Identifying Treatment Targets in Idiographic Psychopathological Networks: A Simulation Study

doi:10.21203/rs.3.rs-2499311/v1

Download PDF

Article

Identifying Treatment Targets in Idiographic Psychopathological Networks: A Simulation Study

https://doi.org/10.21203/rs.3.rs-2499311/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Following current trends for precision medicine, psychology has in recent years turned its attention to treatment personalization. Several technological and methodological advances have contributed to this renewed focus on psychology. One of these methodological advances at the center of this push toward precision medicine is network analysis. One of the core propellers of the network theory of psychopathology is the proposal that highly influential symptoms of the network (i.e., central symptoms) might be promising treatment targets. However, evidence for this proposal remains inconclusive, with evidence from idiographic longitudinal assessments lacking.

In this study, we explore the impact of symptom deactivation in individual networks (named idiographic networks). We tested two types of symptom deactivation: normal and cascade attack, and five measures for highly influential symptom identification (degree, strength, eigenvector, expected influence, and random attack condition). We find that a cascade degree-based attack is more effective at deactivating the symptom network, we next discuss the implications for treatment personalization and precision psychology.

In recent years, the need for personalized treatments in psychology became evident ^1–3 and possible due to the advances in technological and methodological approaches in idiographic longitudinal assessments ⁴, being network analysis the most promising methodological approach for that end ^5–7. In the last decade, network analysis has been growing as a field of study in mental disorders ^8–10. In this approach, mental disorders are seen as a network of symptoms, where symptoms are viewed as nodes and connections between them as edges ¹¹. Network Analysis pinpoints target treatment by identifying highly influential symptoms which might enable evidence-based treatment personalization and therefore improve the psychological treatments.

Mental disorders are maintained by the persistence of the highly influential symptoms, which are deeply embedded in the symptoms network¹². Therefore, identification of those symptoms can be performed by network centrality measures, such as degree, strength, betweenness, and closeness, which uncover each symptoms’ connectivity. However, recent evidence showed that closeness and betweenness centrality are not adequate in psychopathological networks, given their bias in the covariance and sampling variability ^13,14. In other words, despite strength and degree centralities not considering negative connections (commonly present in psychological networks), they are the only viable markers for identifying central in psychopathological networks. Thus, to surpass these limitations, one-step, and two-step expected influence have been introduced as centrality measures in psychopathological networks ¹⁵.

The network theory of psychopathology proposes that when a central symptoms is activated (such as by an external event) and signal diffuses through the network, activating other symptoms it leads to a cascading effect^12,16. Moreover, once the system transitions into a disease state, these central symptoms have an essential role in sustaining the disease given their high number of connections. In this context, network density has been suggested as a network property that can act as a biomarker for predicting mental disorders ^17,18, their severity ¹⁹, while networks with lower density were found to be associated with reduced severity of symptoms ^{7, 20–22}. However, studies examining the central symptoms as psychotherapeutic targets displayed inconclusive results ^23–26.

Previous studies found weak to moderate association between central symptoms for predicting changes in the remaining symptoms ^15,27 and facilitate the evolution of a psychopathological condition ²⁸.. For example, symptom endorsement is a better predictor of change than centrality measures^24–26, and the most common centrality measures used in psychology are not related to causality, except for eigenvector centrality²⁹. Suggesting the need for other centrality measures to identifying therapeutic targets.

In addition, deactivating symptoms according to their centrality or bridging capacity do not significantly reduce network density than the random deactivation of symptoms³⁰. Despite this, and in line with the studies that reported changes in the network structure when comparing psychological networks at different stages ^31–36, when symptoms are deactivated according to their degree centrality there are significant changes in the number of components of the networks³⁷. However, symptoms were deactivated based only on the initial estimation of central symptoms, ignoring changes that might occur in the network structure after the deactivation of a symptom³⁰. Which might have a substantial impact on the results since the deactivating order of the symptoms may not have identified the most central symptom at a certain point in the network.

More importantly, most of these studies were performed using cross-sectional networks ^{23–26, 38,39} which might have hampered their findings, due to individuals idiosyncrasy that is lost in the cross-sectional analysis ⁴⁰. As a potential framework to unveil the individual-level differences one can leverage idiographic networks⁴¹, which allow for a detailed understanding of the associations between symptoms, their directionality, and how different processes (e.g., thoughts, feelings, and behaviors) fluctuate over time ^17,41. It is based on this dynamicity that we can understand the diversity of clinical and symptomatologic trajectories. In turn, this idiographic dynamic view, coupled with the hypothesis of central symptoms being more efficient treatment targets, might promote the development of better personalized treatments. However, the initial proposal regarding the centrality of symptoms as the best therapeutic targets remains unexplored in idiographic networks.

