Challenges in characterizing psychopathologies as unhealthy dynamic systems

Transcription

1 Challenges in characterizing psychopathologies as unhealthy dynamic systems Conference on Complex Systems 2018 Thessaloniki Oisín Ryan 1 & Jonas Haslbeck 2 1 Department of Methodology and Statistics, Utrecht University 2 Psychological Methods, University of Amsterdam

2 Psychological Disorders as Dynamic Systems E D G B C F A

3 What s the appeal? E D G δ B C F A B C D E F G A

4 What s the appeal? D G E B Focus on causal interactions between variables/nodes Characterise structure of interactions that lead to pathology C F A

5 Why a complex dynamical system? Potentially promising mapping between complex systems concepts and psychological theories Bi-stable system Disorder vs no-disorder Hysteresis Disorder triggered by adverse life-events Figure: Wichers et al 2018

6 Theory: Symptom Networks Mental disorders arise from direct interactions betweeen symptoms Unhealthy state: symptoms activated Consistent with medical diagnosis People prone to disorder have different network structures Move more easily/frequently from healthy to unhealthy state Figure: Borsboom 2017

7 Computational Model for Symptom Networks Ising Model Simple proxy model of pairwise interactions S1 Positive Manifold Symptoms mutually activating (0,1) coding: symptoms not mutually disactivating S7 S2 Density marginal probability of unhealthy state S6 S3 Pathology as a function of network topology S5 S4 Cramer et al. (2016)

8 Computational Model for Symptom Networks Ising Model Simple proxy model of pairwise interactions Frequency Positive Manifold Symptoms mutually activating (0,1) coding: symptoms not mutually disactivating Density marginal probability of unhealthy state Pathology as a function of network topology Frequency Frequency

9 Empirical Network Models Dataset on psychological constructs related to some pathology Self-report questionnaire Cross-sectional / time-series Unhealthy vs healthy controls Fit a linear model to the data PMRF or VAR Use estimated parameters to construct a network Compute network metrics Node centrality and Density Look for individual/group differences

10 Empirical Network Models Dataset on psychological constructs related to some pathology Self-report questionnaire Cross-sectional / time-series Unhealthy vs healthy controls Fit a linear model to the data PMRF or VAR Use estimated parameters to construct a network Compute network metrics Node centrality and Density Look for individual/group differences P(Y j = 1 Y \j ) exp( ˆβY \j ) Y N(µ, Σ) ˆΣ 1 Y t = ˆΦY t 1 + e t

11 Empirical Network Models Dataset on psychological constructs related to some pathology Self-report questionnaire Cross-sectional / time-series Unhealthy vs healthy controls E Fit a linear model to the data PMRF or VAR D G B Use estimated parameters to construct a network Compute network metrics Node centrality and Density Look for individual/group differences C A F

12 What s the problem? Empirical applications have outpaced theoretical contributions Theoretical Empirical Nodes Symptoms? Dynamics Bi-stable?

13 Empirical Networks: Nodes Empirical symptom networks Depression (van Borkulo et al 2015) Schizophrenia (van Rooijen et al 2018) S1 Mix of symptoms and non-symptoms Self-efficacy (Santos et al 2018) Working memory (Hoorelbeke et al 2016) S7 S2 Pathology networks with no symptoms Personality traits (Fonseca Pedrero et al 2018) S6 S3 Emotions or mood states (Bringmann et al 2013; Pe et al 2015) S5 S4

14 Empirical Networks: Nodes Empirical symptom networks Depression (van Borkulo et al 2015) Schizophrenia (van Rooijen et al 2018) Mix of symptoms and non-symptoms Self-efficacy (Santos et al 2018) Adapt ER Working memory (Hoorelbeke et al 2016) Maladapt ER Resilience BRIEF_WM PASAT_ACC Pathology networks with no symptoms Personality traits (Fonseca Pedrero et al 2018) Resid Depress Emotions or mood states (Bringmann et al 2013; Pe et al 2015) Figure: Hoorelebeke et al (2016)

15 Empirical Networks: Nodes Empirical symptom networks Depression (van Borkulo et al 2015) Schizophrenia (van Rooijen et al 2018) C Mix of symptoms and non-symptoms Self-efficacy (Santos et al 2018) Working memory (Hoorelbeke et al 2016) R S E W Pathology networks with no symptoms Personality traits (Fonseca Pedrero et al 2018) Emotions or mood states (Bringmann et al 2013; Pe et al 2015) F Figure: Bringmann et al. (2013)

16 Empirical Networks: Dynamics Time-series data typically fitted using VAR(1) models Y t = ΦY t 1 + e t Stationarity assumed Uni-stable dynamics

17 Empirical Theoretical? Theoretical Empirical Nodes Symptoms Miscellaneous Dynamics Bi-stable Uni-stable

18 Open Problem 1: Mapping from node to disorder Meaningful characteristics of symptom networks not meaningful in other domains Symptoms: Density P(Symptom=On) Disorder present Emotion: Density?

19 Open Problem 1: Mapping from node to disorder S7 S1 S2? R C E S6 S3 S W S5 S4 F Symptoms Emotions (Misc) Defines Disorder

20 Open Problem 2: Bi-stable systems from Uni-stable models Theoretical Empirical

21 Open Problem 2: Bi-stable systems from Uni-stable models Healthy Controls Disorder Patients

22 Open Problem 2: Bi-stable systems from Uni-stable models Healthy to Disorder Transition Wichers et al. (2015)

23 Summary S1 C S7? S2 S6 R E S W S3 S5 S4 Symptoms Symptom L Bi-stable Theoretical F Emotions (Misc) Emotion (Misc) Uni-stable Empirical

24 Thanks for listening! ryanoisin.github.io