Describing and Measuring the Pathway toSuicide Attempts: A Preliminary Study

ALEXANDER J. MILLNER, PHD, MICHAEL D. LEE, BS, AND MATTHEW K. NOCK, PHD

To die by suicide, one must think about suicide, make a plan, and thencarry it out. Prior research has examined the presence and predictors of theseoutcomes; however, virtually no studies have characterized how these stepsunfold along the pathway to suicide. A novel instrument was administered to 30recent suicide attempters. Results revealed that although the median onset forsuicidal ideation occurs 1 to 5 years prior to attempting, the median for 6 ofthe 10 steps measured was within 6 hours of attempting. Overall, 86.5% ofproximal planning steps took place within 1 week of attempting and 66.6%occurred within 12 hours.

Suicide is a leading cause of death aroundthe world (Lozano et al., 2012); however,some of the most basic characteristics of thepathway to suicidal behavior have not beenwell examined. Prior research has focusedprimarily on whether people have thoughtabout suicide, whether they have made asuicide plan, and whether they have made asuicide attempt (Nock, Borges, Bromet,Cha, et al., 2008). Although there havebeen qualitative accounts of the lead up tosuicide attempts and deaths (Kidd, 2004;Robins, 1981), there have been no detailed,descriptive, quantitative data characterizingthe pathway to suicide. For example,researchers have not delineated or describedwhat specific steps people take as they tran-sition from suicidal ideation to a suicideattempt or how people move down thispathway to suicide.

There are several important reasonsto describe the pathway to suicide. First,

describing how people move through thepathway to suicide lays a foundation forfuture research examining factors that influ-ence why people move from thinking toattempting suicide. For example, certainrisk factors could increase the likelihood ofsuicide, but only when people are at specificpoints along the pathway. If this were thecase, understanding the pathway would becritical to understanding how certain factorslead to suicide.

Second, clarifying this pathway couldinform suicide risk assessment and suicideprevention. For example, future researchmay find that certain steps along the path-way are associated with greater risk of afuture attempt. In addition, knowledge ofthe steps and average trajectories throughthe pathway could inform clinical decisionmaking.

Third, a description of the pathwaycould help advance the measurement of

ALEXANDER J. MILLNER, Department of Psy-chology, Harvard University, Cambridge, MA, USA;MICHAEL D. LEE, Department of Psychology, TheUniversity of Texas at Austin, Austin, TX, USA;MATTHEW K. NOCK, Department of Psychology,Harvard University, Cambridge, MA, USA.

Address correspondence to Alexander J.Millner, Department of Psychology, HarvardUniversity, William James Hall 1206, 33 KirklandSt., Cambridge, MA 02138; E-mail: amillner@fas.harvard.edu

Suicide and Life-Threatening Behavior 1© 2016 The American Association of SuicidologyDOI: 10.1111/sltb.12284

suicide planning, an important area ofresearch. Prior research has focused on twoelements of the pathway to suicide: generalmeasurement of suicide planning (e.g., thepresence of a suicide plan) and whether anattempt was impulsive or not by, for exam-ple, measuring the amount of time fromdecision to attempt or the extent of plan-ning prior to an attempt. Clearly, there isoverlap between the presence of a plan andwhether an attempt was impulsive—havinga suicide plan indicates that an attempt wasslower and more premeditated—and mea-sures used to assess planning are also usedto determine whether an attempt wasimpulsive or not (Nock et al., 2009). How-ever, a significant issue in this research areais the lack of a definition for a suicide plan andhow to determine whether an attempt wasimpulsive.When has a person crossed the linefrom not having a suicide plan to having one?Similarly, what are the specific aspects of asuicide attempt required to characterize it as“impulsive?” Currently, the field has notaddressed these questions. Outlining anddescribing how people move from thinking toattempting suicide might provide a basis forconsensus operationalization and consistentmeasurement that is necessary for reliableand valid research.

The most common way prior studieshave assessed the existence of a plan is to sim-ply ask whether one has ever made a “suicideplan,” without providing a definition of whatexactly constitutes a plan (Nock, Borges,Bromet, Alonso, et al., 2008). Prior studieshave utilized a wide range of measurementapproaches to assess the extent of suicideplanning prior to an attempt or whether anattempt should be characterized as impulsiveor not (see Anestis, Soberay, Gutierrez,Hern�andez, & Joiner, 2014, for a meta-analy-sis, and see Chalker, Comtois, & Kerbrat,2015, for a review). Several studies use theBeck Suicide Intent Scale–Planning subscale(SIS-Plan; Beck, Weissman, Lester, & Trex-ler, 1976; Beck, Schuyler, & Herman, 1974),which assesses the degree of premeditationand preparation prior to an attempt, butapproaches have varied. Prior studies have

included a different number of items to quan-tify “planning” (e.g., 2, 3, 7, or 8 items;Baca–Garcia et al., 2005; Conner et al.,2005; Weyrauch, Roy-Byrne, Katon, & Wil-son, 2001). Studies have also used myriadapproaches to distinguish between plannedand impulsive attempts. For example, studiesusing the SIS-Plan have had different cutoffrules (i.e., median split; Dombrovski et al.,2011), 75% quartile (Baca–Garcia et al.,2005), three equal-sized groups (Conneret al., 2005), or a particular score (Chenet al., 2007). Other studies have operational-ized an impulsive attempt as: (1) a decision toattempt suicide within 5 minutes of actuallyattempting (Simon et al., 2002; Williams,Davidson, & Montgomery, 1980), (2)30 minutes of “planning,” (Wojnar et al.,2008), (3) 3 hours of contemplation (Spokas,Wenzel, Brown, & Beck, 2012), (4) 7 days ofsuicide ideation (Conner et al., 2006), (5)denial of a lifetime suicidal plan withendorsement of a lifetime suicide attempt(Borges et al., 2006; Kessler, Borges, & Wal-ters, 1999; Nock, Borges, Bromet, Alonso,et al., 2008; Nock, Borges, Bromet, Cha,et al., 2008; Nock et al., 2009), and (6) a lackof plans, fluctuating ideation, and participantsendorsing an “impulsive” suicide attempt(Wyder & De Leo, 2007). In short, mostprior studies use methods that are crude,somewhat arbitrary (i.e., single-item timecutoffs), lack consensus, and fail to illustrateclearly how people transition down the path-way to suicide from thinking to acting (seeBagge, Littlefield, & Lee, 2013, for a recentstudy that collected nonarbitrary timing dataalong the pathway to suicide and argued for a3-hour cutoff between impulsive and nonim-pulsive attempts).

There are several problems with cur-rent measurement approaches for suicideplanning. First, the most critical problem isthe lack of conceptual operationalization ofwhat constitutes an “impulsive” or “planned”attempt, revealed by the varied methodolo-gies. For example, the myriad ways of mea-suring attempt planning with the SIS-Planshare little conceptual overlap with askingwhether the decision to attempt was within

2 THE PATHWAY TO SUICIDE ATTEMPTS

5 minutes of attempting. A recent studyfound modest correlations between differentsuicide planning assessments, leading theauthors to conclude that the approaches wereconceptually distinct (Chalker et al., 2015).The absence of a conceptual operationaliza-tion raises similar questions to those posedearlier, such as: At what point can it be saidthat a person has a plan rather than just tenta-tive suicidal thoughts? If a person comes upwith a “plan” 5 minutes prior to attempting,is that then a “planned suicide attempt” ornot? If not, how long prior to attempting isnecessary for an attempt to be considered“planned?” These questions remain unan-swered and are essential to address.

