Time Series Prediction of Acute Care Hospital Charges

• In Event: Breakfast with Research Round Table Session

Abstract

The management of any business with customers is interested in knowing how much those customers will generate in charges in the future. Acute care hospitals are no different. Such quantities are used for budgeting and resource allocation. The usual budgeting process is performed by using educated estimates based on past experience and future expectations.
Time series analysis begins with the concept that the series of observations in the past reflects all of the influences that combine to make up the series. These influences include seasonal, trend, intervention and cyclical influences that can be extracted and quantified. Once this is done, the only remaining part of the series should be simple white noise. The quantified pieces of the puzzle can then be used to predict the next elements of the series.
The application of time series techniques to health care finance data has been minimal. The majority of work has gone into the prediction of market-based financial data. Individual patient charge data will be obtained from the New York SPARCS (State Planning and Review Cooperative Services) hospital discharge data for a six year period and compiled by facility and by month of discharge . This work will utilize the Box-Jenkins ARIMA (Auto-Regressive Integrated Moving Average) technique to model the time series of charges of six individual hospitals in New York State. These hospitals will be selected at random from strata based on decreasing number of beds, which is a proxy for size and complexity.
Each hospital will be modeled independently. Models will be backcast and forecast, and confidence limits calculated for the forecasts. Predicted values will then be compared with actual and accuracy levels determined. The resulting equations will be compared to determine whether there are similar patterns throughout all of the strata. If there are, it would simplify the process of model building for all New York Hospitals. If not, it would show that hospital time series are actually different from facility to facility and that model construction must be attempted one facility at a time.

Author

• Gerald S. Silberstein, Sage College of Albany