Abstract 118: Factors Associated With Hospital Length of Stay for Heart Failure

Abstract

Objective: To identify which hospital indicators are related to hospital length of stay (LOS) for heart failure (HF) inpatients, and their associations. Methods: Data from Centers for Medicare & Medicaid Services from 2012 to 2015 and the Florida Agency for Healthcare Administration (AHCA) from April 2014 to June 2016 were analyzed. We explored numerous quality and hospital size indicators, which were collected for nonidentical patient populations, for their potential associations with hospital mean LOS of Florida HF inpatients. We applied machine learning methods to screen for metrics most important to predicting LOS. Next, we derived generalized linear models (GLM) using several key screened indicators and multiple imputations for missing data, while weighting hospitals’ metrics according to their numbers of HF inpatients. We considered a total of 175 general acute care hospitals in Florida, and a total of 178,987 inpatient visits to obtain the values of mean hospital LOS. Results: Relationships obtained from the GLM for factors with substantial associations with hospital HF LOS are: Larger values of hospital-wide total staff size, emergency department wait time, and MRSA rates were all significantly associated with larger hospital HF LOS values. Increases in total hospital staff size from 550 to 5924 (the lowest and highest deciles of the hospital staff size distribution), given a fixed value of the MRSA rate, were associated with an average increase in the hospital HF LOS from its mean value over all hospitals of 5.0 to 5.9 (± .1) days. Increases in the mean time spent in the ED for all admitted patients from 222 to 374 minutes (the lowest and highest deciles of the distribution among hospitals for this ED wait time) were associated with an average increase in the hospital HF LOS from its mean of 5.0 to 6.0 (± .1) days. The GLM coefficient estimates (p < .0001) show these associations with the average over Florida hospitals of HF LOS. Additional factors, such as the number of HF inpatients, were also positively associated (though less strongly) with HF LOS, and are highly correlated with the indicators above such as hospital staff size. Conclusion: For Florida inpatients, the mean hospital HF LOS was positively associated with hospital-wide metrics of total staff size, MRSA rate, and time spent in the ED. Since the data were observational and many factors were not controlled, potential causality can not be determined. Further study is needed to discover the mechanisms behind these associations, as well as any potential confounding factors. Disclaimers: Results here were derived from limited data sets of the Florida AHCA. However, Florida AHCA does not claim responsibility for any of the resulting analysis, interpretations or conclusions. The authors and their institutions do not claim that the associations found are causal, nor that any actions taken based on the results here will change outcome metrics.