Second, the lack of conceptualizationcan lead to unreliable responses fromparticipants. Given that there is no consen-sus definition of a “suicide plan” amongresearchers, participants are unlikely to havereliable responses to questions such as,“Have you ever made a suicide plan?”Indeed, a recent study found who a substan-tial proportion of people who denied a “sui-cide plan” reported many of the sameplanning behaviors as those who endorsed aplan (Millner, Lee, & Nock, 2015). Similarly,in a prior study, more than a quarter ofattempters paradoxically described their sui-cide attempt as (1) “impulsive” but “planned”or (2) “not impulsive” but with no “planning”(Wyder & De Leo, 2007). Along the samelines, a recent study found that the amount ofpreparation prior to an attempt was unre-lated to the degree to which participantsdescribed their suicide attempt as motivatedby impulsiveness (May & Klonsky, 2016).Without a consensus approach to measuringsuicide planning, researchers will continue toobtain unreliable responses from partici-pants.

Third, the lack of conceptualizationproduces assessments that cannot be usedacross studies. For example, conducting amedian split on SIS-Plan data establishes acutoff point dividing planned from unplannedthat is unique to that particular study. In onecase, researchers used different items fromthe SIS-Plan to calculate two planning scores.

Although the two scores were highly corre-lated, only one resulted in a significant rela-tionship with a dependent variable of interest(Baca-Garcia et al., 2005). The lack of consis-tent measurement across studies underminesefforts to discover reliable differencesbetween risk factors of more unplanned ver-sus more planned attempts.

Fourth, there is no agreement regard-ing whether suicide planning should be mea-sured on a continuous scale or attemptsshould be dichotomized into “planned” ver-sus “unplanned.” When researchers chooseto measure the construct dichotomously, itassumes that “planned” and “unplanned”attempters come from separate populationson, for example, a particular risk factor. If thisis the case, then discovering the appropriateline to distinguish impulsive and plannedattempts is paramount to uncovering true dif-ferences between these populations.

Fifth, even though researchers havesought to uncover differences betweenthose who make planned versus unplannedattempts, no prior research has provided adescriptive account of the timing of plan-ning steps and the amount of preparationprior to attempts. This is similar to cri-tiques of psychological research in whichcritics have argued that psychological scien-tists jump straight to making inferences andtesting theories, while bypassing a processmost natural sciences start with to uncovera basic description of the phenomena ofinterest (Kagan, 2007; Tinbergen, 1963). Adescriptive account could provide data tohelp resolve several of the aforementionedissues. For example, if suicide planningappears to fall on a bimodal distribution,then dichotomization is suitable and it iden-tifies the appropriate dividing line. If not,then a continuous scale is more appropriate.Gaining a comprehensive description of thepathway to suicide is an essential first steptoward developing a common definition andmeasure to study suicide planning.

The goals of this study were todescribe how people move through thispathway by collecting data on the amountof time attempters spent between different

MILLNER ET AL. 3

planning steps prior to a suicide attempt.We developed a novel interview thatassessed the degree of preparation and thechronology of planning steps that weretaken prior to attempting suicide. Planningsteps were as follows: (1) thinking ofattempting suicide, (2) thinking of themethod, and (3) thinking of the place toattempt suicide. The interview distinguishedbetween the lifetime onset of a planningstep, onset for the current attempt, andtime taken to decide upon a step. We alsoinquired about whether the person (4) hadengaged in several preparatory actions (e.g.,obtaining a method, traveling to a place)prior to attempting. The main goal of thestudy was to provide a description of howpeople move through the pathway tosuicide and collect data on the timing ofspecific behaviors in an effort to overcomethe limitations of prior studies, such asthe use of relatively arbitrary measurementapproaches. However, collecting retrospec-tive data regarding the timing of particularevents over weeks, months, or years has itsown limitations. Particularly, we anticipatedthat participants would estimate timeusing rounded, prototypic values (e.g.,30 minutes, 1 hour, 1 day), and the incre-ment between these rounded values wouldincrease the longer that time had passedfrom the attempt (e.g., 5 minutes to15 minutes to 30 minutes to 1 hour; Hut-tenlocher, Hedges, & Bradburn, 1990).Therefore, a secondary goal of this studywas to closely examine and describe themanner in which people estimate the timingof planning steps to ensure appropriateanalyses and interpretation.

METHOD

Participants

Participants were 30 patients in apsychiatric inpatient unit. All participantsreported having made a suicide attempt in theprior 2 weeks. Additional inclusion criteriawere as follows: being over 18 years of age,

no current or past psychotic disorder or cog-nitive impairment, ability to read English,and correctly answer three questions regard-ing the informed consent form to ensurecomprehension of the nature of the study(100% of participants answered these ques-tions correctly). The Harvard University andPartners Healthcare System institutionalreview boards approved the study, and weobtained informed consent from all partici-pants. Participant demographic informationis presented in Table 1.

Procedure

Inpatient medical staff identifiedparticipants who had engaged in a suicideattempt within the past 2 weeks and met theadditional study inclusion/exclusion criteria.At the beginning of the study, partici-pants granted informed consent and com-pleted a prestudy risk assessment. Then, the

TABLE 1

Demographic Data

Participants (N = 30)

Age M (SD) 26.76 (8.93)Sex n (%)Female 18 (60.00)Male 12 (40.00)

Race n (%)Caucasian 25 (83.33)Mixed race 1 (3.33)Asian 3 (10.00)African American 1 (3.33)

SES M (SD) 3.40 (0.63)Number of participants with past suicidalbehaviors n (%)Aborted attempts(1 or more)

18 (60.00)

Interrupted attempts(1 or more)

7 (23.33)

Mean number of past suicidal behaviors M (SD)Age of onset ideation 17.38 (9.01)Aborted attempts 1.98 (3.83)Interrupted attempts 2.21 (2.38)Attemptsa 1.93 (1.33)

a

One participant who reported 25–50 sui-cide attempts was an outlier and removed fromthe mean number of attempts.

4 THE PATHWAY TO SUICIDE ATTEMPTS

experimenter (AJM) administered the Path-way to Suicidal Action Interview, which tookapproximately 1 hour. Participants com-pleted several other behavioral tasks and self-report questionnaires followed by a post-study risk assessment and, finally, weredebriefed and compensated. This study waspart of a larger study including additionalbehavioral tasks and self-report question-naires not reported here.

Measures

Pathway to Suicidal Action Interview(PSAI). The PSAI was developed to assessseveral different steps people take as theymove from thinking about suicide to actingon their thoughts. The PSAI consists of anintroduction interview and follow-up inter-views. The purpose of the introductioninterview was to collect a history of suicidalideation, nonsuicidal self-injury, and suici-dal actions and to determine which follow-up interview would be used (see below formore information on follow-up interviews).Specifically, the ideation section includes sixthoughts ranging from low-risk passiveideation (wishing to disappear or not existor one was never born) to higher risk pas-sive ideation (thinking life is not worth liv-ing or wanting to be dead) to activeideation (thinking that one should maybe killthemselves or thinking that one should killthemselves). The suicidal actions sectionassesses nonsuicidal self-injury, suicide ges-tures, aborted suicide attempts, interruptedsuicide attempts, and suicide attempts. Theinterview used instructional diagrams andspecific language to help participants cate-gorize their behavior. For an action to beconsidered a suicide attempt, the personhad to have engaged in a potentially lethalor harmful action with some intention ofdying. This definition is consistent withprior work in this area (O’Carroll et al.,1996; Posner et al., 2011; Silverman, Ber-man, Sanddal, O’Carroll, & Joiner, 2007).Although the question assessing prior suicideattempts included the phase “intent to die,”participants that endorsed a suicide attempt

were asked an additional question to ensureintent for the most recent attempt. The inter-viewer (AJM), a senior graduate student at thetime, designed the PSAI and was trained by asenior suicide researcher (MKN) to assesssuicidal behaviors.