In this context, network studies results remain inconclusive and suggest that i) the exploration of new centrality metrics that might be more useful to detect treatment targets and that ii) the application of these methods to idiographic research is needed. Considering this, the hypothesis that central symptoms might be valuable therapeutic targets remains open. Here we address these hypotheses by assessing the impact of symptom deactivation on idiographic networks according to different centrality measures. We do it, by comparing the impact of symptom deactivation using two different types of symptom deactivation procedures, one procedure based only on a single point estimate of centrality measures and a second procedure in which the centrality measures are re-estimated every time a symptom is deactivated. As usual in network science, the impact of symptom deactivation according to the centrality estimates is evaluated in comparison to the random deactivation of symptoms.

Networks

Here we analyze contemporaneous idiographic networks of depression and anxiety symptoms from 40 participants⁶, which, in the original study, aimed to explore the idiographic structure of mood and anxiety symptomatology via contemporaneous and temporal networks. The participants had to have a diagnosis of major depressive disorder (MDD) or generalized anxiety disorder (GAD) and an age of 18 to 65 years, participants with a history of psychosis or mania were excluded. Most of the participants were female (65%), and 25 participants met the criteria for current GAD and 15 for current MDD (Table S1), network basic features are presented in Table 1. Participants rated their symptoms for 30 days, four times per day through an Experience Sample Survey (ESS), which we used here to estimate temporal and contemporaneous networks of MDD and GAD symptoms for each participant. The ESS consisted of DSM-5 symptoms for MDD and GAD, where each participant rated their experience of each symptom in the preceding hours on a 0-100 scale and provided on average 130.43 (SD = 19.27) observations.

Table 1

Network descriptive analysis.
Network Characteristics	Networks (n = 40)
Network Characteristics	M (SD)	Median (MAD)	Minimum-maximum
Nodes	21	21 (0.00)
Edges	96.075 (15.416)	101.00 (10.38)	61–124
Density	0.457 (0.073)	0.48 (0.05)	0.290–0.590
Components	1	1 (0.00)
Average Path Length	1.564 (0.114)	1.52 (0.05)	1.409–1.909

Here we focus on the contemporaneous networks, which reflect the connections between nodes within individuals averaged across time. Contemporaneous networks inform us about the average network topology of each symptom, but opposite to temporal networks, they do not provide the connections directionality. While temporal connections are important to understand how the network developed into a certain state, the connections’ directionality provide limited treatment insights. In other words, if a certain symptom is a common downstream endpoint (i.e., has an elevated number of incoming connections) it might still be a valuable treatment target, despite the symptomatology not originating in that symptom. For example, a person with social anxiety tends to avoid social interactions (downstream endpoint) because of the experienced anxiety, but it’s still a primary target in cognitive-behavioral treatment ⁴². Thus, within-person contemporaneous networks, through the average network structure, might provide better identification of treatment targets, by identifying those symptoms that are more “active” in the network. Therefore, here we focus on contemporaneous idiographic networks to identify psychotherapeutic targets using centrality metrics.

Data analyses

For each participant, in the original study⁶, a contemporaneous correlation matrix was extracted and then a sparse partial correlation network was estimated using the LASSO regularization method¹³. To identify, the central nodes and their effect on individual networks, we perform a two-step analysis: i) the identification of central symptoms and ii) the exploration of the differential impact of symptoms’ deactivation in the network. The network exploration is performed as a cascade-based attack, where symptoms are deactivated in their decreasing order according to their centrality, which is iteratively calculated at every symptom removal. We compare our novel approach to the normal attack³⁰, where symptoms were deactivated according to their original centrality, and to a random symptom deactivation procedure, where symptoms are randomly deactivated.

In the cascade-based attack and in the normal attack procedures’ we identified the central symptoms through five different centrality measures: strength centrality, degree centrality, one-step and two-step expected influence centrality, and eigenvector centrality. These centrality measures were chosen due to their extensive use in psychopathological networks² in addition to being alternatives for better treatment targets identification^29,30. After each symptom deactivation, a set of network properties was measured (i.e., network density, number of components, and average path length). The network density is the sum of all the symptoms’ connections and weights in the network and might give us information regarding symptom activity⁷, while the number of components in the network refers to the number of symptoms or groups of symptoms that are disconnected from the rest of the network and might be able to help us with the identification of groups of active symptoms. Finally, the average path length concerns the mean of the shortest paths in the networks and might identify if network symptoms activation is more likely due to the shortest distance between network symptoms.

For each centrality measure, we assessed the differential impact of symptom deactivation by computing the magnitude and extent of the attack in the network, where the impact magnitude consists of the difference between the maximum values of average path length and components and their initial values. We next measured the attack extent by computing the proportion of nodes that needed to be deactivated to achieve the maximum value of average path length and number of components (Fig. 1). To assess the impact of symptom deactivation on the network density we computed the density of the network at 50% of the symptoms deactivated. All these analyses were performed in the package psychNetsAttack ⁴³ for R ⁴⁴.

To compare each centrality measure attack magnitude and extent on the number of components, average path length, and network density for both normal and cascade attack we used a Friedmann’s Test, we followed by a Kendall’s coefficient of concordance to estimate the effect and Durbin-Conover test for post-hoc test. We performed this analysis in the R ⁴⁴ package ggstatsplot ⁴⁹.

