A systematic review of system dynamics and agent-based models: Mathematical modelling for health systems research

Abstract

Background: Mathematical modelling has emerged as a powerful approach for analysing complex healthcare systems, which exhibit nonlinear interactions, feedback loops, and adaptive behaviours. Among available techniques, System Dynamics Modelling (SDM) and Agent-Based Modelling (ABM) are widely used to capture macro- and micro-level dynamics, respectively. This study systematically reviews the application of these modelling approaches in health systems research. Methods: A systematic literature review was conducted following PRISMA guidelines. Multiple databases—including MEDLINE, EMBASE, Cochrane Library, MathSciNet, ACM Digital Library, HMIC, EconLit, and Global Health—were searched. A structured mathematical framework was used for study selection, screening, and data extraction. Results: A total of 52 studies were included in the final analysis after screening and eligibility assessment. These studies were classified based on modelling approach as follows: System Dynamics Modelling (SDM): 31 studies (59.6%) Agent-Based Modelling (ABM): 15 studies (28.8%) Hybrid Models (SDM–ABM): 6 studies (11.6%) Geographical Distribution: The majority of studies were conducted in high-income countries, with the following distribution: High-income countries: 34 studies (65.4%) Middle-income countries: 12 studies (23.1%) Low-income countries: 6 studies (11.5%) This highlights a persistent geographical imbalance, with limited modelling research in resource-constrained settings. Healthcare Settings: The included studies addressed a range of healthcare system contexts: Emergency and acute care: 16 studies (30.8%) Chronic disease management: 11 studies (21.2%) Public health policy & prevention: 10 studies (19.2%) Hospital resource allocation & patient flow: 9 studies (17.3%) Long-term and elderly care: 6 studies (11.5%) Primary Study Objectives: The main focus areas of the modelling studies were: Policy evaluation and intervention analysis: 18 studies (34.6%) Resource optimisation and capacity planning: 14 studies (26.9%) Disease transmission and epidemiological modelling: 11 studies (21.2%) Healthcare system performance analysis: 9 studies (17.3%) Validation Approaches: Validation methods varied across studies: Sensitivity analysis: 28 studies (53.8%) Empirical/real-world data validation: 19 studies (36.5%) Scenario-based validation: 15 studies (28.8%) Limited or no validation reported: 10 studies (19.2%) Key Findings: SDM models were predominantly used for policy-level and long-term system analysis. ABM models were more common in disease spread and behavioural simulations. Hybrid models, although fewer, demonstrated superior capability in capturing both macro and micro dynamics. A significant gap exists in applying these models to low-resource healthcare systems, particularly in developing countries. Conclusions: There is a critical need to expand modelling applications to low- and middle-income countries. Hybrid SDM–ABM frameworks show strong potential for capturing complex healthcare dynamics. Advances in computational modelling and data integration are expected to further enhance their applicability.