Following the introduction interview,the interviewer administered a follow-upinterview. The goal of the follow-up inter-view was to assess the amount of timebetween planning steps and the suicidalaction. The follow-up interview began byasking the participant to narrate the eventsprior to the suicide attempt for the inter-viewer to understand the general timing ofevents. This was followed by more specificquestions including four pathway compo-nents: (1) thinking about suicide (i.e., idea-tion), (2) thinking of the suicide method, (3)thinking of the place to attempt suicide, and(4) carrying out preparatory actions. For idea-tion, method, and place we measured theamount of time elapsed between the suicideattempt and the (1) lifetime onset, (2) onsetfor the current attempt, and (3) settling on/being sure of the selected option (e.g., formethod, selecting which method to use; forideation, the decision to attempt suicide). SeeTable 2 for all planning steps.

For suicidal ideation, lifetime onsetwas determined by calculating the differ-ence in time between participants’ first life-time thought “I should kill myself” or“Maybe I should kill myself” and the mostrecent attempt. Ideation for the currentattempt was the amount of continuous sui-cidal ideation prior to the attempt. Ideationwas considered continuous if the participantexperienced ideation once within a week.Thus, to measure this step, we asked partic-ipants about their most recent 7-day periodwithout thoughts of suicide (i.e., when therewas a break in ideation) and then calculatedthe amount of time from following thispoint to the attempt.

We probed an additional phase ofideation not asked for in method or place:“mulling over” the decision to attempt sui-cide. Mulling was defined as “strongly con-sidering attempting suicide, perhaps going

MILLNER ET AL. 5

back and forth in one’s mind about whetherto do it or not.” Participants that endorsedmulling were asked how long they mulledover the decision to attempt suicide.Finally, participants were asked how longprior to the attempt they had made thedecision to attempt. To better operational-ize the decision, participants were asked if,at the decision point they specified, theywere (1) leaning toward attempting suicide,(2) pretty sure they were going to attemptsuicide, or (3) sure they would attempt. Ifparticipants reported being less than sure,they were asked to specify when they weresure they were going to attempt. Any differ-ence between the original decision pointand the point at which they were sure wasadded to time spent mulling. In sum, weassessed lifetime onset, onset for the currentattempt, mulling over the decision, and thedecision to attempt for the suicidal ideationcomponent.

For method and place, the experi-menter queried lifetime onset, onset for thecurrent attempt, and when the person “set-tled on” or was sure of the method and placeto attempt suicide. Settling on a method andplace were assessed to distinguish betweentenuous planning, where a person might beconsidering several options, and more con-crete plans, where the person has selectedhow and where to attempt suicide. In someinstances, a person could be sure of a methodor place even when undecided about whetherto attempt suicide or not (e.g., at a certainpoint, a person might have said, “I don’tknow if I’ll ever attempt suicide, but if I do I

know I’ll take pills”). We operationalized asuicide plan as the point at which a personhad thought of and settled on both themethod and the place to attempt suicide.

Finally, we inquired about whetherthe person had engaged in several prepara-tory actions prior to attempting, either tohelp carry out the suicide attempt (e.g.,obtain the method), because they would bedead soon (e.g., make a will), or otherpreparatory actions related to the attempt.

Analyses

Examining Participants’ Time Estimatesof Planning Steps. Other than the prepara-tory actions, all of the responses to thePSAI were free response estimates of thetime between various planning steps and asuicide attempt (we refer to these as esti-mates of relative or elapsed time). Weanticipated that as the elapsed timeincreased, people would use rounded esti-mates (e.g., “30 minutes” rather than“27 minutes 30 seconds”) with increasinglylarger units of time (i.e., weeks, months, oryears which constitute hundred thousandsor millions of seconds) and omit smallertime units (Huttenlocher et al., 1990). Iftrue, this pattern of increasingly larger unitsof time suggests that elapsed time might bereported exponentially. If planning stepswere accurately measured, however, elapsedtime would be linear.

To better understand this potentialbias, we converted all time estimates ofplanning steps to seconds, collapsed all

TABLE 2

Planning Steps Measured in PSAI

Pathway Component Steps

Suicidal ideation (1) Lifetime onset, (2) onset for current attempt,(3) mulling over decision, and (4) decision to attempt

Method (used in most recent attempt) (1) Lifetime onset, (2) onset for current attempt, and(3) being sure of/settling on method

Place (where most recent attempt took place) (1) Lifetime onset, (2) onset for current attempt, and(3) being sure of/settling on place

Preparatory actions Counted based on purpose of action

6 THE PATHWAY TO SUICIDE ATTEMPTS

planning steps across all subjects (i.e., a sin-gle vector containing time estimates for all288 steps across all subjects), sorted thesteps in ascending order, and ranked them(i.e., the shortest time estimate was ranked1, and the longest estimate was ranked288). If time estimates were accurate, plot-ting the time estimates over the rankingsshould reveal a somewhat smooth, diagonalline where, moving from the shortest tolongest intervals, relative time estimatesincrease fairly gradually. However, if timeestimates were exponential, the plots shouldreveal “jumps” near the right side of agraph, indicating a shift to increasingly lar-ger units of time as events move furtheraway from the attempt. To quantify thiscomparison, we fit linear and exponentialfunctions to the data where x was rankingsof the sorted time estimates and y was theamount of the time between each step andthe attempt. The exponential function wascalculated as:

y ¼ ae�bx þ c ð1Þ

where a is the y-intercept and indicates thevertical stretch of the curve, b stretches thecurve horizontally, the negative sign indi-cates that it is a vertical reflection of thefunction without the negative sign, and ccontrols the horizontal shift. The linearfunction was calculated as:

y ¼ aþ bx ð2Þ

where a is the y-intercept and b is the slopeof the line. We compared fits using Baye-sian information criterion (BIC; Raftery,1995), a commonly used model comparisonmetric which penalizes models for eachadditional parameter (Vrieze, 2012). Toexamine whether fit comparison varieddepending on the time interval, we comparedlinear and exponential functions within 11relative time intervals (e.g., within all plan-ning steps that occurred within 30 minutesof the attempt and omitting all steps thatoccurred after 30 minutes; see Figure 1D tosee all the intervals). Shorter time intervals

fit functions to fewer data, with the fewestbeing 45 steps that all occurred in under aminute.

Results of Examining Time Estimates

Figure 1A–C shows sorted relativetime estimates plotted over time with linearand exponential lines fit to the data for threeof the 11 time intervals examined. Asobserved in each of these three plots, timeestimates within 30 minutes, 1 day, and1 week all resemble exponential functions.Indeed, compared with a linear function, BICvalues revealed that an exponential functionprovided a superior fit at all 11 intervals(Figure 1D). In sum, regardless of where timeis cut off, plotting the ordered data revealspatterns similar to those in Figure 1A–C,which are best fit by exponential functions.

