Jeremy Stewart, Queen’s University
@QuERBYLAB
By loading the video, you agree to YouTube’s privacy policy.
Learn more
Full Transcript:
Jeremy G. Stewart:
Good. Okay. So just a couple of housekeeping items before I get started. First off, the talk is going to show some of my stimuli, and so those stimuli show people who presumably or look like they’re making suicide attempts. So just as a heads up. The other thing I want to direct you to is the QR code here, which is going to provide access to recording of the talk, my slides, and I showed again on the final slide. So what I’m going to be tackling in this talk are the potential clinical applications of MouseView to better understanding, and assessing suicidal thoughts and behaviors. Suicide is a serious public health concern, of course, and it claims the lives of about 800,000 people worldwide every year. Specifically it’s the second leading cause of death among mid adolescents and young adults. Although suicide ideation, which is defined as thoughts of killing yourself, affects roughly one in five youth, only 20 to 33% of young people transition from ideation to suicide attempts.
Jeremy G. Stewart:
So identifying predictors of this key transition is critical for refining suicide risk assessment and for developing targeted interventions. But a lot of the putative correlates and predictors of suicide, especially in young people, they’re actually strongly related to suicide ideation, but they show much weaker associations with suicidal behaviors. So in light of this issue, some modern ideation to action theories of suicide, they provide separate explanations for this initial onset of suicide ideation, and then that shift from suicidal thinking to attempts. Although these theories actually differ in their specifics, this generic model I’m showing incorporates the elements that are shared among them. So generally the initial onset and escalation of suicidal thinking is thought to be driven by cognitive and affective variables, like hopelessness and psychological pain. More germane to this presentation though, these theories propose that suicidal actions, like planning and preparation, as well as making a suicide attempt, they require the capability to die by suicide.
Jeremy G. Stewart:
Suicide capability refers to one’s ability to overcome an innate and biological drive for self-preservation, in order to engage in potentially lethal self-directed injury. Capability has some practical aspects, like access to lethal means and knowledge about how to use them. It also includes characteristics that are hypothesized to be acquired through experience. My lab has focused on an acquired aspect of capability called Fearlessness About Death. In these ideation action theories they propose that exposure to risky, dangerous, and or potentially lethal experiences that basically habituates people to the innate fear that suicides should evoke. Then over time, people are thought to develop the fearlessness that’s necessary to undertake suicidal behavior. However, evidence for that direct link between Fearlessness About Death or FAD, and suicidal behavior is quite mixed. A recent review, for instance, reported that only half of the studies found this hypothesized association and the meta analytic effect size here with small. One reason for these inconsistent results might be the widespread practice of measuring capability with questionnaires that have poor psychometric properties.
Jeremy G. Stewart:
One issue is that most people can’t accurately estimate or even fathom the fear they would experience if they were faced by their own imminent death. That might undermine the validity of questionnaire measures of something like FAD. There are also several behavioral tasks that have been developed to capture capability. Just as one example, there’s a suicide version of the Stroop task, and it assumes that greater interference from suicide words on performance is some kind of evidence of attention being preferentially directed to a suicide content. People have speculated that that might indicate higher FAD, but reaction time tasks, like the Stroop actually don’t measure attention directly, of course, and they’re also prone to confounds. My colleagues and I have also recently found that interference on the Suicide Stroop has low internal consistency and poor test-retest reliability. So we’re very fortunate to team up with Tom Armstrong and his lab to improve the measurement of FAD by using eye-tracking.
Jeremy G. Stewart:
Dwell duration recorded by an eye tracker, as many of you might know, yields more differentiated responses to unpleasant stimuli than other behavioral or even psychophysiological measures. So here it might capture suicide specific responding with greater precision. Dwell is also a more direct measure of over of attention, and it has strong psychometric properties. So we used a similar, simple, passive viewing task that you saw on the prior talks, in which participants see pairs of emotional and neutral images, like the pair I have shown here. The emotional images included suicide related, threatening, pleasant, as well as disgusting stimuli, and there were five images representing each category. Each emotional image gets shown four times for 12 seconds in each trial. We record dwell duration, defined as how long a gaze is fixated on each image with an eye tracker.
Jeremy G. Stewart:
Our validation study recruited 140 students and we over sampled for prior suicidal behaviors. So the sample did include 28 people with a prior suicide attempt, and 10 additional students reported making an interrupted attempt. Along with several questionnaires they captured demographics and psychiatric symptoms. Excuse me. Participants completed the suicide ideation scale, which measures the intensity of thoughts of suicide in the past week. That measure actually has two separate sub-scales, on the one hand, there’s Suicidal Desire that reflects hopelessness and kind of a non-specific desire that life would end. The other sub-scale is called Resolved Plans and preparation, and that includes thoughts and actions that indicate greater readiness to make a suicide attempt. So turning to the results, this figure depicts gaze behavior on suicide image trials. The lines basically represent how much participants were fixating on the suicide images relative to neutral.
Jeremy G. Stewart:
So higher positive values indicate longer dwell time on the suicide image. The trial time is divided into 12 one second epochs on the X axis, and the separate lines show the patterns for each presentation of the image from first to fourth. So what you can see in the figure basically is that there’s a significant linear decrease in dwell on suicide images relative to neutral image across the epochs. What we’ve also found and reported in this paper is that gaze behavior towards suicide image seems to be distinct from the patterns of viewing we see for other emotional images. We also found that participants who rate the suicide images as subjectively more frightening and more disgusting, they tend to dwell less on suicide images compared to the neutral images over the course of trials.
