The Path Not Taken: Distinguishing Individuals Who Die by Suicide from Those Who Die by Natural Causes Despite a Shared History of Suicide Attempt

Abstract

Objectives:

This study aimed to identify variables that distinguish suicide risk among individuals with previous suicide attempts.

Method:

Using psychological autopsy procedures, we evaluated 86 decedents who had at least one lifetime suicide attempt prior to eventual death by suicide (n = 65) or natural causes (n = 21).

Results:

The Suicide Death group was more likely to be male, to have alcohol in the toxicology report at time of death, and to have a depression diagnosis, while the Natural Cause Death group was more likely to have personality disorder traits, a polysubstance use disorder, higher reported health stress, and an antidepressant in the toxicology report at time of death. Hopelessness and ambivalence were found to distinguish between groups during the 6 months prior to death.

Conclusions:

These findings suggest important differences between individuals with a shared history of a suicide attempt who die by suicide versus natural causes.

Introduction

Suicide is the 10th leading cause of death in the United States, resulting in more than 47,000 deaths per year, an increase of 33% from 1999 to 2018 (Stone, 2021). Even with a slight decrease in total suicide deaths in the United States in 2019 and provisionally again in 2020 (Ahmad & Anderson, 2021), suicide remains a threat to public health. In 2019, approximately 12 million adults reported serious thoughts of suicide, of which 3.5 million reported making suicide plans and 1.4 million reported attempting suicide (CDC, 2019; SAMHSA, 2020). The high rates of both attempted suicide and deaths by suicide demand the mental health profession to address that our current methods of assessing these issues are not enough, and emphasize the need to develop new ways of identifying and treating individuals at risk of engaging in suicidal behaviors. Life expectancy has been shown to decrease following a suicide attempt by up to 18 years for men and 11 years for women (Jokinen et al., 2018). Given the importance of a previous suicide attempt in predicting future suicide attempts (Franklin et al., 2017), the lack of data regarding previous suicide attempters is a major gap in understanding suicidality. When considering the emotional and behavioral precursors that lead to a suicide attempt or suicide completion, perhaps there is a tipping point in suffering that remains to be identified and targeted with intervention.

Suicide risk can be seen through passive thoughts of death and dying, active thoughts of how to kill oneself, and actions taken to bring about one’s own death (May & Victor, 2018). There do not seem to be many significant risk factors that distinguish those with suicidal ideation from those who engage in suicidal actions (May & Klonsky, 2016). The pathway from one stage to the next is not well understood, and there is some disagreement about when an increased risk becomes worthy of intervention. Prior studies have shown that psychiatric disorders, high stress, substance use, low social support, and being male are all risk factors for suicidal behavior (Kleiman & Liu, 2013; Ritchie et al., 2015; Wang et al., 2012). History of a previous suicide attempt is a significant risk factor for both a future suicide attempt and eventual death by suicide (Brown et al., 2000; Franklin et al., 2017). Additionally, having multiple suicide attempts is related to a diagnosis of three or more mental disorders (Pagura et al., 2008). In patients presenting to an emergency room after a suicide attempt, having multiple suicide attempts had a positive association with previous psychiatric outpatient care, previous psychiatric hospitalization, use of psychiatric medication, higher levels of depression, greater hopelessness, psychiatric comorbidity, substance abuse, psychotic disorders, borderline personality disorder, unemployment, poor relational skills, and poorer global functioning (Forman et al., 2004; Perquier et al., 2017). The profile of individuals with multiple suicide attempts is complex. Understanding why some individuals engage in suicidal action on multiple occasions may help illuminate the distinction between suicidal ideation and action. Having a known history of a previous suicide attempt does not help to distinguish levels of risk among all individuals who share this common risk factor; not everyone who attempts suicide once goes on to attempt again. Having more data regarding those who have multiple suicide attempts would allow for improved efforts for both treatment and prevention.

The risk identified by a previous suicide attempt might increase with the number of previous attempts. Having more than two suicide attempts was associated with greater risk of eventual suicide compared to other causes of death in patients who had a previous suicide attempt (Nordentoft et al., 1993). Younger age at first suicide attempt may strengthen the cumulative effect of multiple attempts, and may relate to the cognitive patterns of hopelessness that become ingrained earlier in development. Individuals with a higher number of previous hospitalizations experienced a significantly lower chance of depression remission compared to individuals with fewer hospitalizations in a sample of 1,014 depressed inpatients (Riedel et al., 2011). Individuals who attempt suicide earlier in life may be more likely to feel hopeless during times of stress, thus viewing suicide as a more viable option for relief from the conflict between wanting to live while also wanting to die.

Ambivalence involves holding contradictory beliefs or opinions simultaneously. In the context of suicide, ambivalence refers to wanting to die while simultaneously wanting to live. Suicidal individuals rarely experience a desire to die with no desire to live. In a study of 106 individuals hospitalized following a suicide attempt, 50% provided ambivalent answers on questions assessing a wish to die and a wish to live (Kovacs & Beck, 1977). Additionally, an ambivalence score was shown to be a unique predictor of risk for eventual suicide in a group of 5,814 psychiatric outpatients followed for 20 years (Brown et al., 2005). Experiencing ambivalence when suicidal may be a risk factor for taking suicidal action. Individuals identified as having ambivalence about living and dying reported a significantly higher level of suicide risk than those with a low wish to die (O’Connor et al., 2012). However, assessing wish to die as a singular construct does not fully capture ambivalence. In another study, individuals identified as having a wish to die were not significantly different in their level of suicidal ideation than those individuals with ambivalence about living and dying, while both the wish to die group and the ambivalent group had significantly higher levels of suicidal ideation than the wish to live group (Lento et al., 2013). Understanding and measuring feelings of ambivalence experienced by suicidal individuals could be a productive way to identify risk above and beyond passive versus active suicidal ideation. Ambivalence may be a crucial step on the path to suicide that could identify risk and mitigate potential for multiple suicide attempts.

