Risk scores did not reliably predict individual risk of mortality for patients with decompensated heart failure

We investigated the performance of four prognostic tools in predicting 180-day mortality for patients admitted for acute decompensated heart failure (ADHF) by calculating sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV) over a range of risk thresholds, in addition to discrimination and calibration.

We studied 1,458 patients. The risk assessment was performed using the Acute Decompensated Heart Failure National Registry (ADHERE) model and the Get With The Guidelines (GWTG), ADHF/NT-proBNP, and Acute Study of Clinical Effectiveness of Nesiritide in Decompensated Heart Failure (ASCEND) risk scores.

C-statistics ranged from 0.727 for the ADHERE model to 0.767 for the ADHF/NT-proBNP score. The ADHF/NT-proBNP risk score, the ADHERE model, and the ASCEND risk score, but not the GWTG risk score, were also well calibrated. Sensitivity and PPV were modest at the >30% risk threshold and ranged from 55% for the ADHF/NT-proBNP risk score to 38.8% for the ADHERE model and from 46.7% for the ADHF/NT-proBNP risk score to 42.1% for the ASCEND risk score, respectively. There was a modest agreement between the risk scores in classifying the patients across risk strata or in classifying those who died as being at >30% risk of death.

Although risk assessment tools work well for stratifying patients, their use in estimating the risk of mortality for individuals has limited clinical utility.