PSAI Scoring

The goal of this study was todescribe the pathway to suicide. To dothis, we converted all elapsed time esti-mates for planning steps to seconds andcreated histograms to illustrate the path-way. However, histograms can give consid-erably different depictions of the data,depending on the number and width of thebins (Scott, 1979; Wand, 1997). Weselected histogram bins based on two crite-ria: (1) approximate a log-transform toachieve linearity and (2) use prototypicalvalues (e.g., 1 day, 1 week) for easy inter-pretation. First, we sought to correct forthe pattern we found where participantsestimated time exponentially. Log-trans-forming exponential data will result in lin-earity (i.e., the logarithmic function is theinverse of the exponential function).Indeed, log-transforming and plotting allresponses over time shifted the data froman exponential to a linear trend (Figure 2A,B). Therefore, we selected bins that fol-lowed an approximately exponential pattern(Figure 2C) and placing the data in thesebins approximated a log-transformation,also resulting in a linear trend (Figure 2D).

MILLNER ET AL. 7

In addition to linearity, a log-transforma-tion also aligns with a conceptual or clini-cal intuition about the pathway to suicide:The further the back in time, the less rele-vant the distinctions of smaller time incre-ments become. For example, once timeintervals reach weeks, months, or yearsprior to a suicide attempt, distinctions ofminutes or seconds are no longer informa-tive or relevant.

One large difference between log-transforming the data and our selected binswas observed for time estimates under aminute. On a logarithmic scale, the differ-ence between 5 seconds and 40 seconds ismore than two units (e.g., notice how high1 minute is on the y-axis on Figure 2B).However, we placed all planning steps thatoccurred within a minute of the suicide

attempt in a single bin under the assump-tion that distinctions of seconds are difficultfor people to remember precisely (Schacter,1999) and of less significance than accordedby a log-transformation.

The second criterion for bin selectionwas interpretability. We chose prototypicaltime bins to increase the interpretability ofthe histogram and best illustrate the path-way to suicide. Providing a histogram withevenly distributed bins of log-transformeddata would have resulted in nonprototypicaltime values (e.g., 13 days and 22 hours) andhampered interpretability.

Preparatory Actions

We used the initial narrative descrip-tion of the events preceding the attempt

-

-10000

-8000

-6000

-4000

-2000

- 0

exponential fitlinear fit

1 5 30 1 1 6 1 31 all

-

1min min min hr day wk mo mos yr yrs

Bay

esia

n In

form

atio

n C

riter

ion

Rank Among All Steps Within 30 Min Ordered by Time

- - - - - - - - - -

BA

DC

Am

ount

of T

ime

Ste

p O

ccur

red

P

rior t

o S

uici

de A

ttem

ptA

mou

nt o

f Tim

e S

tep

Occ

urre

d

Prio

r to

Sui

cide

Atte

mpt

Time Intervals

20 min-

10 min-

30 min-

>1 min-

-

31

-

62

-

94

-

1

All Steps Within 30 Minutes

8 hrs-

16 hrs-

24 hrs-

~ mins-

-54

-

107

-

160

-

1

Am

ount

of T

ime

Ste

p O

ccur

red

P

rior t

o S

uici

de A

ttem

pt

All Steps Within 1 Day

1 wk-

4.6 days-

2.3 days-

-

62-

124

-

186

-

1

All Steps Within 1 Week

~ hrs-

Rank Among All Steps Within 1 Week Ordered by Time

Rank Among All Steps Within 1 Day Ordered by Time

Figure 1. Raw data fit to exponential and linear functions. Data containing the amount of time between theattempt and each planning step were collapsed across all steps and all subjects, sorted in ascending order by theamount of elapsed time and assigned a rank. (A–C) Plots of estimated relative time (y-axis) over rank (x-axis) quali-tatively show that, within different time intervals, an exponential function (green line) provides a superior fit for theraw data, compared with a linear function (blue line). (D) Bayesian information criterion (BIC) quantitatively con-firms this qualitative impression within 11 time intervals (smaller BIC values indicate better fit).

8 THE PATHWAY TO SUICIDE ATTEMPTS

and specific follow-up questions to deter-mine whether preparatory actions were car-ried out for the impending suicide attemptand not ancillary or distal actions. Forexample, if someone wrote a note 6 monthsprior to the attempt and at the time had nointention of killing themselves 6 monthslater, it was not counted. Preparatoryactions were tallied based on the number ofdistinct purposes of the actions, and not onthe number of actions themselves. Forexample, writing one note or five noteswere both counted as a single preparatorystep. Some purposes precluded others. Forexample, if a person traveled to avoid dis-covery, other steps taken to avoid discoverywould not be counted. A codebook with 27distinct purposes, developed in previouswork using an independent sample (Millneret al., 2015), was implemented by the first

author to code and sum the preparationdata.

Materials and Data Availability

The instrument (i.e., PSAI), data, andcomputer code written for analysis as wellas the supplemental materials and Jupyternotebook are available under the firstauthor’s account on the Open ScienceFramework (https://osf.io/4wcxk/). Thedata were analyzed using Python 2.7 pack-ages pandas (McKinney, 2010) for datamanagement; Matplotlib (Hunter, 2007)and Adobe Illustrator for plotting; NumPy(van der Walt, Colbert, & Varoquaux,2011) for math functions; and SciPy (Jones,Oliphant, & Peterson, 2001) and LMFIT(Newville, Stensitzki, Allen, & Ingargiola,2014) for curve fitting.

1 min-6 mo

1 5 10 15

1 yr

5 yrs-

- - - -

[

1 yr-

2 yr-

3 yr-

4 yr-

5 yr-

0 yr-

1 min

30 min3 hrs

1 day1 mo6 mo

1 yr5 yrs

-

-

-

-

--

--

-1 min

30 min

3 hrs

1 day

2 wks

1 yrs

>5 yrs

-

-

-

-

-

-

Bins

Rank Among All Steps Ordered by TimeRank Among All Steps Ordered by Time

Rank Among All Steps Ordered by Time

-

A B

C D

Am

ount

of T

ime

Ste

p O

ccur

red

P

rior t

o S

uici

de A

ttem

pt (l

og)

Am

ount

of T

ime

Ste

p O

ccur

red

P

rior t

o S

uici

de A

ttem

pt (b

inne

d)

Am

ount

of T

ime

Ste

p O

ccur

red

P

rior t

o S

uici

de A

ttem

ptA

mou

nt o

f Tim

e S

tep

Occ

urre

d

Prio

r to

Sui

cide

Atte

mpt

-

96

-

192

-

288

-

1

-

96

-

192

-

288-

1

-

96

-

192

-

288

-

1

Raw Data Log Transformed Data

Binned DataBins

Figure 2. Log-transforming and binning raw data to achieve linearity. (A) All raw data with estimated relative timeon the y-axis and rank on the x-axis show exponential pattern (see Figure 1). (B) After log-transform data are linear.(C) The histogram bins we selected to describe data approximate an exponential function. (D) Data fit to bins arealso approximately linear.

MILLNER ET AL. 9

RESULTS

A summary of methods used toattempt suicide and medical/physical conse-quences of the attempts is listed in Table 3.

What Does the Pathway to SuicideAttempt Look Like?