Hopelessness about the self, the world, and the future may drive individuals who are depressed from suicidal ideation to suicidal action. Higher scores on the Beck Hopelessness Scale (BHS) were shown to be predictive of eventual death by suicide, with a cutoff score of 9 or above identifying 94.1% of the sample who died by suicide (Beck et al., 1990). The pervasiveness of these negative cognitions likely results in depressed individuals feeling torn between wanting to live and wanting to die. Hopelessness may reinforce the idea that suffering will never improve, thus tipping the scales of ambivalence in favor of a wish to die. In a study that followed 142 depressed patients over a period of 10 years, results supported that hopelessness predicts suicidal ideation, but no effect was found for hopelessness helping to distinguish suicidal ideation from suicide attempt (Qiu et al., 2017). Hopelessness appears to fluctuate over time, and a higher level of hopelessness during one depressive episode may predict higher levels of hopelessness during later depressive episodes (Beck et al., 1990). Ambivalence between deciding to die and deciding to live also fluctuates moment-to-moment for those experiencing suicidal ideation (Bergmans et al., 2017). If hopelessness impacts the ambivalence between wanting to live and wanting to die, these two constructs and the fluctuations they exhibit may be connected.

One study of 1,959 individuals presenting to a general hospital followed participants for up to nine years and nine months (Hawton & Fagg, 1988). Factors that distinguished those who died by probable suicide from those who died by natural causes included recent psychiatric hospitalization, persistent psychiatric disorder, behavioral retardation, and poor physical health (Hawton & Fagg, 1988). Another 20-year prospective study followed 6,891 psychiatric outpatients and compared 49 individuals who died by suicide to 170 individuals who died by natural causes (Brown et al., 2000). Significant risk factors for eventual death by suicide included Major Depressive Disorder, Bipolar Disorder, unemployment, hopelessness, current suicidal ideation, previous suicide attempts, previous psychiatric hospitalization, and increasing age (Brown et al., 2000). Once again, a complex psychiatric profile emerges in individuals who die by suicide, but there is no further information regarding the distinction between death by suicide and death by natural causes among individuals with a history of suicide attempt.

Psychological autopsy procedures were originally developed specifically to clarify cause of death during cases of questionable suicide, and are now commonly used in studies investigating suicide death (Litman et al., 1963). The present study used data collection procedures in line with psychological autopsy recommendations (Ebert, 1987). Psychological autopsy has been shown to be an effective method for thoroughly assessing an individual’s state of mind and life circumstances post-mortem in order to investigate suicidal deaths (Weinberger et al., 2018). Psychological autopsy data collection has been shown to be an accurate method for data collection concerning the deceased. In a study assessing suicide attempters and interrater reliability on diagnostic interview, agreement among interviewers had a Kappa = 1.00 with 100% agreement for diagnosis of Major Depressive Disorder (DeJong & Overholser, 2009). Psychological autopsy data uniquely permitted distinguishing individuals who died by natural causes from those who died by suicide, allowing retrospective analysis of risk with a known outcome for cause of death.

The present study addressed factors of hopelessness and ambivalence in individuals who all shared a history of a previous suicide attempt, but who did not all die by suicide. We compared groups on standard demographic and psychological factors, hopelessness, wish to live, wish to die, and factors surrounding first lifetime suicide attempt. Previously suicidal individuals who later died by suicide were expected to be more likely to be male, have a greater number of psychiatric diagnoses, report higher stress, show ante-mortem presence of alcohol and drugs, and possess fewer social-protective factors at time of death. Individuals who died by suicide were expected to have greater severity of hopelessness and ambivalence, be younger at first attempt, have less time between first attempt and eventual death, have greater lethality of first attempt, have received less care after the attempt, and have had a higher number of previous attempts than those who died by natural causes.

Material and Methods

Participants

The present study evaluated 86 adults with a past suicide attempt who died in Cuyahoga County, Ohio, during the years 1989 to 2017 using data collected through an IRB-approved psychological autopsy study (Overholser et al., 2012). Cause of death was determined by the Medical Examiner and the sample was divided into (1) Death by Suicide group (n = 65) and (2) Death by Natural Causes group (n = 21). Causes of natural death included cardiac issues (coronary disease, cardiomyopathy, aortic aneurism, etc.), organ failure, or pulmonary issues. These cases were included in the Natural Cause Death group due to Medical Examiner and consensus agreement that the deaths were not suicidal in nature. Cases were excluded if: (1) cause of death was due to murder or involvement in a murder-suicide or (2) age at time of death was less than 18, or (3) quality of the tissue at autopsy precluded molecular or genomic studies. Seven cases were eliminated due to cause of death being an accidental overdose, as these deaths are unable to be classified as fully suicidal or natural in nature.

Measures

The Structured Clinical Interview for DSM-IV (SCID; First et al., 1995) and the Structured Interview for DSM-IV Personality Disorders (SIDP-IV; Pfohl et al., 1997) were administered to next-of-kin. The SIDP-IV is commonly used to evaluate personality functioning across research and clinical settings (Zimmerman et al., 2008). For an earlier subset of participants, the Structured Clinical Interview for the DSM-III-R (SIDP-R; Pfohl et al., 1989) was used, because the SIDP-IV was not yet developed. Although normally administered directly to a patient, in the current study the SCID and SIDP-IV were adapted for use in psychological autopsy. All of the items were adjusted to be third-person such that they could be applied to assess functioning of the deceased individual. Thus, the diagnoses described throughout the present study can be considered post-mortem diagnoses. Agreement among personal reports and informant interviews when using the SCID and SIDP-IV has been shown to be reliable and valid (Schneider et al., 2004) with Kappa = 0.68 (95% CI [0.35 – 1.0]) with 89% agreement for any Axis-I disorder diagnosis and Kappa = 0.92 (95% CI [0.95 – 1.0]) with 97% agreement for any Axis-II disorder diagnosis.