Jeremy G. Stewart:
So that supports the possibility that our task captures cognitive and effective processes that are relevant to suicide capability or fearlessness about death or both. We next examine whether gaze behavior towards suicide images was associated with individual differences in recent suicide ideation severity. So in this figure, you can see scores on resolved plans and preparations on the Y axis, and on the X axis there’s a measure of overall gaze behavior that we use a lot called the Proportion of Dwell Time on Suicide Images, and this score ranges from zero to one. With equal viewing of the two images, the two categories of images yielding a score of 0.5, and that’s shown in the green dotted line here. Scores above 0.5 indicate longer viewing on the suicide images.
Jeremy G. Stewart:
So what the figure shows then is a small, but statistically significant association between dwell towards suicide images and resolved plans and preparations. Given this substantial skew, you can see depicted here in the outcome variable. We also fit a regression model with several demographic and clinical covariates, and we used a percentile bootstrapping approach. The effect remains statistically significant. Also dwell duration on the other emotional images wasn’t significantly related to resolved plans and preparations. We conducted the same analysis on suicidal desire, and that’s depicted here on the Y axis again. The proportion of dwell on suicide images, wasn’t by variantly associated with suicidal desire. The effect was also non-significant in the multivariate bootstrapped regression models.
Jeremy G. Stewart:
In fact, dwell behavior in general, regardless of the category of emotional image wasn’t associated with suicidal desire in our sample. Turning now to participants history of suicidal behavior. This figure has the proportion of dwell on suicide images this time on the Y axis, and the number of prior suicide attempts on the X axis. The boxes represent the distribution of dwell within each bin of lifetime attempts. So median and dwell toward suicide images, which is represented by the dark lines dissecting each box. You can see that it increases with the number of prior attempts. In fact, when you analyze that in a negative binomial regression model, as we did, greater dwell time on suicide images is associated with a greater rate of past lifetime attempts.
Jeremy G. Stewart:
Finally, we investigated suicidal intent among the 28 suicide attempts there in our sample. In line with what ideation to action frameworks would predict, greater dwell towards suicide images was associated with higher self-reported suicidal intent among the attempters. I’m suspending major inferences, of course, until this is replicated in a much larger sample. But again, we find that the effect for suicide images was not significant. Sorry. The effect on intent was not significant for any other type of emotional image we used. Okay. So to summarize, the participants in this study looked away from suicide images over time when they had the option to look elsewhere, at least on average. However, those who rated suicide images as less aversive dwelled on them longer.
Jeremy G. Stewart:
So this pattern of results suggests that our dwelled time measure might be capturing some important individual differences in something like fearlessness about death. Like other aversive or unpleasant stimuli, suicide images could initially capture attention, but ultimately evoke perceptual avoidance over time. Those with greater FAD, for instance, might engage in less ocular motor avoidance than people with lower FAD. Greater dwell on suicide, but not any other type of emotional image we used was associated with suicide planning and preparation with prior lifetime attempts, and with suicide intent among those with a history of suicidal behaviors. Returning to that ideation to action framework, dwell on suicide images was specifically associated with behaviors that pretend highly felony suicide attempts or dying by suicide.
Jeremy G. Stewart:
Ultimately, this type of measure could have immense clinical utility for guiding treatment and follow up decisions for suicide ideators in psychiatric settings. Of course, COVID-19 really forced our hands and necessitated an online solution for keeping, what I think is a promising line of research going. Fortunately, as we’ve seen today from both Alex and Sonya, there’s already exciting evidence that attention tracking in MouseView is strongly associated with dwell time measures recorded with eye tracking, using similar free viewing tasks as we did. So we’re currently running a study that’s hosted in Gorilla in which we present pairs of images, like the ones to the bottom left of the slide here, and we record gaze with MouseView.
Jeremy G. Stewart:
So far, we’ve recruited 170 participants. We have 69 with a history of suicidal thoughts and behaviors, and we expect to meet our target sample of 250 in the fall. So far, I wish I had data to report, I don’t, but the task is intuitive and it’s acceptable to most participants. The percentage of unusable trials is similar to our in-person study. Once the study is complete, we’re excited to test whether the most few version of the task shows the same pattern of associations with recent suicidal thoughts and lifetime suicidal behaviors, as we saw in our in-person study. If that indeed happens and we find the MouseView version of the task is sensitive to key indicators of suicide capability. There are important implications for just how we study suicide.
Jeremy G. Stewart:
Suicidal behaviors of course have low base rates in the general populations. So studies require very large sample sizes to conduct the complex and detailed research that we need to inform prevention. Because of this most research investigating behavioral and neural correlates of suicidal behaviors in particular, uses small clinical samples in which the rates of suicidal behaviors are much higher, or they enrich smaller samples by oversampling for suicidal behaviors, like we did. Limited statistical power in many studies likely contributes to a modest expected replication rate for studies on suicidal thoughts and behaviors. I have a former student named Brandon Lamb, and he’s working on a project that directly assesses the replicability of work in our field. So far he estimates that about two thirds of existing studies would replicate, and he also finds evidence for an inflated false discovery rate in these studies.
Jeremy G. Stewart:
I think MouseView provides an especially promising solution for this issue. So if we find evidence that this task is a valid measure of suicide cognition, the online delivery format is scalable so that it could be administered to nationally representative samples of thousands. That would yield critical information about the generalizability of our measure, and its utility for understanding and predicting suicide risk. We think, at least preliminarily that this is quite feasible because of the success that other research groups have had with administering online behavioral tasks to large samples. For example, Project Implicit, which is led by Bethany Teachman, Matt Nock, and several others. Actually administered a death suicide version of the Implicit Association Test to volunteers through their website. I think that we could validate our MouseView task in a very similar way. So I’ll leave it there. I’ll end by acknowledging funding sources for this work, my students, and of course my collaborators, including Tom. Thank you.
