Abstract 13321: Usefulness of Classification and Regression Trees (CART) Analysis to Determine Optimal Ranges of Clinical Parameters Associated With Prognosis after Acute Myocardial Infarction
Background: With conventional statistical strategies such as Receiver Operating Curve (ROC) methods, it is difficult to determine optimal target ranges in clinical parameters having U-shaped, but not linear, relationships with outcomes (e.g.; systolic blood pressure (sBP), non high density lipoprotein cholesterol (nHDL-C), and blood glucose (Glu)). In such settings, the Classification and Regression Trees (CART) analysis may be useful to explore target ranges in the clinical settings.
Methods: In consecutive 7,922 patients with first AMI registered to Osaka Acute Coronary Insufficiency Stusy (OACIS) between 1998 and 2011 (75% male, and mean age 65 years), we examined usefulness of the CART analysis to determine the optimal ranges of sBP, nHDL-C, and Glu in association with long-term outcomes after discharge.
Results: The CART analysis clearly identified the two splitting points for discriminating the cardiovascular risks as 111 and 131mmHg for SBP, 111 and 129 mg/dl for nHDL, and 81 and 182 mg/dl for Glu, respectively. As expected, patients with SBP of 111 to 131mmHg, nHDL of 111-129mg/dl, or Glu of 81-182mg/dl were significantly associated with the lowest cardiovascular risks determined by the Kaplan Meier curves and multivariate Cox regression analysis (Figure).
Conclusion: The CART analysis clearly identified optimal ranges in clinical parameters with U-shaped association of cardiovascular risks, and thus may help exploring optimal target ranges in the clinical settings.
- © 2013 by American Heart Association, Inc.