Stressful life events experienced within the 6 months prior to death were measured using the Modified Life Experiences Scale (MLES). Adapted from the original Life Experiences Survey (Sarason et al., 1978), the MLES is a 10-item measure of potentially stressful life events (bereavement, finances, occupation, personal health, family health, personal hospitalization, family hospitalization, relationships, legal issues, and other). The test-retest reliability of the original LES has a reliability coefficient of .64 over a five- to six-week interval (Sarason et al., 1978). The MLES has been used successfully in research on college students (Fisher & Overholser, 2014), psychiatric outpatients with and without suicidality (Fisher et al., 2015), and in psychological autopsy research on suicide decedents (Overholser et al., 2012). MLES data was collected from next-of-kin during the study interview process.

A global subjective rating of hopelessness was measured during chart review with the Hopelessness Severity Scale using information from the diagnostic interview, family history, and medical records. The Hopelessness Severity Scale was developed for the purposes of this study and anchored in the Beck Hopelessness Scale (BHS; Beck et al., 1974) as well as a 9-point hopelessness scale used in Beck et al. (1989). The BHS has obtained adequate interrater reliability of .86 (Beck et al., 1974). In prediction of eventual suicide among 141 psychiatric inpatients, the 9-point hopelessness rating had a specificity of 90.9% (Beck et al., 1989). The Hopelessness Severity Scale offered a broader spectrum of classification in order to better gauge a more realistic range of feelings of hopelessness. The Hopelessness Severity Scale was rated regarding the 6 months prior to death, with scores ranging along a continuous scale from 0 to 150 in a visual analog scale format. The Hopelessness Severity Scale assessed: “During the six months prior to death, how strongly did the person feel generally hopeless (i.e., had negative expectations) about their future across one or multiple life domains (e.g., physical status or health problems, financial status, treatment effectiveness, romantic, familial, and/or social relationships)? This can include the person feeling like they were never going to get well, the future was not going to get any better (dark/pessimistic view of future), feeling like there was nothing to look forward to, feeling like they were not going to achieve their goals, and wanting to give up.”

A rating of wish to die and a rating of wish to live were measured during chart review with the Ambivalence Severity Scale, developed for the purpose of this study and adapted from the first two items of the Modified Scale for Suicidal Ideation (MSSI). The MSSI is an 18-item scale measuring the degree of suicidal ideation and suicidal ideas within 48 hours of administration (Miller et al., 1991). The internal consistency of the MSSI has been shown to be high with an alpha coefficient of .94, and an inter-rater reliability coefficient of .99 (Miller et al., 1986). The construct and discriminate validity of the MSSI were acceptable when compared to scores on the Beck Depression Inventory and physician rating (Clum & Yang, 1995; Miller et al., 1986). The Ambivalence Severity Scale was rated for both a wish to live and a wish to die for each decedent during the 6 months prior to death, offering a broader spectrum of classification in order to better gauge a more realistic range of feelings of ambivalence. Scores ranged along a continuous scale from 0 to 150 in a visual analog scale format. The first item assessed wish to die and asked, “During the six months prior to death, how strongly did the decedent have a wish to die? How strongly did they want to die when they did die? This can include the person having a clear wish to die, having occasional thoughts about dying, or having thoughts about wanting to die in different degrees of frequency and intensity.” The second item assessed wish to live and asked, “During the six months prior to death, how strongly did the decedent have a wish to live? How strongly did they care if they lived or died when they died? This can include the person being unsure about whether he/she wanted to live, having occasional thoughts about living, or having thoughts about wanting to live in different degrees of frequency and intensity.”

For the Hopelessness Severity Scale and both items of the Ambivalence Severity Scale, the relevant construct was rated on a continuous scale from “not at all” to “totally,” with anchors for several discrete levels (not at all, a little bit, somewhat, very much, totally). Using “not at all” as a zero marker, a measurement was taken of the distance from zero to the rating line in millimeters in order to determine the final quantitative severity rating out of 150. For both scales, a confidence rating was provided ranging from 0 (not confident at all) to 3 (very confident) based on rater confidence in having enough information for a reliable score. When the confidence level was rated as 0 (not confident at all) or 1 (slightly confident) from one or both raters, the chart was excluded from the subjective rating analyses.

While every decedent had at least one suicide attempt, it was possible that there were multiple attempts across his or her lifetime. Specific variables concerning the first suicide attempt were collected during chart review of all available records, including: age at first attempt, method of first attempt, time since first attempt until eventual death in months, mental health treatment after the attempt, and total number of lifetime attempts.

Procedures

When notified of an eligible death, next-of-kin were approached and consented to participate in the study on behalf of the deceased. Decedents categorized as having died due to natural causes were considered eligible where the demographic characteristics and quality of the postmortem tissue of the deceased most closely matched those of the decedents determined to have died by suicide. Approximately 100–140 autopsies were performed each month. About 1–3 cases met criteria for inclusion per month after excluding homicides, cases autopsied on weekends, those for which tissue quality was unacceptable, and those who’s next-of-kin declined consent. All interview data were collected from next-of-kin and additional information was gathered (Ebert, 1987), including demographic variables, history of psychological problems, history of psychiatric treatment, and MLES data. Legal and medical records were obtained, where available, as collateral information. Case consensus meetings were attended by a master’s level social worker, a board-certified clinical psychologist, a board-certified psychiatrist, and a neuroscientist and were held regularly to determine authenticity of collected data and agreed upon diagnoses. During retrospective chart review procedures, all decedents who had a previous suicide attempt were split into two groups depending on cause of death (suicide or natural causes). The history of previous suicide attempts was determined from medical records, medical examiner records, and/or informant-based retrospective interviews. Retrospective chart review methods were used to glean further information from participant charts as well as available collateral records. A trained research assistant (SB) and the primary author (EB) completed the chart reviews of available data and coded the Hopelessness Severity Scale, the Ambivalence Severity Scale, and the factors surrounding the index suicide attempt.

