Abstract 15033: A Prediction Model for Short Length of Stay among Patients Admitted with Congestive Heart Failure
Introduction: Risk stratification for length of stay (LOS) among patients hospitalized with heart failure (HF) may be useful for identifying patients likely to have a short LOS for whom management in a short-stay observation unit would be possible.
Methods: We studied 3643 consecutive admissions discharged from a single urban academic medical center between October 2007 and August 2011 with a primary diagnosis of HF based on billing code. We developed a prediction tool for short LOS (defined as ≤ 48 hours) using generalized estimating equations, with three-quarters of the hospitalizations randomly selected as the derivation cohort and one-quarter as the validation cohort. Candidate predictors included sociodemographic and clinical characteristics, initial laboratory results, and vital signs.
Results: The probabilities of short LOS were similar in the derivation [31.5%] and validation [33.8%] cohorts. Admission day of the week, sodium, hematocrit, blood urea nitrogen, creatinine, HF admission within the previous 30 days, and suspicion of ischemia (defined as measurement of troponin on admission) were significantly associated with likelihood of short LOS. The base model incorporating these data had good discrimination (C-statistic of 0.65 in the derivation set and 0.67 in the validation set). The model was well calibrated with similar predicted / observed risks of short LOS across tertiles in the derivation (19% / 18%, 32% / 33%, 43% / 45%) and validation (19% / 20%, 33% / 30%, 46% / 52%) cohorts. Initial vital signs and brain natriuretic peptide (BNP) were available in 1362 and 1842 hospitalizations in the derivation cohort subjects, respectively. Heart rate, systolic blood pressure, respiratory rate, and BNP were associated with short LOS. In these subgroups, adding vital signs or BNP modestly improved the base model’s discrimination (C-statistics 0.68 and 0.67, respectively).
Conclusions: We developed and validated a model predicting short LOS among patients hospitalized with HF using routinely collected biochemical and physiologic data. Given the high cost of inpatient HF care, use of such a model to identify patients who can be managed either as an outpatient or in a short-stay observation unit may present a cost-saving strategy.
- © 2012 by American Heart Association, Inc.