System Nader Amir and Shaan McGhie San Diego State University San Diego CA US Disclosure Dr Amir was formerly a part owner of Cognitive Retraining Technologies LLC ID: 602309
Download Presentation The PPT/PDF document "Post Traumatic Stress Disorder as a Caus..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
Post Traumatic Stress Disorder as a Causal System
Nader Amir and Shaan McGhieSan Diego State University, San Diego, CA US.
Disclosure : Dr. Amir was formerly a part owner of Cognitive Retraining Technologies, LLC (“CRT”), a company that marketed anxiety relief products. Dr. Amir’s ownership interest in CRT was extinguished on January 29, 2016, when CRT was acquired by another entity. Dr. Amir has an interest in royalty income generated by the marketing of anxiety relief products by this entity.
Funding : This project was supported by Grants 1R01MH101118, R01 MH106477 from NIMHSlide2
Traditional models of Post Traumatic Stress Disorder (PTSD)
Latent construct causes the symptoms that quantify it (DSM5; APA, 2013). That is, both categorical and dimensional models conceptualize PTSD symptoms as indicators of an underlying latent variable that is measured by these less than psychometrically perfect indicators The goal is therefore:Make the indicator better: more psychometrically soundSlide3
Traditional models of Post Traumatic Stress Disorder (PTSD)
This has met with limited success We have yet to discover any pathonomonic variables for the latent structuresMaybe we need to look harder, better (more biological) indicatorsMaybe the entire model can be complementedSlide4
Causal Systems Model of Post Traumatic Stress Disorder (PTSD)
Psychometricians have turned to a different approach to understand mental disorders such as PTSD Mental disorders as causal systems (Borsboom & Cramer, 2013) graphical models for causal relations that can complement conventional models
(Greenland, Pearl, & Robins, 1999) Also called influence diagramsrelevance diagrams causal networksSlide5
Causal System model of PTSDA
stressor (trauma) causes a symptom, which may cause other symptoms and in turn be affected by those symptomsThese symptoms themselves constitute the mental disorderSlide6
Causal System model of PTSD
McNally et al. (2014): Examined PTSD symptoms in 362 earthquake survivors (38% met criteria for probable PTSD)Feeling distant from other people was linked to loss of interest in previously enjoyable activities as well as emotional numbing Difficulty sleeping,
hypervigilance, and being easily startled were also clustered as interrelated symptoms.Slide7
McNally et al (2014)Slide8
Current study
Replicated and extended these network analysesUsed a weighted and directed networkThe magnitude of the relation is shown through thickness of the lineThe arrows start at the predictor symptom and end at the predicted
symptomTest the accuracy of the networkBayesian networkSlide9
The symptoms presented in the graphsDistant = feeling of detachment from
others Numb = restricted affectFuture = sense of foreshortened future
Sleep = difficulty falling or staying asleepStartle = exaggerated startled responseLossint = diminished interest in previously enjoyable activities,Avoidth = efforts to avoid thoughts that concern the traumaAvoidact= efforts to avoid activities, places, or people that arouse recollections of the trauma
.
Hyper
=
hyperarousal
in response to cues
Dreams
= traumatic dreams
Intrusion
=
intrusive thoughts, memories or images
Flash
= Flashbacks of the trauma
Upset
= feeling upset in response to reminders of the trauma
Concen
= difficulty concentrating
Physior
= Physiological reactions to reminders of the trauma
Anger
= feeling irritable or having outbursts Slide10
Network Analyses
NodeA symptom in the graphEdgelines connecting nodes, indicating a relationship between two nodes (thicker lines means stronger relationships/higher correlations)Association Networksimple correlations between symptomsConcentration NetworkPartial correlations between symptoms
Relative Importance network Directed and weighted network (shows direction and magnitude of relationships) of linear models.Uses lmg metric in R package Relaimpo (Grömping, 2006) Slide11
Measures of Centrality
Betweenessnumber of times a node is on the shortest path between two other nodesClosenessaverage distance of one node to the others (higher number means closer together)Strength-outThe effect of the node on other nodesSlide12
ResultsSlide13
665 students: 17 PTSD symptoms
Mean PCL-C score: 32.12, SD = 12.3815% (106) meet DSM-V criteria for PTSD
Fig 6. Association network of simple correlations between PTSD symptoms with a cutoff of minimum .30 correlation. Slide14
Fig 8. Concentration network of partial correlations between PTSD symptoms with a threshold of .10Slide15
Fig.