Results

Six cases (29%) from the Natural Cause Death group and 13 cases (20%) from the Suicide Death group were randomly chosen to assess inter-rater reliability for the Hopelessness Severity Scale and Ambivalence Severity Scale. Inter-rater agreement was assessed using intraclass correlations and their respective 95% confidence intervals based on a single-rater, consistency, two-way random-effects model and calculated after eliminating cases with a confidence rating of 0 (not confident at all) or 1 (slightly confident) on subjective measures. The final interrater reliability for the continuous measure of hopelessness was found to have strong agreement between raters (κ = 0.81, 95% Confidence Interval (CI) [0.58–.92], p < .001). Both the wish to die scale (κ = 0.71, 95% CI [0.39–.88], p < .001) and the wish to live scale (κ = 0.60, 95% CI [0.21–.82], p = .003) measuring ambivalence were found to have inter-rater reliability agreement in the moderate range.

Wave 1 Analyses

Wave 1 analyses included all available decedents (N = 86) resulting in 65 cases in the Suicide Death group and 21 cases in the Natural Cause Death group. When comparing the Suicide Death group to the Natural Cause Death group on demographic variables using chi-square tests of independence, only sex (χ² (1, N = 86) = 3.75, p = .05, OR = 2.66, 95% CI [0.97, 7.31]) was found to be significantly different, with more males dying by suicide than by natural causes (Table 1). There were no significant differences between groups for age, race, education, marital status, or employment status. Individuals who died by natural causes were significantly more likely to have exhibited traits associated with personality disorders (χ2 (1, N = 83) = 3.86, p = .05, OR = 0.33, 95% CI [0.11, 0.95]) and to have a polysubstance use disorder (χ² (1, N = 86) = 10.51, p = .001, OR = 0.13, 95% CI [0.03, 0.31]) than those who died by suicide. Compared to the Suicide Death group, individuals who died by natural causes were significantly less likely to have alcohol in their system at time of death (χ² (1, N = 86) = 10.56, p = .001) and more likely to have an antidepressant drug in their system at time of death (χ² (1, N = 84) = 4.62, p = .03, OR = 0.31, 95% CI [0.10, 0.93]; Table 3). Significantly more individuals who died by natural causes experienced personal health stressors during the 6 months prior to death compared to those who died by suicide (χ² (1, N = 86) = 13.79, p < .001, OR = 0.14, 95% CI [0.05, 0.42]; Table 4).

Table 1.

Comparison of demographic characteristics of decedents with a previous suicide attempt who went on to die by suicide or by natural causes.

		Wave 1			Wave 2
		Suicide Death (n = 65)	Natural Cause Death (n = 21)	Test Statistic	Suicide Death (n = 21)	Natural Cause Death (n = 21)	Test Statistic
Age	M (SD)	45.58 (18.01)	50.48 (15.82)	1.11	49.86 (14.61)	50.48 (15.82)	.13
Male sex	n (%)	46 (70.8%)	10 (47.6%)	3.75*	10 (47.6%)	10 (47.6%)	.00
White race	n (%)	58 (89.2%)	19 (90.5%)	.03	19 (90.5%)	19 (90.5%)	.00
> High school diploma	n (%)	32 (30.0%)a	8 (38.1%)	.90	12 (60.0%)b	8 (38.1%)	1.97
Married	n (%)	18 (27.7%)	5 (23.8%)	.12	6 (28.6%)	5 (23.8%)	.12
Employed	n (%)	26 (40.6%)a	5 (23.8%)	1.93	11 (52.4%)	5 (23.8%)	3.64

Note: All test-statistics reported are χ² except for age which is reported as t.

Abbreviations: M, Mean; SD, standard deviation.

^aDue to missing data: Out of 64.

^bDue to missing data: Out of 20.

^*p ≤ .05.

Table 3.

Comparison of medical examiner toxicology results at death of decedents with a previous suicide attempt who went on to die by suicide or by natural causes.

		Wave 1			Wave 2
		Suicide Death (n = 65)	Natural Cause Death (n = 21)	χ2 Statistic	Suicide Death (n = 21)	Natural Cause Death (n = 21)	χ2 Statistic
Alcohol toxicology	n (%)	23 (36.5%)	0 (0.0%)	10.56***	7 (33.3%)e	0 (0.0%)	8.40**
Drugs in toxicology	n (%)	25 (38.5%)	9 (42.9%)	.13	5 (23.8%)	9 (42.9%)	1.71
Antidepressants in toxicology	n (%)	10 (15.9%)e	8 (38.1%)	4.62*	2 (9.5%)	8 (38.1%)	4.73*

Due to missing data:

^eOut of 63

^*p-value ≤ .05,

^**p-value ≤ .01,

^***p-value ≤ .001

Table 4.

Comparison of life stressors during the six months prior to death for decedents with a previous suicide attempt who went on to die by suicide or by natural causes.

		Wave 1			Wave 2
		Suicide Death (n = 65)	Natural Cause Death (n = 20)	χ2 Statistic	Suicide Death (n = 21)	Natural Cause Death (n = 20)	χ2 Statistic
Bereavement	n (%)	4 (6.2%)	1 (0.5%)	.04	1 (4.8%)	1 (5.0%)	.00
Finances	n (%)	17 (26.2%)	4 (20.0%)	.31	6 (28.6%)	4 (20.0%)	.41
Occupation	n (%)	14 (21.5%)	4 (20.0%)	.02	6 (28.6%)	4 (20.0%)	.41
Personal health	n (%)	16 (24.6%)	14 (70.0%)	13.79***	7 (33.3%)	14 (70.0%)	5.51*
Family health	n (%)	5 (7.7%)	2 (10.0%)	.11	2 (9.5%)	2 (10.0%)	.00
Relationships	n (%)	30 (46.2%)	6 (30.0%)	1.64	8 (38.1%)	6 (30.0%)	.30
Legal issues	n (%)	10 (15.4%)	2 (10.0%)	.37	3 (14.3%)	2 (10.0%)	.18
Other	n (%)	10 (15.4%)	2 (10.0%)	.37	3 (14.3%)	2 (10.0%)	.18

^*p-value ≤ .05,

^***p-value ≤ .001

Independent samples t-tests were used to compare groups on subjective ratings of hopelessness, wish to die, and wish to live during the 6 months prior to death (Table 5). Compared to individuals who died by natural causes, those who died by suicide were rated to have significantly more hopelessness (t (78) = 4.22, p < .001, d = 0.89, 95% CI [−34.82, −12.49]), a significantly greater wish to die (t (78) = 9.21, p < .001, d = 1.91, 95% CI [−67.17, −43.29]), and a significantly lower wish to live (t (78) = 4.95, p < .001, d = 1.09, 95% CI [21.04, 49.35]).

