1. Replicability and generalizability of PTSD networks: a cross-cultural multisite study of PTSD symptoms in four samples of trauma patients
Eiko Fried
Department of Psychological Research Methods
University of Amsterdam
VGCT November 2017

2. Marloes Eidhof Sabina Palic Giulio Costantini Mathilde Huisman
Cherie Armour
Claudi Bockting
Iris Engelhard
Karen-Inge Karstoft
VGCT November 2017

3. What didn’t look better was …
Back then introduced r package bootnet to look into stability and accuracy of network models.

4. 2016: Replicability of networks
No single published paper investigated replicability in 7+ years of network literature.
This holds not only for PTSD, but for network modeling in general (e.g. depression). By the way, this also holds for factor modeling where people usually fit a model to one dataset.

5. Replicability of PTSD networks
>10 papers published , but networks fitted to only 1 dataset each
Replicability and generalizability unknown
Networks fitted to primarily small datasets
Potential power issues
Networks fitted on community or subclinical data
Network structure of clinical data unknown (arguably the one we are interested in)
BUT WHERE TO FIND DATA?

9. ^

12. Replicability of PTSD networks
>10 papers published , but networks fitted to only 1 dataset each -> 4 datasets
Replicability and generalizability unknown
Networks fitted to primarily small datasets
Potential power issues
Networks fitted on community or subclinical data
Network structure of clinical data unknown (arguably the one we are interested in)

13. Replicability of PTSD networks
>10 papers published , but networks fitted to only 1 dataset each -> 4 datasets
Replicability and generalizability unknown
Networks fitted to primarily small datasets -> total N=2,782
Potential power issues
Networks fitted on community or subclinical data
Network structure of clinical data unknown (arguably the one we are interested in)

14. Replicability of PTSD networks
>10 papers published , but networks fitted to only 1 dataset each -> 4 datasets
Replicability and generalizability unknown
Networks fitted to primarily small datasets -> total N=2,782
Potential power issues
Networks fitted on community or subclinical data
Network structure of clinical data unknown (arguably the one we are interested in) -> Traumatized patients receiving treatment

15. Cross-cultural multisite study

16. Samples
Very diverse samples: different trauma-types, including civilian-, refugee-, combat-, post-war off-spring-, and professional duty-related trauma.
E.g. sample 3: predominantly male white Danish veterans.
E.g. sample 4: severely traumatized refugees of which 30% experienced persistent trauma-related psychotic symptoms.

17. Methodology
3 different questionnaires (HTQ, PCL-C, PSS-SR); all brought to 1-4 scale on 16 DSM-IV items
Estimate polychoric correlation matrix of symptoms
Estimate four network models jointly using the Fused Graphical Lasso (FGL). This is a novel version of estimating the Gaussian Graphical Model.

18. Descriptives

19. Spearman correlations between the symptom profiles ranged from 0
Spearman correlations between the symptom profiles ranged from 0.43 (datasets 2 and 3) to 0.82 (datasets 1 and 2), with a mean correlation of 0.60.

20. Correlations: 0.43 to 0.82, mean = 0.60
Spearman correlations between the symptom profiles ranged from 0.43 (datasets 2 and 3) to 0.82 (datasets 1 and 2), with a mean correlation of 0.60.

21. Networks

22. Predictability 35% (dataset 2) to 43% (dataset 1).
The four networks featured many consistent edges such as the strong connection between Nightm (2) — Sleep (12) and the moderate connection between Detach (9) — EmoNumb (10); in all networks, Amnes (7) was weakly inter-connected. There were also specific edges that differed considerably across networks, such as Intr (1) — React (4) which was very weak in network 4, moderately strong in networks 1 and 3, and strong in network 2; or Startl (16) — Hyperv (15) which was nearly absent in network 4, moderately strong in network 1, and strong in networks 2 and 3. The only moderately strong negative edge that emerged was in network 3 between Irrit (13) — AvThought (5), which is not too implausible: people that are less likely to avoid thoughts about the trauma may be more irritable.