9 Relative Importance network of PTSD symptoms, showing direction and magnitude of relationshipsSlide16
Centrality plot for the relative importance network.Slide17
Detecting communities of symptoms
Possible to over-interpret the visualization of data (Fried, 2016, Psych Network)Most studies use the Fruchterman-Reingold algorithm to create a layout:Nodes with the most connections / highest number of connections in the center of the graphNode placement just one of many equally ‘correct’ waysSlide18
Detecting communities of symptoms
Better ways:Use Eigen valuesUse spinglass algorithm detects communitiesExploratory Graph Analysis (Golino & Epskamp (2016)
currently under developmentRe-estimates a regularized partial correlation network and uses the walktrap (a random walk) algorithm to find communitiesunlike eigenvalue decomposition it shows directly what items belong to what clustersSlide19
Fig. 11. Identify communities of items in networks using Exploratory Graph Analysis via the R-package
EGA and spinglassSlide20
But how sure are we?
Epskamp, Borsboom & Fried, 2016We canEstimate of the accuracy of edge-weights, by drawing bootstrapped CIInvestigate the stability of (the order of) centrality indices after observing only portions of the dataPerform bootstrapped difference tests between edge-weights and centrality indices to test whether these differ significantly from each otherSlide21
Epskamp, Borsboom & Fried, 2016
Bootstrapped confidence intervals of estimated edge-weights for the estimated network of 17 PTSD symptoms. The red line the sample valuesGrey area the bootstrapped CIs Each horizontal line represents one edge of the network, ordered from the edge with the highest edge-weight to the edge with the lowest edge-weight Slide22
DiscussionSlide23
What does this imply?
Epskamp et al found that: generally large bootstrapped CIs imply that interpreting the order of most edges in the network should be done with care and that upset when reminded of the trauma – upsetting thoughts/images being jumpy – being alertfeeling distant – loss of interest are reliably the three strongest edges since their bootstrapped CIs do not overlap with the bootstrapped CIs of any other edges
Current study:None did not overlap Highest Hypervigilane- startleintrusion--dreamsdistant--numbSlide24
Centrality stability
Stability of centrality indices by estimating network models based on subsets of the dataSlide25
Centrality stability
Betweenness 0.0505closeness 0.0505strength 0.361The CS-coefficient indicatesbetweenness (CS(cor = 0.7) = 0.05) andcloseness (CS(
cor = 0.7) = 0.05) are not stable under subsetting cases.Node strength performs better (CS(cor = 0.7) = 0.36)but does not reach the cutoff of 0.5 from simulation studiesThus: order of node strength is interpretable with some care, while the orders of betweenness and closeness are notSlide26
Testing for significant difference Edges cannot be shown to significantly differ from one-anotherSlide27
Bayesian network
McNally (in press)Parametric method that produces directed acyclical graphs Arrows with directionLacks cycles (feedback loops)An interdisciplinary area the aim of determining causal inferences from observational data. However, requires additional assumptions (Peal, 2014)
Possible? May be (smoking and cancer)Slide28
Each node is printed in square brackets along with all its parents (which are reported after a pipe as a colon-separated list)
[flash][upset|flash][intrusion|upset][physior|upset][
avoidth|upset][dreams|intrusion][avoidact|avoidth][amnesia|avoidth][lossint|avoidact][distant|lossint][numb|distant][concen|distant][
hyper|distant
]
[
future|distant:numb
]
[
sleep|concen
]
[
anger|concen
]
[
startle|hyper
]
Bayesian Networks, with causal aspirationsSlide29
Perturbed and restarted networkSlide30
Discussion
Replicated and extended the results of McNally et al. (2014) in a larger sampleThese results suggest that in a large sample, PTSD symptoms are interrelated especially bidirectionallyThese results suggest that the most central symptoms may be the most important in the disorder and thus, and ideal candidate to target in treatmentSlide31
Discussion
Interventions focusing on social support and interaction for PTSD are likely to influence one symptom cluster, thereby alleviating other symptoms that it affects. These data suggest a potential causal system of symptoms The fact that specific symptoms may give rise to each other highlights a pattern that may lead to their specific chronicitySlide32
Discussion
However, the results of our estimating Network accuracy suggested some cautionReplication and larger samplesBetter characterized samples may provide more clear and accurate networksBegin to create causal models that can be testedSlide33
Thank you