Table 5.

Comparison of ratings of hopelessness and ambivalence during the six months prior to death for decedents with a previous suicide attempt who went on to die by suicide or by natural causes.

		Wave 1			Wave 2
		Suicide Death (n = 58)	Natural Cause Death (n = 20)	t Statistic	Suicide Death (n = 18)	Natural Cause Death (n = 20)	t Statistic
Hopelessness	M (SD)	124.16 (14.93)	100.50 (34.67)	4.22***	122.44 (16.25)	100.50 (34.67)	2.56*
Wish to die	M (SD)	122.38 (15.32)	67.15 (37.88)	9.21***	117.44 (15.06)	67.15 (37.88)	5.26***
Wish to live	M (SD)	34.60 (21.62)	69.8 (40.04)	4.95***	43.67 (27.70)	68.80 (40.04)	2.31*

M = Mean; SD = Standard deviation

^*p-value ≤ .05,

^***p-value ≤ .001

Chi-square tests of independence were used to examine the differences between the Suicide Death group and the Natural Cause Death group for the lethality of the first attempt, evidence of mental health care following the first attempt, and occurrence of two or more lifetime attempts not including the first attempt or the fatal attempt when relevant. Difference in age at first attempt between groups was assessed using an independent samples t-test. Compared to those who died by suicide, those who died by natural causes were alive for a significantly higher number of months after their first attempt (M = 105.61, t (78) = 2.67, p = .01, d = 0.65, 95% CI [21.04, 49.35]). Individuals in the Natural Cause Death group were also significantly more likely to have two or more lifetime attempts, not including the first attempt or the eventual fatal attempt, when compared to those in the Suicide Death group (χ² (1, N = 86) = 3.98, p = .05, OR = 0.36, 95% CI [28.21, 193.69]).

Wave 2 Analyses

Wave 2 analyses included decedents matched on age within 5 years, race, and sex resulting in 21 cases in the Suicide Death group and 21 cases in the Natural Cause Death group. No significant differences were found between the Suicide Death group and the Natural Cause Death group for education, marital status, or employment status (Table 1). There were no longer significant differences between groups concerning personality disorder traits (Table 2). However, individuals in the Suicide Death group were significantly more likely to have a depressive disorder than those in the Natural Cause Death group (Major Depression, Depression NOS, or Adjustment Disorder with depressed mood; χ² (1, N = 42) = 5.08, p = .02, OR = 4.68, 95% CI [1.17, 18.69]), while individuals who died by natural causes more significantly more likely to have a polysubstance use disorder (χ² (1, N = 42) = 5.56, p = .02, OR = 0.10, 95% CI [0.01, 0.91]). In comparison to the Natural Cause Death group, individuals who died by suicide were significantly less likely to have an antidepressant drug in their system at time of death (χ² (1, N = 42) = 4.73, p = .03, OR = 0.17, 95% CI [0.03, 0.94]; Table 3) and significantly more likely to have alcohol in their system at time of death (χ² (1, N = 42) = 8.40, p = .004). Individuals who died by natural causes were more likely to have personal health stressors during the 6 months prior to death than those who died by suicide (χ² (1, N = 41) = 5.51, p = .02, OR = 0.21, 95% CI [0.06, 0.80]; Table 4).

Table 2.

Comparison of clinical diagnoses of decedents with a previous suicide attempt who went on to die by suicide or by natural causes.

		Wave 1			Wave 2
		Suicide Death (n = 65)	Natural Cause Death (n = 21)	χ2 Statistic	Suicide Death (n = 21)	Natural Cause Death (n = 21)	χ2 Statistic
Depressive disorder (% present)	n (%)	44 (67.7%)	10 (47.6%)	2.74	17 (81.0%)	10 (47.6%)	5.08*
Anxiety disorder	n (%)	6 (9.2%)	5 (23.8%)	3.02	4 (19.0%)	5 (23.8%)	.14
Bipolar disorder	n (%)	7 (10.8%)	2 (9.5%)	.03	1 (4.8%)	2 (9.5%)	.36
Posttraumatic stress disorder	n (%)	3 (4.6%)	0 (0.0%)	1.00	1 (4.8%)	0 (0.0%)	1.02
Psychotic disorder	n (%)	10 (15.4%)	6 (28.6%)	1.82	4 (19.0%)	6 (28.6%)	.53
Any personality disorder	n (%)	29 (46.8%)b	10 (47.6%)	.00	12 (57.1%)	10 (47.6%)	.38
Personality disorder traits	n (%)	13 (21.0%)b	9 (42.9%)	3.86*	4 (19.0%)	9 (42.9%)	2.79
Alcohol use disorder	n (%)	33 (51.6%)a	11 (52.4%)	.00	6 (28.6%)	11 (52.4%)	2.47
Drug use disorder	n (%)	21 (33.9%)b	7 (33.3%)c	.01	2 (10.5%)d	7 (35.0%)c	3.29
Polysubstance use disorder	n (%)	4 (6.2%)	7 (33.3%)	10.51***	1 (4.8%)	7 (33.3%)	5.56*

Due to missing data:

^aOut of 64,

^bOut of 62,

^cOut of 20,

^dOut of 19

^*p-value ≤ .05,

^***p-value ≤ .001

Independent samples t-tests were used to compare groups on subjective ratings of hopelessness, wish to die, and wish to live during the 6 months prior to death (Table 5). The finding remained that individuals who died by suicide were rated as having significantly more hopelessness (t (38) = 2.56, p = .02, d = 0.85, 95% CI [−41.10, −4.79]), significantly greater wish to die (t (38) = 5.26, p < .001, d = 1.74, 95% CI [−69.66, −30.92]), and significantly lower wish to live (t (38) = 2.31, p = .03, d = 0.76, 95% CI [3.23, 49.04]) during the 6 months prior to death. None of the factors present at the time of the first suicide attempt remained significantly different between the Suicide Death group and the Natural Cause Death group after controlling for age, race, and sex.