For a normal attack, the extent and magnitude of the attack on the number of components and the average path length of each centrality measure are presented in Table 2. Normal attack distributions for attack extent and magnitude of these properties can be seen in Fig. 2.

Table 2

Descriptives for normal attack magnitude and extent on the number of components and average path length and network density with 50% of the nodes deactivated.
Network characteristic / attack condition	Attack Magnitude			Attack Extent
Network characteristic / attack condition	M (SD)	Median (MAD)	Minimum - maximum	M (SD)	Median (MAD)	Minimum - maximum
Components
Degree	2.30 (1.26)	2.00 (1.48)	1–6	0.72 (0.08)	0.74 (0.04)	0.48–0.81
Strength	1.02 (0.92)	1.00 (0.00)	0–4	0.55 (0.36)	0.71 (0.21)	0–0.90
Expected Influence 1-step	1.02 (0.92)	1.00 (0.00)	0–4	0.55 (0.36)	0.71 (0.21)	0–0.90
Expected Influence 2-step	0.95 (0.85)	1.00 (0.00)	0–4	0.54 (0.37)	0.76 (0.14)	0–0.90
Eigenvector	0.88 (0.82)	1.00 (1.48)	0–3	0.49 (0.40)	0.67 (0.35)	0–0.90
Random	0.95 (0.85)	1.00 (0.00)	0–3	0.51 (0.36)	0.76 (0.14)	0–0.90
Average Path Length
Degree	0.74 (0.34)	0.65 (0.28)	0.28–1.50	0.56 (0.10)	0.57 (0.07)	0.33–0.71
Strength	0.20 (0.15)	0.17 (0.18)	0–0.50	0.52 (0.25)	0.62 (0.21)	0–0.81
Expected Influence 1-step	0.20 (0.15)	0.17 (0.18)	0–0.50	0.52 (0.25)	0.62 (0.21)	0–0.81
Expected Influence 2-step	0.16 (0.13)	0.15 (0.12)	0–0.55	0.48 (0.26)	0.50 (0.25)	0–0.81
Eigenvector	0.23 (0.23)	0.16 (0.17)	0–0.89	0.52 (0.21)	0.55 (0.25)	0.05–0.76
Random	0.28 (0.16)	0.18 (0.18)	0.01–0.68	0.59 (0.15)	0.62 (0.14)	0.19–0.81
	50% of nodes deactivated
	M (SD)		Median (MAD)		Minimum - Maximum
Density
Degree	0.31 (0.09)		0.32 (0.12)		0.09–0.44
Strength	0.47 (0.12)		0.48 (0.12)		0.16–0.67
Expected Influence 1-step	0.47 (0.12)		0.48 (0.12)		0.16–0.67
Expected Influence 2-step	0.48 (0.11)		0.47 (0.13)		0.20–0.67
Eigenvector	0.46 (0.11)		0.49 (0.08)		0.20–0.67
Random	0.47 (0.09)		0.50 (0.08)		0.31–0.69

We observed a statistically significant effect of centrality measures on the magnitude of the normal attack on the number of components (X²_F (5) = 71.02, p < .001, W_k = 0.40, 95% CI [0.36, 0.61]) and on the average path length (X²_F (5) = 79.14, p < .001, W_k = 0.43, 95% CI [0.34, 0.74]), post hoc comparisons suggest that degree centrality had a significant higher attack magnitude on the number of components and on the average path length than all the other centrality measures and the random attack. The extent of the normal attack on the number of components (X²_F (5) = 0.47, p = 0.993, W_k = 0.23, 95% CI [0.20, 0.71]) and on the average path (X²_F (5) = 6.26, p = 0.281, W_k = 0.33, 95% CI [0.32, 0.63]) did not showed statistically significant effects.

Regarding the cascade attack, the extent and magnitude of the different centrality measures are presented in Table 3, and Fig. 3 shows the distributions for the different centrality measures. Magnitude of the attack showed statistically significant effect of centrality measures on the median attack magnitude on the number of components, X²_F (5) = 165.57, p < .001, W_k = 0.35, 95% CI [0.33, 0.59], and on the average path X²_F (5) = 129.37, p < .001, W_k = 0.40, 95% CI [0.32, 0.76]. Post hoc comparisons suggest that degree centrality and eigenvector centrality yielded a significantly higher attack magnitude on the number of components than the random attack and the remaining centrality measures. The random attack also had a statistically higher attack magnitude on the number of components than strength centrality and expected influence one-step and two-step. On the average path length, post hoc comparisons on the cascade attack magnitude were also significantly higher for degree centrality and eigenvector centrality. The remaining centrality measures did not present significant differences between them.

