Improving the Accuracy of Multi-Dimensional Cost Allocation in Public Hospitals Using a GNN-RL Hybrid Strategy Discrete Optimization Algorithm
TL;DR
A hybrid strategy discrete optimization algorithm based on Graph Neural Networks (GNN) and Reinforcement Learning (RL) that mines the complex correlation features between cost elements through GNN and uses RL to realize the learning of optimal allocation strategies in dynamic environments.
Abstract
Cost allocation in public hospitals represents a core link in the optimal allocation of medical resources and the enhancement of operational efficiency. This systematic approach ensures that financial resource is reasonably allocated to form a sustainable framework that supports both clinical excellence and organizational stability. Traditional allocation methods have problems such as a single dimension, reliance on subjective assumptions, and poor dynamic adaptability, leading to insufficient allocation accuracy. This paper proposes a hybrid strategy discrete optimization algorithm based on Graph Neural Networks (GNN) and Reinforcement Learning (RL). It mines the complex correlation features between cost elements through GNN and uses RL to realize the learning of optimal allocation strategies in dynamic environments. Taking the operational data of 3 tertiary public hospitals from 2019 to 2023 as samples, a multidimensional cost allocation index system (including categories and 12 indicators: direct costs, indirect costs, and implicit costs) was constructed, and comparative experi ments were carried out. The results show that the allocation error rate of the proposed algorithm is 18.7% lower than that of the traditional Activity-Based Costing (ABC) and 11.3% lower than that of the Genetic Algorithm (GA). It performs better in scenarios such as cross-departmental cost allocation and adaptation to dynamic cost fluctuations. This study provides new technical support for the accurate cost accounting of public hospitals and the reform of medical insurance payment methods.
