Reducing demand uncertainty in the platelet supply chain through artificial neural networks and ARIMA models

https://doi.org/10.1016/j.compbiomed.2019.103415Get rights and content

Abstract

One of the significant issues in global healthcare systems is improving the supply chain performance and addressing the uncertainties in demand. Blood products, especially platelets, have the most challenging supply chains in the health system given their short shelf life and limited human resources. Therefore, proper management of blood products is critical, and in turn, could reduce loss and health costs, and help preserve these valuable resources. This study aims to predict blood platelet demands based on artificial neural networks (ANNs) and auto-regressive integrated moving average (ARIMA) models in order to reduce the uncertainty in the supply chain. To this end, daily demands for eight types of blood platelets from 2013 to 2018 were used in the current study. Data were collected from treatment centers and hospitals located in Zahedan, Iran. The results of this study indicated that ANNs and ARIMA models were more accurate in predicting the uncertainties in demand than the baseline model used in Zahedan Blood Transfusion Center. The highest and lowest prediction improvements based on ANNs and ARIMA models were associated with type O+ and A+ platelets, respectively. Given that the ANN models can significantly improve the prediction of uncertainties in demand, we highly recommend that the conventional statistical prediction methods in blood transfusion centers be replaced with these models.

Introduction

One of the central issues in global healthcare systems is the improvement of supply chain performance. The health system has one of the most complex and challenging supply chains since it is directly connected to the health of human beings [1]. Issues such as uncertainty in demand, planning for inventory management and ordering, expiration dates, and limited human resources are among key challenges in the health sector, particularly the supply chain of blood and its products [[2], [3], [4], [5]]. Moreover, supply chain management and planning of blood products, especially blood platelets, are essential concerns for human life due to their high perishability [6]. Platelets are the most expensive blood products. As they have a shelf life of three days and high production costs, it is not economical to store them in large quantities in blood centers. Moreover, blood donation is often unpredictable, and demand for its products is random. The uncertainty in the supply chain of platelets has caused decision-makers and experts to face some challenges whenever there is a rise in the platelet demand (a shortage) or a fall in the number of referrals to blood collection centers (a surplus). Besides, issues such as restrictions might appear regarding platelet preservation as well as their excess production whenever there is a decline in demand or an increase in the number of referrals. For this reason, the production of platelets in blood centers must comply with the hospitals and medical centers’ demands [6]. Chopra and Meindl [7] argued that accurate prediction and knowledge of the demand could facilitate planning in the supply chain. Having accurate information on demand, especially when the shelf life of the product is short, can help make a better decision on required products, address supply shortages, and reduce resource and health costs [4,8]. Therefore, this issue is of great importance in the supply chain.

Previous works have studied uncertainty in demands and investigated its effects on the performance of supply chains by applying different methods [[9], [10], [11], [12], [13], [14], [15]]. Also, in order to avoid wasting blood and its products, different research works have been carried out to change the planning and management policies of hospitals and blood banks [[16], [17], [18], [19], [20]], among which only a few studies have used BOX Jenkins or ARIMA to predict blood demand [4,21,22] as discussed below.

In 2016, the monthly demand forecast of blood supply at New York City Blood Center was studied [4], and the optimal method for prediction was determined using MA, ES, ARMA and VARMA models. Results showed that the accuracy of the ARMA models and their simplicity compared to VARMA has turned them into best models in predicting blood demand. Filho et al. [21] sought to predict demand for distribution of blood components in a supply chain with the aim to improve planning and to create a balanced inventory process. To predict these blood components, they used the univariate multiplicative seasonal model of BOX-Jenkins and ARIMA methods. Authors of the prior study believed that using this model instead of traditional methods of moving averages (MA) with weekly lags, could improve the efficiency and accuracy of the planning. Filho et al. [22] in another study proposed a decision-making tool to predict demand for blood components in the supply chain. In their study, instead of adopting a weekly moving average method, a more complex parametric model based on BOX-Jenkins was proposed, and the results proved that the employed model was much better than the moving average. These studies indicate the supremacy of the ARIMA method compared to the linear statistical techniques (MA, ES, VARMA) and the univariate multiplicative seasonal model of BOX-Jenkins and ARMIA methods. However, in order to better diagnose data dynamics and underlying patterns in the observations (demand uncertainties), other methods of artificial neural networks (ANNs) could be used alongside these techniques [23].

