Project
Quality assessment of network models in clinical research and practice using predictive accuracy analysis

The person-specific network model and its adaptations are becoming increasingly popular in clinical research and practice for exploring the temporal relations among symptoms of individuals with mental health concerns. Such networks are developed to improve the understanding and treatment of mental disorders. Yet, they are often used without sufficient quality inspection, facing high risks of overfitting, which results in ungeneralizable and misleading estimates. Therefore, we propose a quality assessment method, “predictive accuracy analysis”, to evaluate a network’s generalizability. More specifically, we will complete three main tasks in four projects:

1. Explore the factors (e.g., sample size) that influence a network’s predictive accuracy (Project 1).
2. Evaluate the extent to which existing network adaptations (i.e., regularized networks, Project 2; fixed moderator networks, Project 3) avoid overfitting.
3. Develop an open-source, easy-to-use software for conducting predictive accuracy analysis on (adapted) networks and advocating for this analysis to become a standard procedure in future network studies (Project 4).

With our output, clinical researchers and clinicians can make sure that the networks they estimate will not overfit the sample and can be used to design personalized treatments.

Supervisors
dr. Laura F. Bringmann
dr. Anja F. Ernst
prof. dr. Peter de Jonge
prof. dr. Ginette Lafit

Financed by
PhD fund, Faculty of Behavioural and Social Sciences, University of Groningen

Period
1 September 2023 – 1 September 2028