27. Comparing the 4 networks
Quantifying differences: The Network Comparison Test
Omnibus: networks are not exactly identical (all p<0.005)
Post-hoc: between 2 to 8 of the 120 edges (1.7%-6.7%) differed significantly across the networks (mean difference: 3.1 edges; 2.6%)
97.4% of the edges were identical
Quantifying similarity: Correlations of edge weights ranged from 0.62 to 0.74 (mean = 0.73), indicating strong similarities

28. Correlations: 0.63 to 0.75, mean = 0.72
Strength centrality is shown in Figure 3; the centrality order was substantially related across the four networks, with correlations ranging from 0.63 (networks 2 and 3) to 0.75 (networks 2 and 4). Amnes (7), EmoNumb (10), and Irrit (13) had consistently low centrality estimates (all standardized centrality estimates considerably below 0), whereas Intr (1), Detach (9), and React (4) emerged as consistently central symptoms.

29. The strongest edges emerged between Intr (1) — Flash (3), AvoThought (5) — AvoSit (6), Nightm (2) — Sleep (12), and Detach (9) — EmoNumb (10), with edges weights of 0.32, 0.32, 0.31 and 0.26 respectively. The most central symptoms were React (4), Detach (9), Intr (1), and Disint (8) with standardized strength estimates of 1.27, 1.06, 0.96, and 0.56; Amnes (7) with a value of was by far the least central symptom.
In the cross-sample variability network, the most variable edges across the four networks were Intr (1) — Flash (3), Hyperv (15) — Startl (16), and Intr (1) — React (4), with standard deviations of 0.15, 0.15, and 0.14 respectively. For the remaining edges, standard deviations were small to negligible.

30. The strongest edges emerged between Intr (1) — Flash (3), AvoThought (5) — AvoSit (6), Nightm (2) — Sleep (12), and Detach (9) — EmoNumb (10), with edges weights of 0.32, 0.32, 0.31 and 0.26 respectively. The most central symptoms were React (4), Detach (9), Intr (1), and Disint (8) with standardized strength estimates of 1.27, 1.06, 0.96, and 0.56; Amnes (7) with a value of was by far the least central symptom.
In the cross-sample variability network, the most variable edges across the four networks were Intr (1) — Flash (3), Hyperv (15) — Startl (16), and Intr (1) — React (4), with standard deviations of 0.15, 0.15, and 0.14 respectively. For the remaining edges, standard deviations were small to negligible.

31. The strongest edges emerged between Intr (1) — Flash (3), AvoThought (5) — AvoSit (6), Nightm (2) — Sleep (12), and Detach (9) — EmoNumb (10), with edges weights of 0.32, 0.32, 0.31 and 0.26 respectively. The most central symptoms were React (4), Detach (9), Intr (1), and Disint (8) with standardized strength estimates of 1.27, 1.06, 0.96, and 0.56; Amnes (7) with a value of was by far the least central symptom.
In the cross-sample variability network, the most variable edges across the four networks were Intr (1) — Flash (3), Hyperv (15) — Startl (16), and Intr (1) — React (4), with standard deviations of 0.15, 0.15, and 0.14 respectively. For the remaining edges, standard deviations were small to negligible.

32. The strongest edges emerged between Intr (1) — Flash (3), AvoThought (5) — AvoSit (6), Nightm (2) — Sleep (12), and Detach (9) — EmoNumb (10), with edges weights of 0.32, 0.32, 0.31 and 0.26 respectively. The most central symptoms were React (4), Detach (9), Intr (1), and Disint (8) with standardized strength estimates of 1.27, 1.06, 0.96, and 0.56; Amnes (7) with a value of was by far the least central symptom.
In the cross-sample variability network, the most variable edges across the four networks were Intr (1) — Flash (3), Hyperv (15) — Startl (16), and Intr (1) — React (4), with standard deviations of 0.15, 0.15, and 0.14 respectively. For the remaining edges, standard deviations were small to negligible.

33. The strongest edges emerged between Intr (1) — Flash (3), AvoThought (5) — AvoSit (6), Nightm (2) — Sleep (12), and Detach (9) — EmoNumb (10), with edges weights of 0.32, 0.32, 0.31 and 0.26 respectively. The most central symptoms were React (4), Detach (9), Intr (1), and Disint (8) with standardized strength estimates of 1.27, 1.06, 0.96, and 0.56; Amnes (7) with a value of was by far the least central symptom.
In the cross-sample variability network, the most variable edges across the four networks were Intr (1) — Flash (3), Hyperv (15) — Startl (16), and Intr (1) — React (4), with standard deviations of 0.15, 0.15, and 0.14 respectively. For the remaining edges, standard deviations were small to negligible.