Discriminant Function Analysis

A discriminant function analysis asked whether the available measures could be combined to accurately predict group membership, using variables found to be significant in Wave 2. Mahalanobis distance indicated that there were no multivariate outliers (z-scores > 3.0) present prior to the analysis. Two cases with missing data were removed from the analyses per requirements of discriminant function analysis. Enough cases remained in the Suicide Death group (n = 20) and Natural Cause Death group (n = 20) to satisfy the required sample size to analyze the nine significant variables from Wave 2 (Poulsen & French, 2008).

The statistical model was shown to be significant (Wilks’ Lambda = .289, χ² (8, N = 40) = 42.23, p < .001) and correctly classified 92.5% of the cases (Table 6). The overall effect size of the model, the canonical correlation value, was .84. The structure matrix presented correlation coefficients for each of the included variables, where positive coefficient values signify a positive correlation between the variable and the Suicide Death group and negative values signify a negative correlation between the variable and the Suicide Death group (Table 6). The standardized canonical discriminant function coefficients provided information regarding the importance of each predictor individually to correct classification. From highest importance to lowest importance the variables were ranked (Table 6): wish to die = 1.208, polysubstance use disorder = .700, wish to live = .540, personal health stressors = .486, antidepressant drug in toxicology = .386, alcohol in toxicology = .218, hopelessness = .119, and depressive disorder = .023.

Table 6.

Discriminant function analysis classification results for predicting cause of death based on significant variables, for model contribution statistics, and for predictor importance to classification.

Count and Percentage
	Predicted Group Membership
	Suicide Death	Natural Cause Death	Correct Classifications
Suicide Death	19 95.0%	1 5.0%	19/20 95.0%
Natural Cause Death	2 10.0%	18 90.0%	18/20 90.0%
Total			37/40 92.5%

Variable	Structure Coefficient	Importance (SCDFC)
Depressive disorder	.097	.023
Polysubstance use disorder	−.258	.700
Alcohol in toxicology	.331	.218
Antidepressants in toxicology	−.235	.386
Personal health stressors	−.278	.486
Hopelessness	.215	.119
Wish to die	.507	1.208
Wish to live	−.219	.540

Note: Wilks’ Lambda = .289, χ2 (8, N = 40) = 42.23, p < .001

SCDFS = Standardized canonical discriminant function coefficients.

Power Analyses

To assess the statistical power present in our analyses, post-hoc power analyses using GPower were conducted for both independent samples t tests and chi-square tests of independence (Faul et al., 2009). The alpha level was set at .05, the effect size was set at .3, and the two groups were entered for Wave 1 (suicidal death group = 65; natural cause death group = 21) and Wave 2 (suicidal death group = 21; natural cause death group = 21) for both types of analyses. For independent sample t tests, GPower output showed the present study to have power (1 - β) equal to .8 for Wave 1 and power (1 - β) equal to .5 for Wave 2. Viewed from an a priori standpoint, achieving statistical power of .8 for t tests would have required each group to include at least 42 cases, totaling 82 cases for the overall sample size. For chi-square analyses, GPower output showed the present study to have power (1 - β) equal to .8 for Wave 1 and power (1 - β) equal to .5 for Wave 2. Viewed from an a priori standpoint, achieving statistical power of .8 for chi-square analyses would have required each group to include at least 44 cases, totaling 88 cases for the overall sample size. Thus, it is possible that some differences between groups assessed were not found due to an inadequate level of statistical power.

For discriminant function analyses, the smallest group must have a sample size greater than the number of predictor variables in order to be adequately powered (Poulsen & French, 2008). A more general rule when using discriminant function analysis is that while having unequal sample sizes are accepted, there should be about 20 participants total for approximately 5 predictor variables or four times the number of predictor variables (Poulsen & French, 2008). The current sample included 20 in the suicidal death group and 20 in the natural cause death group, satisfying the first definition but not the second. Thus, the current results from the discriminant function analysis should be interpreted with caution.

Discussion

The present results suggest that individuals who die by suicide experience a state of distress prior to death that may help identify those at increased risk and offer points of intervention. Despite all decedents sharing a history of suicide attempt, important differences were found between those who died by suicide and those who died by natural causes. Adults with a prior attempt who died by suicide were more likely to have a depression diagnosis and to have alcohol in their blood at time of death. In contrast, decedents who died by natural causes were more likely to have polysubstance use disorder, an antidepressant drug in their system at time of death, and personal health stressors during the 6 months prior to death. Individuals who died by suicide were rated subjectively as having a worsened outlook, with a stronger orientation toward death and away from life. Individuals who eventually died by suicide experienced heightened hopelessness, greater wish to die, and lesser wish to live during the 6 months prior to death compared to those who died by natural causes.