Table 3

Descriptives for cascade attack magnitude and extent on the number of components and average path length and network density with 50% of the nodes deactivated.
Network characteristic / attack condition	Attack Magnitude			Attack Extent
Network characteristic / attack condition	M (SD)	Median (MAD)	Minimum - maximum	M (SD)	Median (MAD)	Minimum - maximum
Components
Degree	4.20 (0.88)	4.00 (0.74)	2.00–6.00	0.68 (0.06)	0.71 (0.07)	0.52–0.76
Strength	0.55 (0.90)	0.00 (0.00)	0.00–4.00	0.23 (0.32)	0.00 (0.00)	0.00–0.86
Expected Influence 1-step	0.55 (0.90)	0.00 (0.00)	0.00–4.00	0.23 (0.32)	0.00 (0.00)	0.00–0.86
Expected Influence 2-step	0.55 (0.90)	0.00 (0.00)	0.00–4.00	0.23 (0.32)	0.00 (0.00)	0.00–0.86
Eigenvector	3.50 (1.24)	3.00 (1.48)	2.00–7.00	0.75 (0.06)	0.76 (0.07)	0.62–0.86
Random	0.98 (0.73)	1.00 (0.00)	0.00–3.00	0.59 (0.36)	0.76 (0.14)	0–0.90
Average Path Length
Degree	1.09 (0.35)	1.11 (0.28)	0.51–2.05	0.52 (0.08)	0.52 (0.07)	0.33–0.62
Strength	0.26 (0.21)	0.18 (0.13)	0.05–0.97	0.55 (0.17)	0.57 (0.18)	0.14–0.81
Expected Influence 1-step	0.26 (0.21)	0.18 (0.13)	0.05–0.97	0.55 (0.17)	0.57 (0.18)	0.14–0.81
Expected Influence 2-step	0.26 (0.21)	0.18 (0.13)	0.05–0.97	0.55 (0.17)	0.57 (0.18)	0.14–0.81
Eigenvector	0.66 (0.26)	0.61 (0.32)	0.26–1.23	0.60 (0.09)	0.62 (0.07)	0.33–0.76
Random	0.25 (0.24)	0.18 (0.18)	0.00–1.14	0.55 (0.19)	0.62 (0.14)	0.00–0.81
	50% of nodes deactivated
	M (SD)		Median (MAD)		Minimum - Maximum
Density
Degree	0.23 (0.07)		0.26 (0.05)		0.07–0.36
Strength	0.45 (0.11)		0.47 (0.10)		0.13–0.69
Expected Influence 1-step	0.45 (0.11)		0.47 (0.10)		0.13–0.69
Expected Influence 2-step	0.45 (0.11)		0.47 (0.10)		0.13–0.69
Eigenvector	0.35 (0.11)		0.38 (0.07)		0.07–0.58
Random	0.47 (0.10)		0.48 (0.12)		0.24–0.64

Similarly, cascade attack extent on the number of components returned a statistically significant effect X²_F (5) = 101.05, p < .001, W_k = 0.22, 95% CI [0.21, 0.56] with the faster centrality measures to achieve maximum value on the number of components being expected influence one-step and two-step and strength centrality. However, as aforementioned, the attack extension is the proportion of symptoms that need to be deactivated to achieve the maximum value. Thus, the lower the score in the attack extension the better. However, expected influence one-step and two-step and strength centrality, as can be seen in the attack magnitude results, did not promote any change in the network (M_dn = 0). Accordingly, the maximum value equals the initial value of the network and, consequently, the median number of symptoms that need to be deactivated to achieve the maximum number of components is 0. The two centrality measures that promoted a change in the number of components (attack magnitude), degree and eigenvector centrality, did not show any significant differences with the random attack on the attack extent. On the average path length, the cascade attack extent presented a statistically significant difference, X²_F (5) = 19.89, p = 0.001, W_k = 0.50, 95% CI [0.45, 0.89] with post hoc comparisons suggesting a better performance of degree centrality over the other centrality measures and random attack.

Concerning network density with 50% of the symptoms deactivated for normal and cascade attacks, distributions can be found in Fig. 4. Normal attack showed significant effects between centrality measures, X²_F (5) = 89.68, p < .001, W_k = 0.70, 95% CI [0.66, 0.94]. Post hoc comparisons suggest that degree centrality has a significantly higher impact in reducing the network density in comparison to all other centrality measures and the random attack. For the cascade attack significant effects were also found, X²_F (5) = 139.08, p < .001, W_k = 0.69, 95% CI [0.67, 0.88]. As observed in the post hoc comparisons, the eigenvector centrality cascade attack promoted a higher decrease in network density than all other centrality measures, except degree centrality, and the random attack. Degree centrality promoted the most significant reduction of the network density, outperforming all other measures, including eigenvector centrality, and random attack.

Attending to these results, a post hoc analysis was made to compare the magnitude and extent of the effect of degree centrality under normal and cascade attacks on the network characteristics. Results suggest that the degree centrality cascade attack outperformed the normal attack in all the network properties examined, yielding a higher magnitude, a lower extension, and density. Results from this analysis are presented in Figs. 4 and 5.

