Demand forecasting is one of the important issues related to operations management in health sector. Forecasting patient volume in hospitals provides an important input regarding the correct planning of financial resources, human resources, and material resources. In this chapter, the authors first discuss forecasting patient volume in hospital services and then present a case study involving patient volume forecasting for a local hospital in Turkey. Different traditional statistical methods and machine learning methods are applied to both inpatient and outpatient demand from six polyclinics and a surgery room. Results show that damped trend exponential smoothing method outperforms other methods based on overall performance.
Top1. Introduction
The health sector plays an important role in the society, in the economy and in the continuity of human civilization. There is constant innovation and technological advancement in the sector. Management of health care operations is vital in this fast-paced, dynamic and competitive environment. Demand forecasting is an important aspect in operations management in health sector.
Demand in health sector can be examined from two perspectives, patient demand and material demand. Patient demand in a hospital can be divided into two categories: emergency services and regular services. Patient demand for emergency services usually exhibits higher demand uncertainty than the demand for regular services such as those provided in polyclinics, laboratories and radiology departments. Forecasting inventory items are also critical and consequences of errors can be vital in the health sector. While the demand for some inventory items is independent demand, some of them can be considered as dependent demand. For example, demand for some medical supply, medicine, blood, and serum may be affected by a particular type of surgery. In other words, patient demand for a certain surgery affects the demand for certain types of medical supplies.
One of the major issues in managing operations in healthcare services is the presence of capacity shortage especially in public hospitals in many countries (Barros et al., 2011). Capacity shortages emphasize the value of accurate demand forecast for efficient use of capacity. Today, analytical forecasting tools become a basic need in most industries. However, demand forecasting in health sector has not received as much attention as in other sectors (Jones et al., 2008). For example, Barros et al. (2011) states that 75% of the hospitals in Chile do not forecast demand at all.
Patient demand forecasting approaches consists of traditional statistical approaches and artificial intelligence approaches (Maddigan & Susnjak, 2022). On the statistical side, Cheng Wang and Li (2008) proposed the Expectation Method and Grade-Selection Method based on weighted-transitional matrix to forecast outpatient demand in a hospital. Hertzum (2017) applied ARIMA model to forecast hourly patient arrivals in an emergency room in Denmark. The main purpose was to manage the capacity in the emergency room. Yucesan, et al. (2020) aimed to forecast demand in an emergency room in Turkey. They used linear regression, ARIMA, exponential smoothing and ANN. They also used hybrid methods as well such as ARIMA-ANN and ARIMA-LR. The best performance is obtained by ARIMA-ANN. Klute et al. (2019) compared traditional statistical methods and machine learning methods in order to forecast outpatient appointments for two locations where one exhibited a cyclical and trend patterns and the other has cyclical pattern only. The results showed that traditional methods could outperform machine learning methods in their particular environment.