Abstract 807: Incremental Value of Clinical Data Beyond Claims to Predict 30-Day Outcomes Following Heart Failure Admission: Results From the AHA’s Get With the Guidelines-Heart Failure Registry
BACKGROUND: The value of clinical data to improve the prognostic ability of existing claims-based models for mortality and readmission among hospitalized heart failure patients is unknown.
METHODS: We linked hospitalizations from the Get With the Guidelines-Heart Failure registry with 100% Medicare inpatient claims. For each outcome–30-day mortality from admission and 30-day readmission after discharge–we fit two logistic models. The first included only claims-based comorbidities. The second also included ejection fraction, hemoglobin, serum creatinine, serum sodium, heart rate, blood pressure, and weight. For each model, we calculated the c-statistic, the generalized R2 statistic, and the difference in observed outcome rates among the highest and lowest deciles of predicted risk. We also fit hierarchical logistic models and classified sites into performance deciles based on the predicted random site effects.
RESULTS: There were 33,390 patients hospitalized from 2003– 06 from 342 sites eligible for the mortality model, of which 31,223 were discharged alive and eligible for the readmission model. Observed 30-day mortality was 10.9% and readmission was 21.5%. Mortality model discrimination substantially improved with the addition of clinical data (Table⇓) yet readmission model discrimination was not meaningfully changed. In the hierarchical mortality model, 5 of 34 sites identified as top-performing by the claims model were not considered such by the claims-clinical model; and 4 of 34 bottom-performing sites changed designation. There was almost complete agreement on site rankings between the claims and claims-clinical readmission models.
CONCLUSIONS: Adding clinical data to claims data substantially improved prediction of mortality and shifted the mortality performance rankings for some sites. Clinical data did not meaningfully improve the claims-based readmission model, but neither model reliably predicted readmission.
This research has received full or partial funding support from the American Heart Association, National Center.