Abstract 230: Improved Prediction of 30-day Mortality after Heart Failure Hospitalization by Utilization of Admission Vital Signs and Laboratory Findings.
Background: Heart failure (HF) is associated with high mortality so early identification of those at high risk may be useful in reducing mortality rates. We sought to develop a simple risk score from data readily available at the time of HF admission. Methods: We studied Veterans hospitalized from January 2004 to December 2009, who were discharged alive with a primary diagnosis of HF (based on ICD-9 coding). For the purposes of these analyses, only the 869 who had complete data for admission vital signs, laboratory parameters, ECG, co-morbidities and ejection fraction were included. Univariate analyses were employed to identify variables associated with mortality at a p-value < 0.10. These variables were then analyzed by MVA to determine independent predictors of mortality. The c-statistic was utilized to determine cutpoints for continuous variables. A risk score was developed by assigning risk points based on the odds ratio for each variable. Model calibration was assessed by the Hosmer-Lemeshow (H-L) statistic and by plotting observed vs. expected mortality (Figure below). Results: Of the 869, 35 (4.0%) died within 30-days of discharge. Twelve independent predictors of 30-day mortality were identified: admission pulse, systolic and diastolic BP, BNP, troponin-I, sodium, glucose, ALT, atrial fibrillation and absence of dyslipdemia diagnosis, and not admitted on an oral anticoagulant or a calcium channel blocker. All vartiables were assigned 1 risk point, except for not on an oral anticoagualnt, which was asigned two points. Patients with a risk score < 6 (57% of the group) had a mortality rate of only 0.4%, while a risk score > 6 (20% of the group) was associated with a 15.3% mortality, corresponding to a 49-fold range. Conclusion: 30-day mortality can be predicted by the developed risk score which has excellent discrimination (c-statistic = 0.87), adequate calibration (H-L = 0.11) and covers a mortality range from 0% to nearly 50%. This risk score also has the advantage of being easily calculated within 24 hours of admission. Validation of the model will be an important next step.