Investigating the Feasibility of Idiographic Network Models

Recent times have seen a call for personalized psychotherapy and tailored communication during treatment, leading to the necessity to model the complex dynamics of mental disorders in a single subject. To this aim, time-series data in one patient can be collected through ecological momentary assessment and analyzed with the graphical vector autoregressive model, estimating temporal and contemporaneous idiographic networks. Idiographic networks graph interindividual processes that may be potentially used to tailor psychotherapy and provide personalized feedback to clients and are regarded as a promising tool for clinical practice. However, the question whether we can reliably estimate them in clinical settings remains unanswered. We conducted a large-scale simulation study in the context of psychopathology, testing the performance of personalized networks with different numbers of time points, percentages of missing data, and estimation methods. Results indicate that sensitivity is low with sample sizes feasible for clinical practice (75 and 100 time points). It seems possible to retrieve the global network structure but not to recover the full network. Estimating temporal networks appears particularly challenging; thus, with 75 and 100 observations, it is advisable to reduce the number of nodes to around six variables. With regard to missing data, full information maximum likelihood and the Kalman filter are effective in addressing random item-level missing data; consequently, planned missingness is a valid method to deal with missing data. We discuss possible methodological and clinical solutions to the challenges raised in this work. Translational Abstract Repeated measures of a subject-using smartphone apps-combined with sophisticated network-based modeling techniques have been proposed as promising methods in both psychological research as well as clinical practice. The promise is that a personalized network model can be obtained per person, giving unique insights in the dynamics of that person's life and psychological mechanisms. However, it is not currently known if such methods are at all tractable with data that can realistically be obtained. This may be especially problematic in clinical practice as, in that setting, data is likely scarce and contains many missing responses. We investigated the feasibility of these methods using large-scale simulation studies and found that while missing data can adequately be handled, more data may be required than previously assumed to reliably estimate personalized networks. We end the article with concrete suggestions to clinical practitioners as well as researchers aiming to study personalized network models.

Automated summary

Recent times have seen an increasing demand for personalized psychotherapy and tailored communication during treatment. This has led to the need to model the complex dynamics of mental disorders in individual patients. Time-series data can be collected through ecological momentary assessment and analyzed using the graphical vector autoregressive model to estimate personalized networks. These networks can be used to customize psychotherapy and provide personalized feedback to clients, making them a promising tool for clinical practice. However, it remains unclear whether these networks can be reliably estimated in clinical settings. A large-scale simulation study was conducted, and the results showed that sensitivity is low with sample sizes feasible for clinical practice. While the global network structure can be retrieved, the full network may not be recoverable. Estimating temporal networks is particularly challenging, and it is recommended to reduce the number of nodes to around six variables when using 75 and 100 observations. Full information maximum likelihood and the Kalman filter are effective in handling missing data, with planned missingness being a valid method. Methodological and clinical solutions to the challenges raised in this study are discussed.