Publicação

EXPLAINABLE MACHINE LEARNING FOR HEALTHCARE COST OPTIMIZATION: A TIME-DRIVEN APPROACH

Ver documento

Detalhes bibliográficos
Resumo:Efficient cost and resource allocation in healthcare is essential for the sustainability of hospital operations and patient-centered initiatives. However, this can be a complex issue due to the extensive scope of work and the difficulty in maintaining economic models designed for these evaluations. In recent years, machine learning (ML) has been increasingly adopted to support healthcare decision-making, but despite its predictive power, a major limitation remains: the lack of interpretability in many models, which hinders trust and usability by medical personnel. In this study, we use the electronic medical records of 2800 cardiothoracic surgery patients of Santa Marta’s Hospital and propose a novel approach that integrates ML with mathematical optimization to provide interpretable insights for healthcare cost analysis. Our methodology incorporates a set covering model within a clustering algorithm to identify representative patient cohorts, addressing the explainability gap in current ML approaches. Subsequently, we apply the Time-Driven Activity-Based Costing model to estimate the cost of each patient type by mapping clinical activities to time-based resource consumption. The results show that patients can be grouped into meaningful cohorts that share clinical and resource-use characteristics, while still showing differences in costs. This allows for comparison of care pathways across patient types and provides cost estimates that capture both similarities and variations within the population. To further demonstrate the adaptability of our framework, the clustering model is applied to another case study regarding Blockchain technology, highlighting its ability to extract interpretable patterns and optimize decision-making in domains beyond healthcare. By combining optimization techniques with interpretable machine learning, our approach provides a transparent framework for complex cost and resource analysis. This supports more informed decision-making, aligns with the growing demand for explainable ML, and enhances communication between technical and domain-specific stakeholders.
Autores principais:Loureiro, Dulce da Silva
Assunto:Machine Learning Optimization Data-Driven Explainability Cost Analysis Time-Driven Activity-Based Costing
Ano:2025
País:Portugal
Tipo de documento:dissertação de mestrado
Tipo de acesso:acesso aberto
Instituição associada:Universidade Nova de Lisboa
Idioma:inglês
Origem:Repositório Institucional da UNL
Descrição
Resumo:Efficient cost and resource allocation in healthcare is essential for the sustainability of hospital operations and patient-centered initiatives. However, this can be a complex issue due to the extensive scope of work and the difficulty in maintaining economic models designed for these evaluations. In recent years, machine learning (ML) has been increasingly adopted to support healthcare decision-making, but despite its predictive power, a major limitation remains: the lack of interpretability in many models, which hinders trust and usability by medical personnel. In this study, we use the electronic medical records of 2800 cardiothoracic surgery patients of Santa Marta’s Hospital and propose a novel approach that integrates ML with mathematical optimization to provide interpretable insights for healthcare cost analysis. Our methodology incorporates a set covering model within a clustering algorithm to identify representative patient cohorts, addressing the explainability gap in current ML approaches. Subsequently, we apply the Time-Driven Activity-Based Costing model to estimate the cost of each patient type by mapping clinical activities to time-based resource consumption. The results show that patients can be grouped into meaningful cohorts that share clinical and resource-use characteristics, while still showing differences in costs. This allows for comparison of care pathways across patient types and provides cost estimates that capture both similarities and variations within the population. To further demonstrate the adaptability of our framework, the clustering model is applied to another case study regarding Blockchain technology, highlighting its ability to extract interpretable patterns and optimize decision-making in domains beyond healthcare. By combining optimization techniques with interpretable machine learning, our approach provides a transparent framework for complex cost and resource analysis. This supports more informed decision-making, aligns with the growing demand for explainable ML, and enhances communication between technical and domain-specific stakeholders.