The network theory of psychopathology has suggested that central symptoms might be valuable therapeutic targets, due to their proposed ability to fasten the deactivation of connections between symptoms ^{11,15, 50–55}. This is one of the core propellers of the network theory of psychopathology, that lead to its growth in recent years ⁸. However, evidence for this hypothesis is still scarce with studies focusing on cross-sectional networks and grounding the identification of possible therapeutic targets on the initial estimations of centrality measures ^{24–26, 30}. Due to these inconclusive results, it has been recognized that there are changes in symptoms centrality that occur during treatment ⁵⁶ and that idiographic networks might be more appropriate to identify treatment targets ⁴¹. This might have important implications for treatment personalization.

In this context, we explored the impact of deactivating symptoms in contemporaneous idiographic networks through two distinct procedures. The first is based on a single time point estimate of network centrality (normal attack), and a second procedure, where, after each symptom deactivation, centrality measures are estimated again (cascade attack). The impact of symptom deactivation was assessed through a set of network properties since it has been suggested changes in the network density might be able to differentiate between different clinical presentations ^17,18. However, due to the conflicting results in previous studies regarding the association between symptoms remission and networks density ^23,38 and the identified changes in the network topology ^31–36 we have explored the impact of symptom deactivation in two more network properties, average path length and the number of components.

Globally, our results suggest that changes in psychopathological network structure are best achieved through degree centrality. In comparison with the most common centrality metrics in psychopathological networks (i.e., strength centrality and expected influence one-step and two-step), the deactivation of symptoms by the absolute number of connections (i.e., degree centrality) seems to have a higher impact on the network structure. A previous study using cross-sectional networks ³⁰ also found that degree centrality was the only centrality measure that was able to produce significant changes in the network structure. However, this study ³⁰ only found significant changes in the number of components of the networks. In turn, the present study suggests that for contemporaneous within-person networks all the three network properties examined are transformed through a degree-based attack. These results suggest that different properties might be of interest according to the nature of the network (nomothetic or idiographic). Previous research has also pointed to this need for further exploration and clarification of psychological network structure ⁵⁷ and the impact of the networks’ structural properties in the selection of centrality measures ⁵⁸.

With the field focusing on the strength centrality and expected influence measures to identify important symptoms in the network, it’s of relevance that neither of these measures was able to promote significant changes in the network structure. In fact, the random deactivation of symptoms revealed a significantly higher impact magnitude in the number of components than a cascade attack through expected influence one and two-step and strength centrality. Thus, if changes in a person symptomatology are identifiable by changes in the network structure, the traditional psychopathological centrality metrics do not seem able to induce significant changes. Consequently, this might explain the inconclusive results in previous studies that explored if these centrality measures were related to changes in symptomatology ^24–26.

It has been suggested that all centrality measures make implicit assumptions about the network processes of node-to-node transmission and the type of trajectories followed ^58,59. The case may be that common centrality metrics in psychopathological networks are not accessing the specific processes that occur in these networks or are accessing some other processes that are not related to network transformation. For example, they might be identifying emergent phenomena in the network that needs to be addressed (e.g., a very active symptom) but not phenomena related to disorder maintenance (e.g., symptoms that sustain the disorder). However, in psychological networks, the processes within the networks that generate and maintain mental disorders are still unknown. Interestingly, with a cascade attack, eigenvector centrality produced significant changes in the network structure. This might be due to its suggested proximity to the causality structure of the network ²⁹ and might mean that this measure is tapping into a specific process in psychological networks. Understanding these processes will advance the identification of treatment targets by enabling an enhanced selection of centrality metrics.

Besides exploring which network centrality metric promoted changes in the network structure, we have also tested two types of attacks, normal and cascade. Although degree centrality had a better performance than any other measure in both attacks. In the cascade attack, the magnitude and the extension were significantly higher than in a normal attack. This suggests that might be of importance to estimate symptoms’ centrality each time before an intervention is deployed to act on the symptom with the highest degree at any given time point. The dynamic fluctuations of central symptoms during a psychotherapeutic process have been highlighted by previous studies ⁶⁰ and our results suggest that assessing and intervening in which symptom is central at any given time-point might produce faster recoveries. Consequently, the estimation at a single time-point of centrality measures to establish treatment targets for intervention might not be the most effective procedure to promote changes in the network structure.

This has important implications for treatment personalization. Our results suggest that to promote more effective treatments the assessment of the central symptoms must be done each time before the intervention is done. Meaning that, in the context of idiographic networks, symptomatology needs to continuously be assessed through, for example, ecological momentary assessments ^61,62 for the duration of treatment to determine, at each session, in which symptom the treatment should focus. However, this leads to another important question, do psychotherapeutic strategies and psychopharmacological treatments have the specificity needed to act on a specific central symptom at each time? And is this a negative constraint of the treatments or is it a positive consequence? The answers to these questions are still unknown but the first results seem promising ⁶³ and network analysis has all the right tools to promote more effective and personalized treatments.