Plots illustrating the median timepoints for each step in the pathway to sui-cide attempt are presented in Figure 3. Inthe current sample, the median first onsetof suicidal ideation occurred 5 years beforethe attempt and first thoughts about thesuicide method used began 1 to 5 yearsprior to the attempt. A number of other

steps occurred much closer in time to theindex attempt. More specifically, the med-ian onset of continuous ideation began

TABLE 3

Summary of Methods Used and Medical/PhysicalConsequences of the Attempt

n Method

17 Overdose on medication/pills4 Overdose on medication/pills and alcohol3 Cut wrist2 Asphyxiation1 Asphyxiation and medication/pills1 Suffocation (e.g., gas)1 Cut wrists and medication/pills1 Medication/pills and suffocation

Method

Place

Ideation

Preparatory Actions

Lifetime

Onset for Current Attempt and Sure

Lifetime

Onset for CurrentAttempt Sure

Lifetime

Onset for CurrentAttempt Sure\DecisionMulling

5 M

in

30 M

in

2 H

rs

6 H

rs

1 W

k

2 W

ks

5 Y

rs

>5 Y

rs

Atte

mpt

Figure 3. Median of each pathway step. Themedian number of preparatory actions was 2. The timing of when prepara-tory actions occurred was not collected, and therefore, preparatory actions bar plot does not use the same x-axis as thetime estimates. Median lifetime ideation and lifetime thoughts about the method used both occurred years prior toattempting suicide. Median onset of ideation for the current attempt occurred 2 weeks prior to the attempt, and the life-time thoughts about where to attempt were 1 week before attempting. Median mulling over the decision started 6 hoursprior to attempting, whereas the median time for starting to think about the method, settling on the method, and think-ing of the place to attempt was all 2 hours prior to attempting. Median time for settling on the place was 30 minutesprior and median time for making the decision to attempt was only 5 minutes prior to attempting.

10 THE PATHWAY TO SUICIDE ATTEMPTS

2 weeks prior to the attempt, the onset ofthoughts about where to attempt suicidebegan 1 week prior to the attempt, theonset of mulling occurred 6 hours prior tothe attempt, and the onset of certaintyabout method occurred 2 hours prior to theattempt. The median time for certaintyabout the place to attempt was 30 minutesprior to the attempt, whereas the final deci-sion to attempt occurred 5 minutes beforethe attempt. See Figures S1 and S2 for aplot of each participant’s pathway to a sui-cide attempt along with a key to interpretthese plots.

Variability in the Timing of Steps in thePathway to Suicide Attempt

The median timing of each step pro-vides a measure of central tendency for the

timing of each step; however, we also wantedto examine individual differences in the tim-ing of each of these steps. The spread ofscores for each pathway step as well as a “sui-cide plan” is presented in Figure 4. Lifetimeonset of suicidal thoughts and method fre-quently occur more than 1 year prior to theattempt (see Figure 4). Once ideation for thecurrent attempt begins, the vast majority(86.5%) of the steps (across participants) iscarried out within a week of the attempt and66.6% of steps were carried out within12 hours of the attempt. For most distribu-tions, the modal time bin is less than1 minute. The most skewed distribution isthe decision point, in which 18 of 30 attemp-ters (60.0%) report deciding to attemptwithin 5 minutes and no one deciding morethan 3 days prior. For preparatory actions,most people carry out between zero and four

Method (Most Recent Attempt) Sure of Method

Place (Most Recent Attempt) Sure of Place Preparatory Actions (Most Recent Attempt)

Ideation (Most Recent Attempt) Mulling Over Decision Decision (Most Recent

Attempt)

Suicide Plan (Most Recent Attempt)

30 M

in

2 H

rs

6 H

rs

1 W

k

2 W

ks

5 Y

rs

>5 Y

rs

1 Y

r6

Mnt

h1

Mnt

h

3 D

ays

1 D

ays

12 H

rs

3 H

rs

1 H

r

15 M

in5

Min

<1 M

in

30 M

in

2 H

rs

6 H

rs

1 W

k

2 W

ks

5 Y

rs

>5 Y

rs

1 Y

r6

Mnt

h1

Mnt

h

3 D

ays

1 D

ays

12 H

rs

3 H

rs

1 H

r

15 M

in5

Min

<1 M

in

30 M

in

2 H

rs

6 H

rs

1 W

k

2 W

ks

5 Y

rs1

Yr

6 M

nth

1 M

nth

3 D

ays

1 D

ays

12 H

rs

3 H

rs

1 H

r

15 M

in5

Min

<1 M

in

30 M

in

2 H

rs

6 H

rs

1 W

k

2 W

ks

5 Y

rs

>5 Y

rs

1 Y

r6

Mnt

h1

Mnt

h

3 D

ays

1 D

ays

12 H

rs

3 H

rs

1 H

r

15 M

in5

Min

<1 M

in

30 M

in

2 H

rs

6 H

rs

1 W

k

2 W

ks

5 Y

rs

>5 Y

rs

1 Y

r6

Mnt

h1

Mnt

h

3 D

ays

1 D

ays

12 H

rs

3 H

rs

1 H

r

15 M

in5

Min

<1 M

in

30 M

in

2 H

rs

6 H

rs

1 W

k

2 W

ks

5 Y

rs

>5 Y

rs

1 Y

r6

Mnt

h1

Mnt

h

3 D

ays

1 D

ays

12 H

rs

3 H

rs

1 H

r

15 M

in5

Min

<1 M

in

30 M

in

2 H

rs

6 H

rs

1 W

k

2 W

ks

5 Y

rs

>5 Y

rs

1 Y

r6

Mnt

h1

Mnt

h

3 D

ays

1 D

ays

12 H

rs

3 H

rs

1 H

r

15 M

in5

Min

<1 M

in

30 M

in

2 H

rs

6 H

rs

1 W

k

2 W

ks

5 Y

rs

>5 Y

rs

1 Y

r6

Mnt

h1

Mnt

h

3 D

ays

1 D

ays

12 H

rs

3 H

rs

1 H

r

15 M

in5

Min

<1 M

in

1 2 3 4 5 6 7 8 0190 11

1

2

3

4

5

6

7

8

0

1

2

3

4

5

6

7

0

1

2

3

4

5

6

7

0

1

2

3

4

5

6

7

0

1

2

3

4

5

6

7

01

2

3

4

5

6

7

0

1

2

3

4

5

6

7

0

1

2

3

4

5

6

7

0

2

4

6

8

0

10

12

Figure 4. Histograms of the onset of pathway steps prior to the suicide attempt. Plan is defined as the point atwhich the person had thought of and settled on the method and place for the most recent attempt.

MILLNER ET AL. 11

actions, with a small group of participantscarrying several preparatory actions. For sui-cidal planning, although the modal responsewas that all four elements of a plan cametogether in under a minute, there was a largerange with some participants reporting hav-ing a plan as far as 3 days prior, althoughmost of the time the plan is formed within thesame day as the attempt.

Order of Pathway Steps

We also determined the sequence ofplanning steps (i.e., which step came firstand then second) regardless of the precisetiming. Even when looking at only fiveproximal steps (timing of ideation, method,place, mulling, and decision for the cur-rent attempt), we found that the pathwayto suicide is heterogeneous: There were 17unique sequences among the 30 attemp-ters. The most frequent sequence was tohave four items occur simultaneously:onset of ideation, thinking about method,thinking about place, and making the deci-sion to attempt suicide (16.7%). As wouldbe expected, participants generally thoughtof lifetime steps prior to thinking of stepsduring the current attempt and generallythought of current attempt steps beforesettling on method, place, and the decisionto attempt. About half of participants firstthought about the method before thinkingof the place and nearly half first thought(lifetime) about the method and place atthe same time. For nearly three-quartersof participants, thinking of the method forthe current attempt occurred at the sametime as settling on the method. This per-centage was similar for thinking of themethod for the current attempt and firstthinking of the place for the currentattempt or settling on the place toattempt. Thinking of the place for the cur-rent attempt and settling on the placeoccurred simultaneously for 80% of partic-ipants. See the Supporting Informations(Tables S1 and S2) for a table containingpercentages of bivariate chronology foreach the planning steps.

