Abstract
Platelet supply chain (PLT SC) management has always been a challenging task for the healthcare system due to the nature of platelets (PLTs). PLTs have an extremely short shelf life after being extracted from a human body and the demand is highly uncertain, which may lead to a high percentage of wastage and shortage in the PLT SC. The purpose of this study is to investigate the opportunity of incorporating frozen PLTs (FPLTs) into the PLT SC to see how it can improve the performance of the PLT SC in which the PLTs are used only in liquid form with respect to the platelet shortage, wastage and substitution in transfusions.This thesis first proposes a two-stage stochastic programming (2SSP) model for the PLT SC incorporating FPLTs. The model is designed to assist the decision maker(s) in making decisions on the production, storage and distribution of the PLTs and FPLTs in a blood centre so that the costs of production, shortage, wastage and substitution of the PLTs and FPLTs can be reduced. In the 2SSP, the first stage decisions are tactical level decisions, which are to decide the stock target levels for PLTs and FPLTs for each day of the week in a given planning horizon. The second stage decisions are operational level decisions on daily PLT and FPLT production and inventory levels to satisfy the uncertain demand in different scenarios. In this stage, an assumption is made that the daily demand for blood follows Zero Inflated Poisson (ZIP) distribution with known parameters and the demand scenarios are generated using a Monte Carlo sampling approach. To derive a reliable number of scenarios to use in the 2SSP model, the in-sample and out-of-sample stability tests are performed. The performance of the 2SSP model incorporating FPLs is evaluated with systematic experiments.
Secondly, an extended goal programming model is built based on the proposed 2SSP model (2SSP-EGP) to investigate a more realistic situation when clear targets of blood shortage, wastage and substitution are set by the healthcare practitioners. These targets are, more often than not, conflicting and they cannot be expressed by constraints, which may lead to an infeasible solution. Furthermore, we collected six months of historical demand for blood data and estimated the parameters for ZIP using method of moments estimation (MME) and maximum likelihood estimation (MLE) approaches. The Chi-Square test has been used to check if the hypothesis that the demand data is modelled with the correct specification of the ZIP regression is valid. A Monte Carlo based scenario generation method is used to generate scenarios. The stability tests are also applied to derive a reliable number of scenarios to use in the 2SSP model. Then the performance of the 2SSP-EGP model incorporating FPLs is compared with its counterpart – a 2SSP-EGP model without incorporating FPLs to identify the improvements. The optimisation models, scenario generation process and stability tests are coded in Spyder Python, and the models are solved to optimality using the Gurobi Optimizer package. By comparing the PLT SC model with and without incorporating FPLTs, the PLT SC model incorporating FPLTs shows high reduction rates of PLTs wastage, shortage and substitution in transfusion and provides managerial insight for the practitioners.
In all the aforementioned models, the decision variables are taking fractional values. However, the production, storage and substitution of the PLTs or FPLTs take integer values in reality. Thus, in the final stage of the thesis, an improved hybrid algorithm is developed to search for high quality “good” solutions for the 2SSP-EGP models with integer decision variables, since the 2SSP-EGP models can only be solved to optimality with Gurobi if the decision variables are taking fractional values. No solution can be found for 2SSP-EGP models with integer decision variables after running Gurobi for 6 hours. The improved hybrid algorithm is a combination of a proposed Matheuristics algorithm and a modified ABC algorithm. It can be seen, from the experimental results, the improved hybrid algorithm shows a good convergence rate and performs well.
| Date of Award | 18 Aug 2023 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Xiang Song (Supervisor) & Dylan Jones (Supervisor) |