Star­ing Down Death: Apply­ing Eye and Mouse Track­ing to Unpack Sui­cide-Rel­e­vant Cognition

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.