DISCUSSION

Data on how people transition fromthinking about suicide to attempting suicideare sparse, yet crucial for understanding riskfactors associated with why people movedown the pathway to suicide as well asbuilding a consensus approach to measuringsuicide planning. In this article wedescribed the degree of preparation and thetiming of steps people take as they movealong the pathway to suicide and providedan operationalization for “suicide planning.”This approach could establish a frameworkfor an empirical operationalization of animpulsive attempt, although we were unableto accomplish this in the current studygiven the small sample size.

There were three main findings inthis study. First, the onset of suicidal idea-tion and selection of a suicide method oftenoccur years prior to an attempt; however,the vast majority of proximal steps downthe pathway to suicide occur within a weekand most within 12 hours of the attempt.Second, the order of steps that people takein planning a suicide attempt is quiteheterogeneous. The most common patternwas that several of the steps occurredtogether. For instance, thinking of and set-tling on a method and place frequentlyoccurred simultaneously. Third, we foundthat, compared with a linear function, anexponential function best fit participants,estimates of the time between planningsteps and the suicide attempt looking within11 different time intervals. As planningsteps occurred further into the past fromthe attempt, participants used increasinglylarger units of time (e.g., seconds to min-utes to hours to days) and omitted smallerunits, which created biased data. If timewere estimated accurately, the data wouldbe fairly linear. Several aspects of the studywarrant further comment.

The first key finding was that the vastmajority of proximal pathway steps occurredwithin a week before the attempt, and mostwithin 12 hours. Furthermore, many stepsshowed a modal time of less than one minute

12 THE PATHWAY TO SUICIDE ATTEMPTS

prior to the attempt, with two of the quickeststeps being planning and making the decisionto attempt. In addition, several peoplereported an absence of suicidal ideation untiljust prior to attempting. Taken together,these data suggest that, although logisticallydifficult, research aimed at examining psy-chological processes occurring during thedays (or day, if possible) before a suicideattempt would provide enormous insightinto why people attempt suicide because thisappears to be the timescale in which themajority of planning and decisions are made(see Bagge, Glenn, & Lee, 2013; Bagge, Lee,et al., 2013; Bagge, Littlefield, Conner,Schumacher, & Lee, 2014, for a retrospectiveapproach).

The vast majority of participants inthis study experienced suicidal ideationyears prior to the attempt yet the transitiondown the pathway to an attempt, often wasprecipitous, particularly the decision toattempt. Given the presence of both long-and short-term processes, we recommendthat clinicians thoroughly assess lifetimesuicidal ideation and help patients with asuicide history create safety plans (Stanley& Brown, 2012) to prevent or slow downthe typically rapid progress down the path-way to suicide. The speed with which theparticipants in this study transitioned to asuicide attempt suggests that people maynot have access to their own degree of riskdays or perhaps even hours prior toattempting. Therefore, a crucial area ofclinical research is developing methods thatassist risk assessment but do not involve thepatients’ report (Nock et al., 2010).

The current results are consistentwith a prior study that assessed the timingof suicidal behaviors leading up to anattempt. Bagge, Glenn, and Lee (2013),Bagge, Lee, et al. (2013), and Bagge, Little-field, & Lee (2013) used a Timeline Fol-low-Back procedure to collect data on“contemplation,” “plan,” and “decision”within the 48 hours prior to a suicideattempt on a large sample (N > 200). Thepercentage of participants reporting theseindexed behaviors within 3 hours is

strikingly similar between the two studies:decision (current sample: 83%; Bagge et al.:85%), plan (current sample: 67%; Baggeet al.: 67%), and ideation (current sample:30%; Bagge et al.: 42%) (Bagge, Littlefield,et al., 2013).

The second key finding was that themost common order was for several steps tooccur simultaneously and steps involving amethod and place frequently occurred atthe same time. Although it makes concep-tual sense to characterize settling on amethod and place as a “suicide plan,” thefact that we found that these two stepsoften occur together provides additionalsupport for this designation. Finally, for allparticipants, the decision point representedthe last step of the pathway to an attemptand therefore represented the most proxi-mal step prior to the attempt within thissample.

A third finding was that when recount-ing the amount of time between planningsteps and an attempt, participants reportedelapsed time exponentially. We could find noprior study that reported this pattern,although most studies focus on accuracy ofveridical judgments of elapsed time (e.g.,comparing reports with diary entries;Friedman, 1993). The exponential patternrepresents a bias and introduces error intotime estimates, which has important implica-tions for descriptions, interpretations, andanalyses of these data. Depending on thequestion these data are addressing, we urgecaution in interpreting raw time estimates orincluding raw data in descriptive or inferen-tial analyses. For example, a log-transforma-tion or a specialized regression model (e.g.,Weibull) might be more appropriate thancanonical statistics (e.g., t test) with raw data.We specified histogram bins that approxi-mated a log-transform to achieve linearitywhile using prototypical values (e.g., 1 day,1 week) for easy interpretation. In addition tolinearity, transforming the data permits timeestimates ranging from less than a minute tomore than 5 years to be plotted on one graphand aligns with the intuition that smallerunits of time (e.g., seconds, minutes) are

MILLNER ET AL. 13

relevant for steps carried out close to theattempt but are less relevant for steps carriedout weeks, months, or years prior to theattempt.

The accuracy of the time estimatesparticipants provided is unclear. There isno consensus model of how people incorpo-rate time into autobiographical memories(Friedman, 1993), but prior research hasdocumented several characteristics thatinfluence the recall of the time elapsed sincean event. For example, accuracy is improvedwhen the event was more recent (Rubin &Baddeley, 1989), evoked a more extremeemotional reaction (Betz & Skowronski,1997), or when one can utilize generalknowledge from one’s life (Burt, 2008; e.g.,when I was in college). In the current study,most steps being estimated occurred withinweeks of the interview and involved anevent that likely was emotionally intense(i.e., involving suicide), and therefore, weassume estimates were fairly accurate. Formore distal aspects, such as the onset of sui-cidal ideation, time estimates are morelikely to be inaccurate. Prior research sug-gests that every week that passes since anevent, time estimates errors increase linearlyby about 1 day (Thompson, 1982) for atleast 6 years (Rubin, 1982).

Appropriately measuring preparatoryactions is challenging. We tallied prepara-tory actions based on the purpose of theaction. This system prevented us from pars-ing the precise number of actions, whichreduced reliability in pilot coding sessions(i.e., Should a “long” note be one actionbecause it was one note or more because itwas “long?”). The limitation of thisapproach is that scribbling a quick note andwriting a 10-page note are counted equally.We considered alternative approaches, suchas assessing the amount of time spent onthe preparatory acts, but all had more sig-nificant limitations (e.g., for obtaining amethod, when would the preparatory actionstart and end?). Another limitation of ourapproach is that we did not collect informa-tion about the amount of time that passedbetween preparatory actions and suicide

attempt to better characterize whenpreparatory actions occurred along thepathway. Future studies could collect this.

Given the small sample size, we couldnot produce a valid, data-driven method toaggregate the data into a composite score.We hypothesize that, with the appropriatesample size, the steps along the pathway tosuicide can be reduced into a single forma-tive composite, similar to socioeconomicstatus. If confirmed, this composite variablecould be valuable in future studies examin-ing an array of questions regarding suicide.For example, on the one hand, future stud-ies can examine whether this compositevariable falls on a unimodal or bimodal dis-tribution, which could be used to determinean empirical cutoff for “impulsive” versus“nonimpulsive” suicide attempt. On theother hand, large-scale studies or aggre-gated pathway data might reveal a continu-ous distribution. If this was the case, futureresearchers may split a sample into “impul-sive” and “nonimpulsive” based on an esti-mate of the population distribution that canbe used across studies rather than a splitwithin a given sample. Additionally, futurestudies can examine whether the pathwaycomposite interacts with certain risk factorsand suicide attempts. For example, somerisk factors may predict a more protractedpathway (e.g., internalizing disorder; Bagge,Littlefield, et al., 2013), while other factorsmay predict a rapid pathway resulting in asuicide attempt.