Artificial neural networks, one of the components of computational intelligence, have recently received extensive attention from scholars in various fields of science. These networks are robust and competent tools for decision-making since they are intelligent, adaptable to environmental changes, generalizable to nonlinear complex systems, and able to process data at high speed [[24], [25], [26]].

In previous studies, a demand forecast of blood supply has been done through linear statistical techniques and BOX-Jenkins models. On the other hand, currently, the averaging method is being used to predict demands in blood transfusion organizations. Also, the above methods consider blood demand (data behavior) as consistent with relatively regular alterations, whereas in reality, nonlinear behaviors are observed as well. Thus, blood issues, especially the blood platelet demand, are a combination of both linear and nonlinear behaviors in which the nonlinear part has been ignored in earlier studies. Moreover, linear methods cannot address the complexities associated with changes in demand. Hence, in order to fill this gap, the present study is an attempt to use ANNs and ARIMA models to reduce the demand uncertainty and predict demand in the platelet supply chain of Zahedan Blood Transfusion Center. This prediction can reduce waste and production costs and alleviate the supply shortage.

The rest of this paper is organized as follows: Section 2 presents an overview of ARIMA and neural network methods. Section 3 investigates findings on existing methods where the best models appropriate for various blood types are chosen. Finally, conclusion and managerial implications are presented in Section 4 Discussion, 5 Conclusion and future directions.

Section snippets

Data collection

The data for this study were collected from the education department of Blood Transfusion Center in Sistan and Baluchistan province, Zahedan, Iran. This center is the foremost supplier of blood and its products for hospitals and medical centers in the province. To plan for producing and supplying the demanded platelets, the staff in this center uses an averaging method and the available data on demands from previous days to forecast requests for upcoming days. We have used the data on platelet

Results

Fig. 2 demonstrates a time series graph used in the ARIMA model to predict O+ type platelet demands. As the series for different blood type platelet demands are similar to one another, only the graph for this platelet is presented. The series includes platelet demand for 1826 days between 2013 and 2018.

For ARIMA modeling, the stationarity of the blood platelets series was first tested using Dickey-Fuller and Phillips Perron tests. The results are presented in Table 2.

H0

There is unit root for the

Discussion

Although the ARMA method is linear similar to the averaging method, it has a better performance. As the results indicate, the artificial neural network method is more sensitive to changes in demands and the actual trend in data. It proves that this method could improve the prediction significantly (as shown in Fig. 2 of the Appendix). In fact, one main reason that ANNs result in more accurate prediction compared to linear statistical methods is that ANNs can reproduce the dynamic interaction of

Conclusion and future directions

This study strives to minimize the production of waste and costs by predicting platelet demands more precisely and consequently reducing uncertainty in demands. Artificial neural networks and ARIMA models were used for this purpose. The results showed that the models have high accuracy in predicting the demands for this blood product. According to the neural network and ARIMA, the highest accuracy was associated with the type O+ platelet. These models could improve the predicted outcomes by 66%

Conflicts of interest

The authors declared no conflicts of interest.

References (41)

  • O.S. Filho et al.

    A decision- making tool for demand forecasting of blood components

    Bucharest

    (2012 MAY)
  • M. Qi et al.

    An investigation of model selection criteria for neural network time series forecasting

    Eur. J. Oper. Res.

    (2001 Aug)
  • F.F. Jahantigh et al.

    Development of a supply chain model for healthcare industry

  • M.Y. Attari et al.

    Developing multi-choice goal programming for reducing wastages and shortages of blood products at hospitals with considering cross matching ratio

    Int. J. Appl. Optim. Stud.

    (2018)
  • M. Fattahi et al.

    Integrated strategic and tactical supply chain planning with price-sensitive demands

    Ann. Oper. Res.

    (2016)
  • S. Chopra et al.

    Supply Chain Management: Strategy, Planning, and Operation

    (2013)
  • B. Malmir

    A new model for supply chain quality management of hospital medical equipment through game theory

  • A. Aghighi et al.

    Designing distribution networks of perishable products under stochastic demands and routs

  • A. Goli

    A multi-objective invasive weed optimization algorithm for robust aggregate production planning under uncertain seasonal demand

    Computing

    (2019)
  • A. Goli

    A robust just-in-time flow shop scheduling problem with outsourcing option on subcontractors

    Prod. Manuf. Res.

    (2019)
  • Cited by (70)

    View all citing articles on Scopus
    View full text