TY - JOUR T1 - Computational methods to model complex systems in sports injury research: agent-based modelling (ABM) and systems dynamics (SD) modelling JF - British Journal of Sports Medicine JO - Br J Sports Med DO - 10.1136/bjsports-2018-100098 SP - bjsports-2018-100098 AU - Adam Hulme AU - Scott Mclean AU - Paul M Salmon AU - Jason Thompson AU - Ben R Lane AU - Rasmus Oestergaard Nielsen Y1 - 2018/11/17 UR - http://bjsm.bmj.com/content/early/2018/11/17/bjsports-2018-100098.abstract N2 - ‘Systems thinking’,1 2 complexity theory3 and the ‘complex systems approach’4–7 are gaining momentum among leading sports injury researchers. One reason for this is a growing recognition that traditional risk factor identification methods (eg, stepwise regression modelling) fail to reflect the complex mechanisms of sports injury causation.1 4 8 9 Effective sports injury prevention requires us to understand the complex relationships that occur among a ‘web of interacting determinants’,4 rather than try to isolate the causal effect of individual factors.1 To better understand sports injury mechanisms, researchers are exploring several different approaches. One of them—and the focus of this editorial—is using computational methods that have the potential to describe and simulate the complex and dynamic nature of sports injury causation and prevention in ‘complex systems’.1 4 5 7The main characteristics of complex systems are described in table 1. These characteristics justify the use of two computational modelling approaches: agent-based modelling (ABM) and systems dynamics (SD) modelling. It is important to note that static modelling (eg, computer-based spreadsheets), differential equations, automata and process algebraic models, Bayesian networks, machine learning, neural networks, social network analysis and Monte Carlo methods are, unlike ABM and SD modelling, not able to model or simulate dynamic causal feedback among fundamentally different factors. Rather, predictive and simple statistical modelling, as well as mathematical algorithms that forecast the probability of future events, are useful for understanding certain aspects of complex systems at fixed time points and/or across one or more levels. These latter approaches are still computational in nature; however, they serve a purpose for a specific type of problem. We also alert the reader that the use of advanced statistical modelling to better understand the complex relationships between time-varying exposures (eg, training/match workloads) and sports injury development (eg, … ER -