Unveiling Insights in Healthcare Operations: Analysing Dynamic Inventory Management and Prescription Patterns with Hidden Markov Models

Abstract

This thesis explores three key aspects of healthcare operations management: hospital inventory management, oncology drug prescription patterns, and physician prescription behaviour. The first study addresses hospital inventory management using the Hidden Markov Model (HMM) and Markov Decision Processes (MDP). The main contribution of this study is the method used to determine the optimal number of hidden demand states in the HMM. This approach explores the interaction between the HMM and the MDP during the development of the inventory management model. We illustrate the performance of our approach through a case study, showing that additional hidden states increase inventory costs beyond an optimal threshold. The optimal number of hidden states depends on demand patterns, inventory shortage, and wastage costs. Therefore, it suggests that inventory clusters should be managed with models designed for various demand patterns and service levels. The second study investigates the prescription patterns of oncology Named Patient Drugs (NPDs), a complex issue due to uncertainties. The study introduces an approach using convolutional HMMs. The model enables comparison and analysis of cancer diagnoses and treatment patterns over time, uncovering prescription patterns by calculating state transition and joint probabilities. This provides information on the evolution of NPD selection strategies, improving decision-making in inventory management, informing prescription guidelines, and improving patient care and resource allocation, especially under budget constraints. The third study uses a multivariate HMM to analyse the complexities of doctors’ prescription behaviours when treating breast cancer patients. This differs from the second study as it focuses on how these behaviours have evolved and the factors influencing the sequential nature of these behaviours. The study analyses data from a Sri Lankan hospital, revealing sequential patterns in prescribed treatments and how oncologists’ prescription decisions vary based on the unit price and dosage of drugs and the age of patients. The model uncovers heterogeneity in prescription behaviour, highlighting the need for personalised treatment strategies that consider unique prescribing behaviours. Overall, the three articles improve hospital inventory management, explain the hospital’s drug prescription patterns, and uncover patterns in the physicians’ prescription behaviour.