A network analysis of risk factors for suicide in Iraq/Afghanistan-era veterans

Abstract

Suicidal ideation (SI) is a prevalent issue in the veteran population. A number of factors have been identified as risk factors for suicidal ideation (SI) in veterans, including suicide attempts, depression, posttraumatic stress disorder (PTSD), and drug use. However, clinicians' ability to predict suicide is poor, particularly given the interplay between various factors such as previous suicide attempts. As such, there is a gap in our knowledge of which factors most saliently predict suicide risk and which should be targets for interventions designed to lower SI. Network analysis, a method allowing for an examination of how variables relate within the context of a network of factors, may bridge this gap by simultaneously evaluating the interrelationships between risk factors for suicide in veterans. Current study used network analysis and data from 2268 Iraq/Afghanistan-era military veterans to examine the relationships between suicidal ideation and several factors related to suicide risk, such as past suicide attempts, PTSD symptoms, depression, drug use, trauma exposure. Partial correlation network results showed suicidal ideation to be strongly related to depression, with smaller connections to past suicide attempts and anger. Additionally, past suicide attempts was strongly related to history of childhood trauma and weakly related to problematic drug use and PTSD symptoms. These results offer valuable information for both predicting suicide risk and differentiating targets for interventions lowering suicide risk in veterans.

Keywords: Network analysis; Risk factors; Suicide; Veterans.

Figure 1

Partial Correlation Network Note. Visually, positive relationships are depicted with blue lines, while negative relationships are shown in red. Additionally, the strength of the relationship correlates with the width and saturation of the line; thicker and darker lines indicate stronger relationships. Abbreviations are as follows: TLEQ-CA = Child Abuse Exposure; TLEQ-AIP = Adult Interpersonal Trauma Exposure; SCL-H = Anger; PSQI = Sleep Difficulties; MOS = Social Support; DTS = PTSD Symptoms; DAST = Problematic Drug Use; CES = Combat Exposure; BSS = Suicide Risk; BDI-II = Depressive Symptoms; AUDIT = Alcohol Use; Attempt = Past Suicide Attempts.

Figure 2

Centrality Plot Note. The plot above shows the centrality values for strength and expected influence. Individual nodes are listed on the y-axis, with their value of centrality on the x-axis. In this case, depressive and PTSD symptoms show the highest centrality across both measures.

Figure 3

Centrality Difference Note. This chart shows the difference in centrality for each node. Black boxes indicate a significant difference between nodes, while gray indicates a non-significant difference.

Figure 4

Edge stability Note. The plot above illustrates the stability of edge weights following bootstrapping. The red line shows the edge weights (seen on the x-axis) of the sample found from network analysis, with each horizontal line on the y-axis signifying one edge weight. Often, connected black dots are illustrated as well to show the edge weights found from bootstrapping. The 95% confidence intervals are displayed by the gray lines. When considering the stability of the edges, it is vital to compare the confidence intervals of edges to see if they truly vary from each other. When doing this, one should first examine how much confidence intervals overlap. Should they overlap greatly, this indicates that most edges likely do not vary from each other and thus results should be interpreted with care. If some confidence intervals do not overlap with each other, those are the edges that can be the most confidently interpreted (Epskamp et al., 2017). Ultimately, it is not unlikely that, regardless of adjustments and steps made toward securing edge stability, some edges can be interpreted confidently while other cannot.

Figure 5

Centrality Stability Note. This plot shows the stability of centrality by showing the change in centrality as participants are removed from the sample. A stable network should show little change as participants are removed.