Besides this, some limitations of our study should be pointed out. First, we used only centrality measures, leaving another important concept of psychopathological networks, bridge symptoms, outside of our study. Bridge symptoms have been proposed as symptoms that connect two different disorders and that acting in these symptoms might promote a faster disintegration of a comorbidity network ^48,64. Our network is a comorbidity network comprising symptoms of depression and anxiety and identifying and deactivating bridge symptoms might have led to faster disintegration of the network. Another important limitation in our work is that we assume, as has been proposed in previous studies ^17,18, that network properties identify psychological states, although this hypothesis is currently lacking consistent evidence ^23,65. In previous studies that focused on the idiographic network, a relationship between network density and psychopathological states was found ^17,18 and our results show that there are clear changes in the network density after the deactivation of 50% of the symptoms. However, we also use two network properties rarely studied in psychopathological networks and without a clear theoretical formulation, although we think it’s important to explore these new properties, we also acknowledge that there’s a need to theoretically frame these properties.

Our study provides the first simulation study in idiographic networks to examine symptom deactivation through several centrality measures. The emergence of degree centrality as a measure more suitable to transform the network might be of relevance for further studies trying to identify treatment targets through network analysis. Further exploration of network properties is needed, but, if changes in the network structure are aligned with psychopathological and healthy states, deactivating symptoms through a cascade attack based on degree centrality might promote faster and more effective treatments.

Author’s Note

Correspondence concerning this paper should be addressed to Daniel Castro, Social Sciences Department, University of Maia, Avenida Carlos Oliveira Campos – Castêlo da Maia, 4475-690, Maia, Portugal. Email: [email protected]

This work was funded by the Center for Psychology at the University of Porto, Portuguese Science Foundation (FCT UIDB/00050/2020), and EU FEDER through COMPETE 2020 program (POCI-01-0145-FEDER-007294). The first author is supported by the Portuguese Foundation for Science and Technology through the Ph.D. grant: SFRH/BD/148884/2019.

Data Availability

Datasets analyzed in the current study are publicly available (available at https://osf-io/mgdp6) and was not collected by the authors of this manuscript. Thus, the approval of the protocol, the accordance of all the methods with relevant guidelines and regulations, and the informed consent were obtained in a previous published study by different authors⁶. The authors of this manuscript had no control over these data collection procedures.