34. The strongest edges emerged between Intr (1) — Flash (3), AvoThought (5) — AvoSit (6), Nightm (2) — Sleep (12), and Detach (9) — EmoNumb (10), with edges weights of 0.32, 0.32, 0.31 and 0.26 respectively. The most central symptoms were React (4), Detach (9), Intr (1), and Disint (8) with standardized strength estimates of 1.27, 1.06, 0.96, and 0.56; Amnes (7) with a value of was by far the least central symptom.
In the cross-sample variability network, the most variable edges across the four networks were Intr (1) — Flash (3), Hyperv (15) — Startl (16), and Intr (1) — React (4), with standard deviations of 0.15, 0.15, and 0.14 respectively. For the remaining edges, standard deviations were small to negligible.

35. The strongest edges emerged between Intr (1) — Flash (3), AvoThought (5) — AvoSit (6), Nightm (2) — Sleep (12), and Detach (9) — EmoNumb (10), with edges weights of 0.32, 0.32, 0.31 and 0.26 respectively. The most central symptoms were React (4), Detach (9), Intr (1), and Disint (8) with standardized strength estimates of 1.27, 1.06, 0.96, and 0.56; Amnes (7) with a value of was by far the least central symptom.
In the cross-sample variability network, the most variable edges across the four networks were Intr (1) — Flash (3), Hyperv (15) — Startl (16), and Intr (1) — React (4), with standard deviations of 0.15, 0.15, and 0.14 respectively. For the remaining edges, standard deviations were small to negligible.

36. The strongest edges emerged between Intr (1) — Flash (3), AvoThought (5) — AvoSit (6), Nightm (2) — Sleep (12), and Detach (9) — EmoNumb (10), with edges weights of 0.32, 0.32, 0.31 and 0.26 respectively. The most central symptoms were React (4), Detach (9), Intr (1), and Disint (8) with standardized strength estimates of 1.27, 1.06, 0.96, and 0.56; Amnes (7) with a value of was by far the least central symptom.
In the cross-sample variability network, the most variable edges across the four networks were Intr (1) — Flash (3), Hyperv (15) — Startl (16), and Intr (1) — React (4), with standard deviations of 0.15, 0.15, and 0.14 respectively. For the remaining edges, standard deviations were small to negligible.

37. Comparison to McNally et al. 2015

38. Stability
Stability analyses indicated that all four networks were accurately estimated, with small to moderate confidence intervals around the edge weights. The CS-coefficient for strength centrality for the four networks was 0.60, 0.59, 0.75 and 0.52 respectively (the recommended threshold for stable estimation is >0.5 [25]). Details are available in the Supplementary Materials.

39. Open science
While we cannot share the data, we have made all correlation matrices and many other objects (means, standard deviations, network parameters) freely available on the Open Science Framework
The paper is fully reproducible.
You can use the data for factor and network models!
Stability analyses indicated that all four networks were accurately estimated, with small to moderate confidence intervals around the edge weights. The CS-coefficient for strength centrality for the four networks was 0.60, 0.59, 0.75 and 0.52 respectively (the recommended threshold for stable estimation is >0.5 [25]). Details are available in the Supplementary Materials.

40. Accepted at Clinical Psychological Science
Stability analyses indicated that all four networks were accurately estimated, with small to moderate confidence intervals around the edge weights. The CS-coefficient for strength centrality for the four networks was 0.60, 0.59, 0.75 and 0.52 respectively (the recommended threshold for stable estimation is >0.5 [25]). Details are available in the Supplementary Materials.
Accepted at Clinical Psychological Science

42. Marloes Eidhof Sabina Palic Giulio Costantini Mathilde Huisman
Cherie Armour
Claudi Bockting
Iris Engelhard
Karen-Inge Karstoft
VGCT November 2017

43. Thank you!

44. Slides & more information available at:
Eiko Fried
Slides & more information available at: