Healthcare Demand Estimation Using Time Series Analysis in a Family Medicine Practice
Zeliha Yelda Özer, Erkan Tiyekli, Hatice Kurdak
Keywords: Family Practice, Time Series Analysis, Estimation, Forecasting, Health Care Demand, Optimization
Background:
There is a substantial increase in the number of patients and the need for services in health institutions. Therefore, healthcare leaders regard this increase among significant problems and need forecast data. Making the necessary amount of preparation according to the demand intensity is essential in increasing patient satisfaction and optimizing outpatient clinic services. Time series analysis is the series formed by ordering the observation values of any event according to time and is one of the models that can be used effectively in short and long-term demand forecasts.
Research questions:
This study aims to forecast healthcare demand for a Family Medicine Practice (FMP) with a time series analysis of at least four years of data.
Method:
Two years of data were drawn from the FMP database, established on 28 June 2018. Healthcare demand forecasts were produced as prescription, physical examination, sick-leave reports, health reports, immunization, periodic health exams for reproductive-age women, infant-child follow-ups, school-aged children follow-ups from FMP big data. The data were made stationary, and the autoregressive integrated moving average (ARIMA) was used for demand estimation. A preliminary 6-month healthcare demand estimation for the first half of 2022 was simulated with the model created with the last two-years data.
Results:
Partial-autocorrelation was used to determine the autoregressive (AR) terms of the ARIMA models. The autocorrelation functions were used to determine the moving average (MA) terms. By comparing MSE criteria, the most suitable model for the data set was selected among the models created for daily data (p values for Constant:0.007, AR{1}:0.01, AR{2}:0.002, MA{1}:0.0011, MA{2}:0.01). As a result, among these models, it was deemed appropriate to use the autoregressive integrated moving average model ARIMA (2,2,2) for daily data.
Conclusions:
Although the preliminary results seem satisfactory with two-year data, a more robust model can be obtained with a more extended time series.
Points for discussion:
Can this demand estimation be used for the optimization of the training schedule of the FMP?
Can an optimization model be developed for appropriate personnel shifts based on the demand frequency of family medicine practice?
Can this model be used with live script without creating a new model?