High-sensitivity troponin as an outcome predictor in acute medical admissions.

BACKGROUND

Troponin estimation is increasingly performed on emergency medical admissions. We report on a high-sensitivity troponin (hscTn) assay, introduced in January 2011, and its relevance to in-hospital mortality in such patients.

AIM

To evaluate the impact of hscTn results on in-hospital mortality and the value of incorporating troponin into a predictive score of in-hospital mortality.

METHODS

All patients admitted as general medical emergencies between January 2011 and October 2012 were studied. Patients admitted under other admitting services including cardiology were excluded. We examined outcomes using generalised estimating equations, an extension of generalised linear models that permitted adjustment for correlated observations (readmissions). Margins statistics used adjusted predictions to test for interactions of key predictors while controlling for other variables using computations of the average marginal effect.

RESULTS

A total of 11 132 admission episodes were recorded. The in-hospital mortality for patients with predefined cut-offs was 1.9% when no troponin assay was requested, 5.1% when the troponin result was below the 25 ng/L 'normal' cut-off, 9.7% for a troponin result ≥25 and <50 ng/L, 14.5% for a troponin result ≥50 and <100 ng/L, 34.4% for a troponin result ≥100 and <1000 ng/L, and 58.3% for a troponin result >1000 ng/L. The OR for an in-hospital death for troponin-positive patients was 2.02 (95% CI 1.84 to 2.21); when adjusted for other mortality predictors including illness severity, the OR remained significant at 2.83 (95% CI 2.20 to 3.64). The incorporation of troponin into a multivariate logistic predictive algorithm resulted in an area under the receiver operating characteristic curve to predict an in-hospital death of 0.87 (95% CI 0.85 to 0.88).

CONCLUSIONS

An increase in troponin carries prognostic information in acutely ill medical patients; the extent of the risk conferred justifies incorporation of this information into predictive algorithms for hospital mortality.