Individuals who died by natural causes displayed a profile distinguishable from those who died by suicide. The prevalence of polysubstance use disorder in the natural cause death group aligns with the association of substance use and attempted suicide (Poorolajal et al., 2016). However, the rate of substance use in individuals who died by natural causes does not confirm the previously supported association of substance use with eventual death by suicide (Poorolajal et al., 2016). Adults who have attempted suicide but eventually die of natural causes may be more likely to cope with life circumstances by using substances. The self-medication hypothesis argues that use, abuse, and dependence upon substances occurs because of the subsequent relief from distress and numbing effects (Khantzian, 2003). Individuals who died by natural causes were more likely to have two or more lifetime suicide attempts not counting their first attempt, however this finding was no longer significant after controlling for age, race, and sex. Thus, the struggle with suicidal thoughts does not end for individuals who eventually die by natural causes, but may persist for an extended period of time signified by a greater number of suicide attempts as well as a greater inclination to use psychoactive substances.

Decedents who died by natural causes were more likely to have an antidepressant drug in their blood at time of death. Perhaps these individuals were more engaged in mental health treatment or had a tendency to seek medical treatment for their depression, although we did not have data to analyze this possibility. The relationship between antidepressant drug use and risk for suicide is a complicated one (Courtet & Lopez-Castroman, 2017). The use of newer selective serotonin reuptake inhibitors may have helped decrease the rates of suicidal overdose on antidepressant medications due to their lesser toxicity compared to older tricyclic medications (Grunebaum et al., 2004). However, there has been an overall increase in the number of antidepressant drug suicidal overdoses, despite there being fewer fatalities (McKenzie & McFarland, 2007). The present findings may support the benefit of antidepressant medications for combatting major depression. However, the present study was unable to assess the timing or duration of antidepressant drug use in this population; therefore, conclusions are made with caution.

The higher likelihood of having personal health stressors during the 6 months prior to death in those who died by natural causes after controlling for age points to the medical issues that may have led to their eventual death. In a case controlled study of 2,674 individuals who died by suicide, 17 different physical health conditions were significantly associated with increased risk for suicide (Ahmedani et al., 2017). The current results do not imply an association between health stress and suicide in the individuals who died by natural causes toward the end of their life. Individuals who died by natural causes displayed significantly lower levels of hopelessness and did not resort to suicidal action, even while next-of-kin reported elevated health-related stress. A lower level of hopelessness or less ambivalence about living may have boosted resilience in individuals who died by natural causes. In cancer patients undergoing chemotherapy, hopefulness was found to be significantly and negatively correlated with symptoms of grief (Gökler-Danışman et al., 2017). Alternatively, individuals who died from natural causes may have had an increased risk of health stress due to a longer amount of time lived after their first suicide attempt compared to those who died by suicide.

Individuals in the current study who went on to die by suicide showed a different pattern of risk factors, validating some already known risk determinants (e.g. being male and in the matched sample having a diagnosis of depression), while also supporting the influence of important constructs better describing their distress (e.g. hopelessness and ambivalence). Depression is commonly found to be higher in individuals who die by suicide (Conner et al., 2019). Individuals suffering from depression may be more inclined to suicidal action as a result of living with a negative view of self, a negative interpretation of ongoing experiences, and a negative view of the future. High levels of depression severity and increased variability in symptom severity have been found to be associated with risk of attempting suicide (Melhem et al., 2019). In a comparison of suicidal depressed vs. non-suicidal depressed inpatients, the time to recovery of depression symptoms was significantly longer for those individuals who were suicidal (Overholser et al., 1987). Thus the road to recovery and improvement in depressive symptoms may be delayed in depressed patients who are also suicidal. Individuals who died by suicide were significantly more likely to have alcohol in their blood at time of death compared to those who died by natural causes. Acute alcohol consumption significantly increases the risk of suicide attempt, with higher levels of consumption leading to higher levels of risk (Borges et al., 2017). Consuming alcohol may lower a person’s inhibitions, thus increasing the odds of acting on suicidal urges. Of note, while there was a significant difference between groups, the majority of decedents in the Suicide Death group did not have alcohol in their toxicology screen at time of death, suggesting these results are applied to a minority of the present sample. Although assessing depression and alcohol use as risk factors for suicide is not new, increased focus on these risk factors in individuals who have a history of a suicide attempt may help identify those at greater risk of eventual suicide death.

In the present study, individuals who died by suicide had a significantly higher level of hopelessness, a greater wish to die, and a lower wish to live. Hopelessness has been strongly linked to risk of suicide attempt (Beck et al., 1974), particularly for individuals who have multiple suicide attempts (Forman et al., 2004). Despite having greater health stress and higher occurrence of polysubstance use disorders, the Natural Cause Death group had less hopelessness than the Suicide Death group. In 934 military personnel, grit, defined as persistently pursuing goals even when faced with obstacles or challenges, decreased the strength of the relationship between hopelessness and suicidality (Pennings et al., 2015). There may be an important difference in outlook that influences individuals who act on suicidal thoughts. Hopelessness alone relates to suicide risk. Yet, in combination with an ambivalent perspective of life this construct may prove increasingly detrimental to the ability of a suicidal individual to continue to wish to live, even if that wish to live outweighs the wish to die.

Having a greater wish to die and a lower wish to live distinguished those who went on to die by suicide from those who died by natural causes. The range of scores seen on these measures suggests true variability within these constructs supporting the distinction of a low wish to live from a high wish to die. Higher hopelessness or a wish to die might overpower any reasons for living, tipping a more balanced perspective toward death. However, the ratings of hopelessness and ambivalence spanned the 6 months prior to death, meaning there may be ample time for recognizing and intervening to prohibit progression to a suicidal crisis. An acute suicidal crisis is thought to last approximately 30 minutes (Kattimani et al., 2016). However, the ongoing chronic state of experiencing suicidal thoughts lasts an unknown amount of time that varies between individuals. The current markers of ambivalence and hopelessness may help to identify the more chronic mindset of individuals who struggle with ongoing or repeated suicidal ideation. A recent study found that suicidal ideation can occur anywhere from 1 to 5 years before a suicide attempt, while the onset of steps toward a current attempt can begin between 2 weeks and a few hours before a suicide attempt (Millner et al., 2017). Thus, a window of opportunity exists before an individual gets to the point of crisis. The construct of ambivalence, and its relationship with hopelessness, may contribute to the trajectory individuals travel down during this time.

