Real-time Application of an Inpatient Heart Failure Mortality Model to Predict 30-Day Mortality

Abstract

Background A CART model using admission systolic blood pressure (SBP), blood urea nitrogen (BUN), and serum creatinine (SCr) is proven to discern HF in-patient mortality risk. However, this model was not able to evaluate post-discharge outcomes. Objective We sought to 1) investigate an established HF in-patient mortality CART model's ability to discern 30-day all-cause mortality risk at an academic medical center and 2) investigate the potential to implement it in a real-time HF dashboard within the EMR Methods We have previously derived and validated a region-specific risk tool using CMS data on Medicare beneficiaries, age >65 years admitted to Vanderbilt University Medical Center with a primary diagnosis of acute HF from July 2009 to June 2016. We then applied the pre-specified threshold values from the aforementioned CART model for admission SBP, BUN, and SCr to evaluate 30-day all-cause mortality across the 5 resultant risk groups. We explored the presence of the model's data points at admission, the ability to automatically extract the data, and the ability to present a risk score in a real-time HF dashboard within the EMR. We utilized the CMS mortality report to identify death occurring within 30 days of hospital discharge. Results There were 2065 patients in the study population with 192 deaths occurring from July 2009-June 2016. The total population mean age was 77 ± 7.6 years with 54% male, 48% HFrEF, and 77% ischemic cardiomyopathy, not significantly different from the patients who died, 78 ± 7.6 years with 58% male, 57% HFrEF, and 77% ischemic cardiomyopathy. At admission, the median BUN was 28 (19,44) mg/dl, SCr was 1.4 (1.1, 2.0) mg/dl, and SBP was 122 (107,138) mmHg. The three data points were available within 24 hours of admission in all patients and the risk score was able to be presented in the EMR in real-time. In-patient mortality occurred in 2.5% of the population and the 30-day mortality rate was 9.3% at a median of 6 (0,15) days from discharge. The Highest Risk cohort in our CMS regional-specific model had increased 30-day mortality compared to the Low (OR 9.4; 95% CI 5.5-15.9), Intermediate Risk 3 (OR 6.5; 95% CI 3.7-11.4), and Intermediate Risk 2 (OR 3.5, 95% CI 2.0-6.2) cohorts but not the Intermediate Risk 1 (OR 1.60; 95% CI 0.9-2.8) cohort. Conclusion The high mortality risk stratification tool is capable of discerning 30-day mortality risk among hospitalized HF patients upon admission. This could be utilized to identify patients who may benefit from palliative care referral.