David, S. J., Marshall, A. J., Evanovich, E. K. & Mumma, G. H. Intraindividual Dynamic Network Analysis – Implications for Clinical Assessment. Journal of Psychopathology and Behavioral Assessment 40, 235–248 (2018).
Roche, M. J., Pincus, A. L., Rebar, A. L., Conroy, D. E. & Ram, N. Enriching Psychological Assessment Using a Person-Specific Analysis of Interpersonal Processes in Daily Life. Assessment 21, 515–528 (2014).
Hofmann, S. G. & Hayes, S. C. The Future of Intervention Science: Process-Based Therapy. Clinical Psychological Science 7, 37–50 (2019).
Wright, A. G. C. & Woods, W. C. Personalized Models of Psychopathology. Annual Review of Clinical Psychology 16, 49–74 (2020).
Bos, F. M. et al. Exploring the emotional dynamics of subclinically depressed individuals with and without anhedonia: An experience sampling study. Journal of Affective Disorders 228, 186–193 (2018).
Fisher, A. J., Reeves, J. W., Lawyer, G., Medaglia, J. D. & Rubel, J. A. Exploring the idiographic dynamics of mood and anxiety via network analysis. Journal of Abnormal Psychology 126, 1044–1056 (2017).
van Borkulo, C. et al. Association of Symptom Network Structure With the Course of Depression. JAMA Psychiatry 72, 1219 (2015).
Contreras, A., Nieto, I., Valiente, C., Espinosa, R. & Vazquez, C. The study of psychopathology from the network analysis perspective: A systematic review. Psychotherapy and Psychosomatics 88, 71–83 (2019).
Fried, E. I. et al. Mental disorders as networks of problems: a review of recent insights. Social Psychiatry and Psychiatric Epidemiology 52, 1–10 (2017).
Robinaugh, D. J., Hoekstra, R. H. A., Toner, E. R. & Borsboom, D. The network approach to psychopathology: A review of the literature 2008-2018 and an agenda for future research. Psychological Medicine (2019). doi:10.1017/S0033291719003404
Borsboom, D. & Cramer, A. O. J. J. Network Analysis: An Integrative Approach to the Structure of Psychopathology. Ssrn 9, 91–121 (2013).
Borsboom, D., Cramer, A. O. J. & Kalis, A. Brain disorders? Not really: Why network structures block reductionism in psychopathology research. Behavioral and Brain Sciences 42, (2019).
Epskamp, S., Borsboom, D. & Fried, E. I. Estimating psychological networks and their accuracy: A tutorial paper. Behavior Research Methods 50, 195–212 (2018).
Hallquist, M. N., Wright, A. G. C. & Molenaar, P. C. M. Problems with Centrality Measures in Psychopathology Symptom Networks: Why Network Psychometrics Cannot Escape Psychometric Theory. Multivariate Behavioral Research 0, 1–25 (2019).
Robinaugh, D. J., Millner, A. J. & McNally, R. J. Identifying highly influential nodes in the complicated grief network. Journal of Abnormal Psychology 125, 747–757 (2016).
Borsboom, D. A network theory of mental disorders. World psychiatry : official journal of the World Psychiatric Association (WPA) 16, 5–13 (2017).
Wichers, M. & Groot, P. C. Critical Slowing Down as a Personalized Early Warning Signal for Depression. Psychotherapy and Psychosomatics 85, 114–116 (2016).
Pe, M. L. et al. Emotion-Network Density in Major Depressive Disorder. Clinical Psychological Science 3, 292–300 (2015).
Koenders, M. A. et al. A network approach to bipolar symptomatology in patients with different course types. PLoS ONE 10, 1–16 (2015).
Santos, H., Fried, E. I., Asafu-Adjei, J. & Jeanne Ruiz, R. Network structure of perinatal depressive symptoms in latinas: Relationship to stress and reproductive biomarkers. Research in Nursing and Health 40, 218–228 (2017).
Stone, D. I., DuBois, R., DeJesus, B., Rodgers, R. F. & Rizzo, C. Brief report: A network analysis of self-cutting risk among late adolescent girls exposed to dating violence. Journal of Adolescence 61, 12–16 (2017).
van Rooijen, G. et al. A symptom network structure of the psychosis spectrum. Schizophrenia Research 189, 75–83 (2017).
Bos, F. M. et al. Cross-sectional networks of depressive symptoms before and after antidepressant medication treatment. Social Psychiatry and Psychiatric Epidemiology 53, 617–627 (2018).
Rodebaugh, T. L. et al. Does centrality in a cross-sectional network suggest intervention targets for social anxiety disorder? Journal of Consulting and Clinical Psychology 86, 831–844 (2018).
Spiller, T. R. et al. On the validity of the centrality hypothesis in cross-sectional between-subject networks of psychopathology. BMC Medicine 18, 0–14 (2020).
Papini, S., Rubin, M., Telch, M. J., Smits, J. A. J. & Hien, D. A. Pretreatment Posttraumatic Stress Disorder Symptom Network Metrics Predict the Strength of the Association Between Node Change and Network Change During Treatment. Journal of Traumatic Stress 33, 64–71 (2020).
Elliott, H., Jones, P. J. & Schmidt, U. Central Symptoms Predict Posttreatment Outcomes and Clinical Impairment in Anorexia Nervosa: A Network Analysis. Clinical Psychological Science 8, 139–154 (2020).
Boschloo, L., Van Borkulo, C. D., Borsboom, D. & Schoevers, R. A. A Prospective Study on How Symptoms in a Network Predict the Onset of Depression. Psychotherapy and Psychosomatics 85, 183–184 (2016).
Dablander, F. & Hinne, M. Node centrality measures are a poor substitute for causal inference. Scientific Reports 9, (2019).
Castro, D. et al. The Differential Role of Central and Bridge Symptoms in Deactivating Psychopathological Networks. Frontiers in Psychology 10, (2019).
Curtiss, J., Fulford, D., Hofmann, S. G. & Gershon, A. Network dynamics of positive and negative affect in bipolar disorder. Journal of Affective Disorders 249, 270–277 (2019).
Groen, R. N., Wichers, M., Wigman, J. T. W. & Hartman, C. A. Specificity of psychopathology across levels of severity: a transdiagnostic network analysis. Scientific Reports 9, 18298 (2019).