There are several important limita-tions to this study. First and foremost, thesample size is small. Studies with larger sam-ples need to confirm several of the findings.Second, as discussed, there is likely error inpeople’s reports of specific time estimatesdue to how people report time (omittingsmaller time units) and their memory for thetiming of events is imprecise (Schacter,1999). Third, although we interviewed peo-ple within 2 weeks of attempting suicide,they might have difficulty rememberingspecific details from the event or the timeleading up to it or have been motivated towithhold information because we were in an

14 THE PATHWAY TO SUICIDE ATTEMPTS

inpatient unit. Fourth, the participants in thisstudy were generally young (i.e., aged 18–35 years) and White, and therefore, thepathway steps should be tested across a morediverse sample. For example, in a prior studywith people over aged 50 years, older agewas associated with increased suicide plan-ning (Conner et al., 2007), although a differ-ent study with a large, representative samplefound no relationship between age andplanning (Hjelmeland et al., 2000). Althoughunclear, there is a chance these data overrep-resent the percentage of attempts that are

less planned compared with a sample thatincluded older adults. Fifth, although wehave provided a rationale for the timing bins,the choice of bins influenced the results andinterpretations. One advantage of the currentapproach is that other researchers can obtainthe raw time interval data collected in thecurrent study and use other bins or otheranalysis approaches. Despite these limita-tions, this preliminary study makes severaladvances over previous studies examiningsuicide planning.

REFERENCES

ANESTIS, M. D., SOBERAY, K. A., GUTIER-

REZ, P. M., HERN�ANDEZ, T. D., & JOINER, T. E.(2014). Reconsidering the link between impulsiv-ity and suicidal behavior. Personality and SocialPsychology Review, 18, 366–386.

BACA–GARCIA, E., DIAZ–SASTRE, C., GARC�IA

RESA, E., BLASCO, H., BRAQUEHAIS CONESA, D.,OQUENDO, M. A., ET AL. (2005). Suicide attemptsand impulsivity. European Archives of Psychiatry andClinical Neuroscience, 255, 152–156.

BAGGE, C. L., GLENN, C. R., & LEE, H.-J. (2013). Quantifying the impact of recent nega-tive life events on suicide attempts. Journal ofAbnormal Psychology, 122, 359–368.

BAGGE, C. L., LEE, H.-J., SCHUMACHER, J.A., GRATZ, K. L., KRULL, J. L., & HOLLOMAN,G. (2013). Alcohol as an acute risk factor forrecent suicide attempts: A case-crossover analy-sis. Journal of Studies on Alcohol and Drugs, 74,552–558.

BAGGE, C. L., LITTLEFIELD, A. K., CON-

NER, K. R., SCHUMACHER, J. A., & LEE, H.-J.(2014). Near-term predictors of the intensity ofsuicidal ideation: An examination of the 24 hprior to a recent suicide attempt. Journal of Affec-tive Disorders, 165, 53–58.

BAGGE, C. L., LITTLEFIELD, A. K., & LEE,H.-J. (2013). Correlates of proximal premedita-tion among recently hospitalized suicide attemp-ters. Journal of Affective Disorders, 150, 559–564.

BECK, A. T., SCHUYLER, D., & HERMAN, I.(1974). Development of suicidal intent scales. InA. T. Beck, H. L. P. Resnik and J. Lettieri(Eds.), The prediction of suicide (pp. 45–56). Bowie,MD: Charles Press.

BECK, A. T., WEISSMAN, A., LESTER, D.,& TREXLER, L. (1976). Classification of suicidalbehaviors: II. Dimensions of suicidal intent.Archives of General Psychiatry, 33, 835–837.

BETZ, A. L., & SKOWRONSKI, J. J. (1997).Self-events and other-events: Temporal datingand event memory. Memory & Cognition, 25,701–714.

BORGES, G., ANGST, J., NOCK, M. K., RUS-

CIO, A. M., WALTERS, E. E., & KESSLER, R. C.(2006). A risk index for 12-month suicide attemptsin the National Comorbidity Survey Replication(NCS-R). Psychological Medicine, 36, 1747.

BURT, C. D. B. (2008). Time, language,and autobiographical memory. Language Learn-ing, 58, 123–141.

CHALKER, S. A., COMTOIS, K. A., & KER-

BRAT, A. H. (2015). Impulsivity and suicidalbehavior: How you define it matters. Interna-tional Journal of Cognitive Therapy, 8, 172–192.

CHEN, E. Y. H., CHAN, W. S. C., CHAN,S. S. M., LIU, K. Y., CHAN, C. L. W., WONG, P.W. C., ET AL. (2007). A cluster analysis of thecircumstances of death in suicides in HongKong. Suicide and Life-Threatening Behavior, 37,576–584.

CONNER, K. R., DUBERSTEIN, P. R., BECK-

MAN, A., HEISEL, M. J., HIRSCH, J. K., GAMBLE,S., ET AL. (2007). Planning of suicide attemptsamong depressed inpatients ages 50 and over.Journal of Affective Disorders, 97, 123–128.

CONNER, K. R., HESSELBROCK, V. M.,SCHUCKIT, M. A., HIRSCH, J. K., KNOX, K. L.,MELDRUM, S., ET AL. (2006). Precontemplatedand impulsive suicide attempts among individualswith alcohol dependence. Journal of Studies onAlcohol and Drugs, 67, 95.

CONNER, K. R., PHILLIPS, M. R., MEL-

DRUM, S., KNOX, K. L., ZHANG, Y., & YANG, G.(2005). Low-planned suicides in China. Psycholog-ical Medicine, 35, 1197–1204.

DOMBROVSKI, A. Y., SZANTO, K., SIEGLE, G.J., WALLACE, M. L., FORMAN, S. D., SAHAKIAN, B.,

MILLNER ET AL. 15

ET AL. (2011). Lethal forethought: Delayed rewarddiscounting differentiates high- and low-lethalitysuicide attempts in old age. Biological Psychiatry, 70,138–144.

FRIEDMAN, W. J. (1993). Memory for thetime of past events. Psychological Bulletin, 113,44–66.

HJELMELAND, H., NORDVIK, H., BILLE-BRAHE, U., DE LEO, D., KERKHOF, J. F., L €ONNQ-

VIST, J., ET AL. (2000). A cross-cultural study ofsuicide intent in parasuicide patients. Suicide andLife-Threatening Behavior, 30, 295–303.

HUNTER, J. D. (2007). Matplotlib: A 2Dgraphics environment. Computing in Science andEngineering, 9, 90–95.

HUTTENLOCHER, J., HEDGES, L. V., &BRADBURN, N. M. (1990). Reports of elapsedtime: Bounding and rounding processes in esti-mation. Journal of Experimental Psychology: Learn-ing, Memory, and Cognition, 16, 196–213.

JONES, E., OLIPHANT, E., & PETERSON, P.(2001). SciPy: Open source scientific tools for python.Retrieved August 15, 2015, from http://www.scipy.org/.

KAGAN, J. (2007). A trio of concerns. Per-spectives on Psychological Science, 2, 361–376.

KESSLER, R. C., BORGES, G., & WALTERS,E. E. (1999). Prevalence of and risk factors forlifetime suicide attempts in the National Comor-bidity Survey. Archives of General Psychiatry, 56,617–626.