The discriminant function analysis identified who would go on to die by suicide with 92.5% accuracy using significant differences between the groups, with wish to die, presence of polysubstance use disorder, and wish to live as the factors with highest discriminant power. The use of a discriminant function analysis combined with psychological autopsy methods to test the prediction of death type provided a rare opportunity to determine the utility of the assessed constructs. While wish to die was the highest weighted variable in the discriminant functional analysis, the successful group identification resulted from a combination of variables. The wish to die and wish to live were in the top three strongest discriminators of eventual type of death, highlighting the usefulness of better understanding and effectively measuring the construct of ambivalence. The relationship between an individual’s wish to live and wish to die is inherently complex and more work needs to be done in order to understand how change in one of these constructs influences change in the other as well as suicide risk overall.

The present sample was rather small, particularly after examining groups matched for age, race, and sex, resulting in lowered power of the analyses to detect significant differences. Given the nature of the present psychological autopsy data, collecting additional cases to increase sample size and improve power was not an option, however, results should be interpreted with caution given this limitation. Additionally, the majority of the sample consisted of White males, thus limiting the generalizability of the findings. By controlling variability due to race and sex in the matched groups, these findings allowed a closer look at psychosocial variables instead of demographic differences. Finally, the present sample was chosen due to the important risk of eventual suicide involved in having a prior suicide attempt. As a result, the Natural Cause Death group may not represent individuals who die by natural causes and who do not have a prior suicide attempt.

Psychological autopsy research does involve some innate limitations. The basic methodology has been questioned because of the intrapsychic nature of many psychiatric symptoms (Hjelmeland et al., 2012). The data used in the present study had been collected over a 28-year period which included some changes and advancements in clinical knowledge, staff personnel, and diagnostic criteria. Additionally, informant report and available medical records may have lacked required information for a complete case conceptualization. For example, decedents may have had a higher number of lifetime suicide attempts than the informants knew about to report or was present in the collateral information. Nevertheless, psychological autopsy methods have demonstrated reliability and a variety of studies have examined the validity of post-mortem interviews.

In general, research has supported the use of post-mortem interviews with knowledgeable family members. Psychological autopsy has been shown to be an effective method for thoroughly assessing an individual’s state of mind and life circumstances post-mortem in order to investigate suicide deaths (Weinberger et al., 2018). Thus, family member informants are able to provide accurate information about the decedent’s medical history (Klinkenberg et al., 2003), the presence of psychotic disorders and mood disorders (Sundqvist et al., 2008), and even internal states such as hopelessness (Ma et al., 2020). Further, research has found a high level of agreement when psychological autopsy procedures are evaluated using a sample of living depressed psychiatric inpatients as compared to their family member informants (DeJong & Overholser, 2009). Psychological autopsy procedures have been recommended for use in cases where cause of death may be suspicious (Saxena & Saini, 2017), such as cases of accidental overdose (Hakansson & Gerle, 2018). If conducted in a thorough and standardized manner, psychological autopsy procedures have the scientific validity to be acceptable as evidence in a court of law (Snider et al., 2006). Psychological autopsy should rely on an interdisciplinary collaboration (Solomon, 2018) that incorporates information from several informants and multiple sources, such as prior medical records (Connor et al., 2012).

When considering the psychological autopsy procedures as well as the retrospective chart review, certain limitations must be considered concerning subjective ratings. Neither the interviewed next-of-kin nor the trained chart review raters were blinded to the cause of death. When a family member dies, their loved one’s memory of the deceased life, strengths, and personality may shift. An informant’s report may be influenced by the time since the death, the intimacy of the bond, and the frequency of contact with the decedent (Pouliot & De Leo, 2006). After death by suicide, the surviving family members experience higher levels of shame, rejection, and social stigma as compared to other forms of bereavement (Kõlves et al., 2020; Sveen & Walby, 2008). The grief reactions may be further complicated because of the limited social support that sometimes follows a death by suicide (Spillane et al., 2017). These qualities may have an impact on how the informants in our study perceived the psychiatric symptoms of the decedent. Additionally, the two chart reviewers were not blinded to cause of death, which could have influenced the way in which subjective scores of hopelessness and ambivalence were collected. However, blinding raters to the suicidal or natural cause of the subject death would have been impossible given the information available to provide the ratings. For each subject there was a great deal of information capturing the months leading up to death and regular mention of details of the death itself. The ongoing events in each subject’s life during the months leading up to death provided crucial information concerning the level of hopelessness and ambivalence experienced.

In a clinical context, a psychologist’s subjective opinion on these matters is one portion of clinical judgement and exploring the relationships of subjective ratings remains useful. Clinicians treating suicidal clients are not blind to the ongoing thoughts of death that plague these individuals. Whether the eventual death of a client is due to suicide or natural causes, orientation toward wanting to live and wanting to die remains vital information in terms of the therapeutic process. Utilizing subjective ratings supports the importance of clinical judgement when gauging risk and the current study showed one method of working to increase the objective process of determining important subjective clinical ratings.

Conclusion

The current study provides new information regarding ways in which individuals who die by suicide are distinct from those who die by natural causes despite a shared history of attempting suicide. More remains to be done to better define the construct of ambivalence and to further assess the nuances in the relationships among clinical variables over time. Hopelessness and ambivalence offer changeable targets for suicide prevention in a therapeutic context. Gaining information about how these states manifest both separately and in conjunction could assist the field of psychology in decreasing the rapidly rising rates of suicide.

The present study explored potential differences between individuals who went on to die by suicide compared to those who died by natural causes, despite a shared history of suicide attempt. Through the use of psychological autopsy methods and prospective studies, research and clinical work should continue to examine hopelessness and ambivalence to understand their role in the distressed state of suicidal individuals. Aiming to change the ambivalence between life and death experienced by a person who considers taking their own life may help make the path to suicide a path not taken.