Network neuropsychology: principles, pitfalls, and promises

In ‘network neuropsychology’, graph theory is applied to the neuropsychological task performance of patients with neurological disorders to investigate cognitive functioning and/or neuropsychological measurement. This approach has been used to study cognition in Alzheimer’s disease, vascular encephalopathy, temporal lobe epilepsy, paediatric epilepsy, and stroke, as well as the measurement properties of screening instruments. Several authors have commented on the strengths and limitations of applying graph theory in neuropsychology; however, a more comprehensive appraisal has not been performed. This review briefly summarises the available literature. It highlights some putative principles; specifically, loose assumptions that cognition can be thought of as a network and that cognitive reorganisation can occur in neurological disorders. Potential benefits of network neuropsychology (e.g., holism, consistency with clinical practice, supplementing the latent variable approach to cognition, and hypothesis generation) are then outlined. Additionally, several potential pitfalls with network neuropsychology are discussed. These include validity (e.g., variable inclusion and content validity), statistical (e.g., statistical power, differential variability, etc.), and theoretical (e.g., parallelism, the need for theoretical as well as statistical models, the nature of cognitive functions probed by tasks) issues. Some secondary analyses of open data are presented to support these discussions. It is suggested that the lack of parallelism between neuropsychological task performance and cognitive functioning and the paucity of cognitive theory in network neuropsychology are the most prominent limitations. It is noted that many pitfalls also apply to other approaches in neuropsychology, whereas the promises of network neuropsychology are fairly unique. It is concluded that, in addition to statistical network models, cognitive theory - whether it is network theory or not - is needed to realise the potential benefits of network neuropsychology.