A PREDICTION MODEL FOR LENGTH OF STAY IN FLORIDA HOSPITALS: BIG DATA ANALYSIS

Big data is an extremely useful tool for exploring relationships among observed variables. This paper investigates a big data and suggests a possible predictive model to interpret its outcomes. We examine a data set whose predictor variables are nominal in nature and the response variable is counted as several mutually exclusive categories of ordinal variables. We use a generalized linear model for the count data known as loglinear model. For this type of model, we need both the explanatory and response variables to be qualitative. The model outcome includes expected frequencies for each cell and the estimates of cell counts in an exponential form of average cell frequencies. Calculations of the odds of the cell responses and the odds ratios between cells are done by simple calculations using these estimates from the fitted model. We apply this technique to the inpatient hospital visits in the state of Florida in 2011, considering length of stay prior to discharge, gender, primary diagnosis, and severity of illness as significant variables. The results show some strong evidence of predicting prevalence of certain conditions over others. Gender and diagnosis play an important role in predicting the severity of the patients’ conditions and the length of stay in hospitals prior to discharge.