Symposium: Complex Systems Analysis in Mental Health Research
Dr. Fred Hasselman
Prof. Denny Borsboom
Dr. Eiko Fried
Early warning signals (EWS) are used widely across fields such as ecology and virology to anticipate transitions like lake biodiversity changes and virus dissemination, and have recently shown promise as signals for mental health transitions. The statistical signals indicating an upcoming transition are often mathematically derived from dynamical system models, such as increases in variance as a marker of critical slowing down. As of yet, EWS have largely been applied to simple transitions such as the saddle-node bifurcation, yet it is widely conjectured that more complex transitions occur within systems as non-linear and high-dimensional as those found within psychopathology. To narrow this gap, we compare the performance of generic EWS in characterizing and anticipating more complex, higher-dimensional transitions between different dynamical regimes. In a numerical study of a four-dimensional Generalised Lotka-Volterra model under varying observational noise intensities, we focus on a noisy, periodic, and chaotic regime, which are traversed by two types of transitions: the birth of a limit cycle in a Hopf bifurcation and the creation of a chaotic attractor via a period-doubling cascade. Our simulation study approximates Ecological Momentary Assessment data collection, where data may be analysed in real-time without access to the full timeseries to detect a transition. In addition, to address the challenges arising in the move from theory to real-world psychological data, such as high dimensionality, non-linearity, noise, and non-stationarity, we include a relatively unexplored method in the EWS literature, namely Recurrence Quantification Analysis (RQA). RQA is a popular model-free nonlinear timeseries method which identifies recurrent patterns in line structures of the timeseries’ distance matrix. Our study emphasizes the limitations of EWS with respect to more complex transitions: Do these measures anticipate upcoming changes or merely characterize the regime change that has already occurred?
This is an in-person presentation on July 19, 2023 (11:20 ~ 11:40 UTC).
Ms. Kyra Evers
Early warning signals are indicators that some (major) change is about to happen. In almost all situations the indicators are obtained from (multiple) time series. Research into such indicators in climatology and engineering has met with some success. Such indicators of large changes in mental health would obviously be useful in psychopathology. However, it has been shown that simply applying standard techniques for indicators of large changes to any kind of system will often fail, i.e., it may fail to indicate a change when it is coming, but more often it will indicate change when it is not coming. To avoid such mistakes in early warning signals, we propose a framework using network theory to determine the type of indicators of large (qualitative) changes in psychopathology. In this framework we require assumptions about the type of network (relations between variables) and how the network is affected by external influences (big and small events in life). Applying this framework narrows down the type of indicators that are useful to function as early warning signals. We then focus, using the network framework, to determining extreme values. There are strong connections between the fields of extreme value theory and dynamical systems. The connections between these fields can be used to obtain early warning signals.
This is an in-person presentation on July 19, 2023 (11:40 ~ 12:00 UTC).
Panic disorder is a highly prevalent mental health condition that significantly impacts patients' quality of life. However, current treatments are not universally effective and have shown limited improvement since their introduction decades ago. This lack of progress is due in part to our incomplete understanding of the system underlying panic disorder and how different treatments intervene on it. Existing theories suggest that this system comprises multiple components that interact in non-linear ways over different time scales. Because of the counter-intuitive behavior of such systems, verbal theorizing can only provide limited understanding about them. To address this issue, we extended an existing computational model of panic disorder with a typical Cognitive Behavioral Therapy (CBT) treatment. Simulating treatment outcomes allows us to study how different treatment components interact with each other. Based on this analysis, we develop a new CBT treatment and demonstrate that its simulated outcomes are superior. Next, we introduce inter-individual variation in key parts of the model, and study which treatment work best for which type of patients, which leads us to personalized treatment plans. We close by discussing how computational models can advance treatment research and may lead to the development of better and more personalized treatments.
This is an in-person presentation on July 19, 2023 (12:00 ~ 12:20 UTC).
The measurement of physiological signals using wearable sensors (heart rate, electrodermal activity, skin temperature) as well as motor activity, has become a reliable, nonintrusive and affordable method for monitoring e.g. general arousal and physical activity. In a clinical context, such measurements may provide a useful tool in the care of patients who are unable to verbally report on their current emotional or psychological state. In the present study we analyse multivariate physiological time series that were simultaneously recorded in dyads of caregivers and youth with severe mental disability in residential care. We investigate whether synchronization between physiological signals can be used to predict incidents that occur during the day, which often concern some form of aggression towards objects, self, or others. So-called Inter-system Recurrence Networks (ISRN, an extension of Cross Recurrence Quantification Analysis) can be used to determine whether a coupling direction exists between the physiological signals of caregiver and client networks. This gives insight in leading, following or bi-directional interactions of physiological signals, within and between dyad members. Such synchronization measures may serve as early warning signals of incidents and could serve as indicators for interventions to prevent incidents (e.g. take a break, disengage, change of context). We compare our results to ISRN based on simulations of delay-coupled dynamic system models.
This is an in-person presentation on July 19, 2023 (12:20 ~ 12:40 UTC).