Hartung, T. J., Fried, E. I., Mehnert, A., Hinz, A. & Vehling, S. Frequency and network analysis of depressive symptoms in patients with cancer compared to the general population. Journal of Affective Disorders 256, 295–301 (2019).
Kraft, B. et al. Attention Bias Modification in Remitted Depression Is Associated With Increased Interest and Leads to Reduced Adverse Impact of Anxiety Symptoms and Negative Cognition. Clinical Psychological Science 7, 530–544 (2019).
Montazeri, F., de Bildt, A., Dekker, V. & Anderson, G. M. Network Analysis of Behaviors in the Depression and Autism Realms: Inter-Relationships and Clinical Implications. Journal of Autism and Developmental Disorders 0, 0 (2019).
Silk, T. J. et al. A network analysis approach to ADHD symptoms: More than the sum of its parts. PLoS ONE 14, 1–17 (2019).
Ferreira, F., Castro, D., Araújo, A. S., Fonseca, A. R. & Ferreira, T. B. Exposure to Traumatic Events and Development of Psychotic Symptoms in a Prison Population: A Network Analysis Approach. Psychiatry Research 286, (2020).
Beard, C. et al. Network analysis of depression and anxiety symptom relationships in a psychiatric sample. Psychological Medicine 46, 3359–3369 (2016).
Schweren, L., van Borkulo, C. D., Fried, E. & Goodyer, I. M. Assessment of Symptom Network Density as a Prognostic Marker of Treatment Response in Adolescent Depression. JAMA Psychiatry 75, 98 (2018).
Fisher, A. J. Toward a dynamic model of psychological assessment: Implications for personalized care. Journal of Consulting and Clinical Psychology 83, 825–836 (2015).
Bringmann, L. F. et al. A Network Approach to Psychopathology: New Insights into Clinical Longitudinal Data. PLoS ONE 8, (2013).
Hofmann, S. G. & DiBartolo, P. M. Social Anxiety. Social Anxiety: Clinical, Developmental, and Social Perspectives: Third Edition (Elsevier, 2014). doi:10.1016/C2011-0-06934-X
Castro, D. psychNetsAttack. (2021).
R Core Team. R: A Language and Environment for Statistical Computing. (2017).
Csardi, G. & Nepusz, T. The igraph software package for complex network research. InterJournal Complex Sy, 1695 (2006).
Epskamp, S., Cramer, A. O. J., Waldorp, L. J., Schmittmann, V. D. & Borsboom, D. Qgraph: Network visualizations of relationships in psychometric data. Journal of Statistical Software 48, (2012).
Mueller, L. A. J., Kugler, K. G., Graber, A., Emmert-streib, F. & Dehmer, M. Structural Measures for Network Biology Using QuACN. (2011).
Jones, P. J., Ma, R. & McNally, R. J. Bridge Centrality: A Network Approach to Understanding Comorbidity. Multivariate Behavioral Research 0, 1–15 (2019).
Patil, I. Visualizations with statistical details: The ‘ggstatsplot’ approach. Journal of Open Source Software 6, (2021).
Bekhuis, E., Schoevers, R. A., Van Borkulo, C. D., Rosmalen, J. G. M. & Boschloo, L. The network structure of major depressive disorder, generalized anxiety disorder and somatic symptomatology. Psychological Medicine 46, 2989–2998 (2016).
Bryant, R. A. et al. Acute and chronic posttraumatic stress symptoms in the emergence of posttraumatic stress disorder a network analysis. JAMA Psychiatry 74, 135–142 (2017).
Knefel, M., Tran, U. S. & Lueger-Schuster, B. The association of posttraumatic stress disorder, complex posttraumatic stress disorder, and borderline personality disorder from a network analytical perspective. Journal of Anxiety Disorders 43, 70–78 (2016).
McNally, R. J. et al. Mental disorders as causal systems: A network approach to posttraumatic stress disorder. Clinical Psychological Science 3, 836–849 (2015).
Olatunji, B. O., Levinson, C. & Calebs, B. A network analysis of eating disorder symptoms and characteristics in an inpatient sample. Psychiatry Research 262, 270–281 (2018).
Richetin, J., Preti, E., Costantini, G. & De Panfilis, C. The centrality of affective instability and identity in Borderline Personality Disorder: Evidence from network analysis. PLoS ONE 12, (2017).
Park, S. C. et al. Network analysis of the symptoms of depressive disorders over the course of therapy: Changes in centrality measures. Psychiatry Investigation 18, 48–58 (2021).
Christensen, A. P. & Golino, H. Random, factor, or network model? Predictions from neural networks. (2020).
Bockholt, M. & Zweig, K. A. Towards a process-driven network analysis. Applied Network Science 5, (2020).
Borgatti, S. P. Identifying sets of key players in a social network. Computational and Mathematical Organization Theory 12, 21–34 (2006).
Park, M. & Mcdonald, D. W. Perception Differences between the Depressed and Non-Depressed Users in Twitter. 476–485
Shiffman, S., Stone, A. A. & Hufford, M. R. Ecological Momentary Assessment. Annual Review of Clinical Psychology 4, 1–32 (2008).
Robinaugh, D. J. et al. Towards a precision psychiatry approach to anxiety disorders with ecological momentary assessment: The example of panic disorder. General Psychiatry 33, 1–7 (2020).
Blanken, T. F., Borsboom, D., Penninx, B. W. & Van Someren, E. J. Network outcome analysis identifies difficulty initiating sleep as a primary target for prevention of depression: a 6-year prospective study. Sleep 43, (2020).
Cramer, A. O. J., Waldorp, L. J., Van Der Maas, H. L. J. & Borsboom, D. Comorbidity: A network perspective. Behavioral and Brain Sciences 33, 137–150 (2010).
Beard, C. Cognitive bias modification for anxiety: Current evidence and future directions. Expert Review of Neurotherapeutics 11, 299–311 (2011).

No competing interests reported.

IdentifyingtreatmenttargetsSM.docx.zip

Download PDF

Version 1

posted

You are reading this latest preprint version

Identifying Treatment Targets in Idiographic Psychopathological Networks: A Simulation Study

Status:

Version 1

Abstract

Figures

Introduction

Method

Networks

Data analyses

Results

Discussion

Conclusion

Declarations

Data Availability

References

Additional Declarations

Supplementary Files

Status:

Version 1