Keywords: scatterplots, mortality, clinical, predictive, inpatient, ICD-10, ICD-9
Our case study uses administrative claims data from an all-payer data source. We compare ICD-9 to ICD-10 diagnosis and procedure codes used in predicting inpatient mortality. The results are presented in various ways to assess the impact of the coding systems on an identical data set along with graphical displays to visualize the variability and model performance. Using national mortality rates at the Clinical Classifications Software (CCS) category level we show differences that may impact model prediction. These differences also demonstrate possible instability due to the way patient records are coded.