights.
Conclusions: Event vs. nonevent weighting modulates reclassification yield when considering a new test. Use of a clinical model to identify intermediate risk individuals may optimize yield of new tests.
- © 2011 by American Heart Association, Inc.