A model for troponin I as a quantitative predictor of in-hospital mortality

Waxman DA, Hecht S, Schappert J, Husk G. J Am Coll Cardiol. 2006 Nov 7;48(9):1755-62. Epub 2006 Oct 17. Division of Cardiology, Department of Emergency Medicine, Beth Israel Medical Center, New York

OBJECTIVES: We evaluated log transformed troponin I as a predictor of mortality in 2 independent populations.

BACKGROUND: The troponin I result is typically dichotomized by a single diagnostic cutoff. Its performance as a continuous prognostic variable has not previously been well-characterized.

METHODS: We studied the first troponin I sent from the emergency department (ED) as a predictor of all-cause inpatient mortality, with retrospectively gathered data. We performed our study in 2 stages, deriving our model with data from a single medical center and validating it with data from another. Subjects included every patient who had a troponin I sent from the ED during the period from November 2002 to January 2005. We assessed prognostic independence by including other potential confounders in nested logistic regression models. The troponin assay was identical at both sites (Ortho-Clinical Diagnostics, Rochester, New York).

RESULTS: There were a total of 34,227 patients (12,135 derivation and 22,092 validation). Odds ratio for mortality as a function of log10-troponin was 2.08 (95% confidence interval [CI] 1.85 to 2.32) in the derivation set and 2.07 (95% CI 1.92 to 2.24) for the validation set. Troponin I remained a strong predictor after inclusion of age, electrocardiogram normality, renal insufficiency, arrival mode, chief complaint, admission diagnosis, and abnormal vital signs into bivariate and nested multivariate models.

CONCLUSIONS: The presence of any detectable troponin I at ED presentation is associated with increased inpatient mortality. In 2 distinct clinical populations, the odds of death approximately doubled with any 10-fold increase in troponin result. This held true at levels below current diagnostic cutoffs. The placement and utility of dichotomous cutoffs might merit reconsideration.