KIDD, S. A. (2004). “The walls were clos-ing in, and we were trapped”: A qualitative anal-ysis of street youth suicide. Youth & Society, 36,30–55.

LOZANO, R., NAGHAVI, M., FOREMAN, K.,LIM, S., SHIBUYA, K., ABOYANS, V., ET AL. (2012).Global and regional mortality from 235 causesof death for 20 age groups in 1990 and 2010: Asystematic analysis for the Global Burden of Dis-ease Study 2010. The Lancet, 380, 2095–2128.

MAY, A. M., & KLONSKY, E. D. (2016).“Impulsive” suicide attempts: What do we reallymean? Personality Disorders, 7, 293–302. doi:10.1037/per0000160.

MCKINNEY, W. (2010). Data structuresfor statistical computing in Python. In Proceedingsof the 9th Python in Science Conference (pp. 51–56).Available at http://conference.scipy.org/proceed-ings/scipy2010/pdfs/mckinney.pdf

MILLNER, A. J., LEE, M. D., & NOCK, M.K. (2015). Single-item measurement of suicidalbehaviors: Validity and consequences of misclas-sification. PLoS One, 10, e0141606.

NEWVILLE, M., STENSITZKI, T., ALLEN, D.B. & INGARGIOLA, A., (2014). LMFIT: Non-linearleast-square minimization and curve-fitting forPython. Available at https://zenodo.org/record/11813

NOCK, M. K., BORGES, G., BROMET, E. J.,ALONSO, J., ANGERMEYER, M., BEAUTRAIS, A.,ET AL. (2008). Cross-national prevalence and riskfactors for suicidal ideation, plans and attempts.The British Journal of Psychiatry, 192, 98–105.

NOCK, M. K., BORGES, G., BROMET, E. J.,CHA, C. B., KESSLER, R. C., & LEE, S. (2008).Suicide and suicidal behavior. EpidemiologicReviews, 30,133–154.

NOCK, M. K., HWANG, I., SAMPSON, N.,KESSLER, R. C., ANGERMEYER, M., BEAUTRAIS, A.,ET AL. (2009). Cross-national analysis of theassociations among mental disorders and suicidalbehavior: Findings from the WHO World Men-tal Health Surveys. PLoS Medicine, 6, e1000123.

NOCK, M. K., PARK, J. M., FINN, C. T.,DELIBERTO, T. L., DOUR, H. J., & BANAJI, M. R.(2010). Measuring the suicidal mind: Implicitcognition predicts suicidal behavior. PsychologicalScience, 21, 511–517.

O’CARROLL, P. W., BERMAN, A. L., MARIS,R. W., MOSCICKI, E. K., TANNEY, B. L., & SIL-VERMAN, M. M. (1996). Beyond the tower ofBabel: A nomenclature for suicidology. Suicideand Life-Threatening Behavior, 26, 237–252.

POSNER, K., BROWN, G. K., STANLEY, B.,BRENT, D. A., YERSHOVA, K. V., OQUENDO, M.A., ET AL. (2011). The Columbia–Suicide Sever-ity Rating Scale: Initial validity and internal con-sistency findings from three multisite studieswith adolescents and adults. American Journal ofPsychiatry, 168, 1266–1277.

RAFTERY, A. E. (1995). Bayesian modelselection in social research. Sociological Methodol-ogy, 25, 111–163.

ROBINS, E. (1981). The final months: Astudy of the lives of 134 persons who committed sui-cide. New York: Oxford University Press.

RUBIN, D. C. (1982). On the retentionfunction for autobiographical memory. Journal ofVerbal Learning and Verbal Behavior, 21, 21–38.

RUBIN, D. C., & BADDELEY, A. D. (1989).Telescoping is not time compression: A model.Memory & Cognition, 17, 653–661.

SCHACTER, D. L. (1999). The seven sinsof memory: Insights from psychology and cogni-tive neuroscience. American Psychologist, 54(3),182.

SCOTT, D. W. (1979). On optimal anddata-based histograms. Biometrika, 66, 605–610.

SILVERMAN, M. M., BERMAN, A. L., SAND-

DAL, N. D., O’CARROLL, P. W., & JOINER, T. E.(2007). Rebuilding the tower of Babel: A revisednomenclature for the study of suicide and suici-dal behaviors part 2: Suicide-related ideations,communications, and behaviors. Suicide and Life-Threatening Behavior, 37, 264–277.

SIMON, T. R., SWANN, A. C., POWELL, K.E., POTTER, L. B., KRESNOW, M., & O’CARROLL,

16 THE PATHWAY TO SUICIDE ATTEMPTS

P. W. (2002). Characteristics of impulsive sui-cide attempts and attempters. Suicide and Life-Threatening Behavior, 32(Suppl. 1), 49–59.

SPOKAS, M., WENZEL, A., BROWN, G. K.,& BECK, A. T. (2012). Characteristics of individ-uals who make impulsive suicide attempts. Jour-nal of Affective Disorders, 136, 1121–1125.

STANLEY, B., & BROWN, G. K. (2012).Safety planning intervention: A brief interventionto mitigate suicide risk. Cognitive and BehavioralPractice, 19, 256–264.

THOMPSON, C. P. (1982). Memory forunique personal events: The roommate study.Memory & Cognition, 10, 324–332.

TINBERGEN, N. (1963). On aims andmethods of ethology. Zeitschrift F€ur Tierpsycholo-gie, 20, 410–433.

VRIEZE, S. I. (2012). Model selection andpsychological theory: A discussion of the differ-ences between the Akaike information criterion(AIC) and the Bayesian information criterion(BIC). Psychological Methods, 17, 228–243.

VAN DER WALT, S., COLBERT, S. C., &VAROQUAUX, G. (2011). The NumPy array: Astructure for efficient numerical computation.Computing in Science and Engineering, 13, 22–30.

WAND, M. P. (1997). Data-based choiceof histogram bin width. The American Statistician,51, 59–64.

WEYRAUCH, K. F., ROY-BYRNE, P., KATON,W., & WILSON, L. (2001). Stressful life eventsand impulsiveness in failed suicide. Suicide andLife-Threatening Behavior, 31, 311–319.

WILLIAMS, C. L., DAVIDSON, J. A., &MONTGOMERY, I. (1980). Impulsive suicidal behav-ior. Journal of Clinical Psychology, 36, 90–94.

WOJNAR, M., ILGEN, M. A., JAKUBCZYK, A.,WNOROWSKA, A., KLIMKIEWICZ, A., & BROWER, K.

J. (2008). Impulsive suicide attempts predict post-treatment relapse in alcohol-dependent patients.Drug and Alcohol Dependence, 97, 268–275.

WYDER, M., & DE LEO, D. (2007).Behind impulsive suicide attempts: Indicationsfrom a community study. Journal of AffectiveDisorders, 104, 167–173.

Manuscript Received: November 20, 2015Revision Accepted: May 9, 2016

SUPPORTING INFORMATION

Additional Supporting Information may befound in the online version of this article:Figure S1. Orientation to pathway to suicideplots (A) Components of the pathway to sui-cide: suicide method = red, place = blue,ideation = green, preparatory actions = or-ange. (B) Method, place and ideation/deci-sion are on a timescale.Figure S2. Individual pathway to suicideplots for every participant (n = 30).Table S1. Percentages of subjects thatthought of one planning step at the sametime as another step.Table S2. Percentages of subjects thatthought of one planning step prior to or afteranother step.

MILLNER ET AL. 17