ORIGINAL ARTICLE
Predictors of Future Suicide Attempts Among Individuals Referred to Psychiatric Services in the Emergency Department A Longitudinal Study Yunqiao Wang, MA,* Joanna Bhaskaran, MA,* Jitender Sareen, MD,*†‡ JianLi Wang, PhD,§k Rae Spiwak, MSc,‡ and James M. Bolton, MD*†‡
Abstract: This study examined which factors predict future suicide attempts (SAs) among people referred to psychiatric services in the emergency department (ED). It included consecutive adult (age >18 years) presentations (N = 6919) over a 3-year period to the two tertiary care hospitals in Manitoba, Canada. Medical professionals assessed each individual on 19 candidate risk factors. Stepwise logistic regression and receiver operating characteristic curves examined the association between the baseline variables and future SAs within the next 6 months. A total of 104 individuals re-presented to the ED with future SAs. Of the 19 baseline variables, only two independently accounted for the variance in future attempts. High-risk scores using this two-item model were associated with elevated odds of future SA (odds ratio, 3.22; 95% confidence interval, 1.62–6.42; p < 0.01), but this was tempered by a low positive predictive value. Further evaluation is required to determine if this two-item tool could help identify people requiring more comprehensive risk assessment referred to psychiatry in the ED. Key Words: Suicide attempt, suicidal ideation, emergency department, risk factors, assessment, SAD PERSONS (J Nerv Ment Dis 2015;203: 507–513)
S
uicide is a leading cause of death worldwide (Centers for Disease Control and Prevention, 2012). Suicide attempts (SAs) are believed to be 20 times more frequent than completed suicide, and in the United States there are 2.5 million SAs each year (US Department of Health and Human Services (HHS) Office of the Surgeon General and National Action Alliance for Suicide Prevention, 2012; World Health Organization, 2010). SAs are among the strongest risk factors for suicide completion (Nordentoft et al., 2011; Sokero et al., 2003), and those who attempt suicide exhibit great similarities with individuals who die by suicide (Suominen et al., 2004). Every year, there are thousands of patients who visit the emergency department (ED) with suicidal ideation and SA as their major complaints (Olfson et al., 2012). Presenting to the ED with self-harm behavior increases an individual’s risk of suicide 6-fold (Crandall et al., 1997), and those who visited the ED for deliberate self-harm have shown to be at higher short-term risk for revisiting the ED with the same complaint (Olfson et al., 2013; Steeg et al., 2012). The recently released US Surgeon General’s 2012 National Strategy for Suicide Prevention highlights specific goals of evaluating research on screening of high-risk individuals and emphasizes the ED as a key clinical area to managing suicide risk (US Department of Health and Human Services (HHS) Office of the Surgeon General and National Action Alliance for Suicide Prevention, 2012). As such,
Departments of *Psychology, †Psychiatry, and ‡Community Health Sciences, University of Manitoba, Winnipeg, Manitoba; and Departments of §Psychiatry and kCommunity Health Sciences, University of Calgary, Calgary, Alberta, Canada. Send reprint requests to James M. Bolton, MD, PZ430-771 Bannatyne Avenue, Winnipeg, Manitoba, Canada R3E 3N4. E-mail: [email protected]. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved. ISSN: 0022-3018/15/20307–0507 DOI: 10.1097/NMD.0000000000000320
the ED is a critical component of the broader public health priority to reduce the rate of suicide. Most of existing risk assessment scales used in the ED are unable to predict future suicidal behaviors (Bolton et al., 2012; Randall et al., 2011). For example, Bolton et al. used the receiver operating characteristic (ROC) curves in a sample of 4019 ED presentations and demonstrated that the SAD PERSONS scale, a widely used instrument in assessing one’s risk of suicide, was no better than chance in predicting future SAs (Bolton et al., 2012). This often leaves many physicians doubting the results of their assessments and wondering what factors would dependably contribute to future risk (Ronquillo et al., 2012). Furthermore, the systematic review of Randall et al., which examined 12 studies on the predictability of future suicidal behavior, revealed that although these instruments have impressive psychometric properties, there is very little clinical utility. Unfortunately, the study did not perform a meta-analysis, and the presence of publication or selection biases is unknown. A recent study showed improved accuracy using a statistically derived tool among people presenting with self-harm (Steeg et al., 2012). Several other studies have found that a positive SA screening predicts future ED presentations among various age groups in a number of settings (Ballard et al., 2013; Ting et al., 2012). Ting et al. specifically found that alcohol and other substance use was the most significant predictor for future SA and noted that many assessment tools do not assess this issue. However, the screening process may vary substantially between EDs and many self-harm and suicidal presentations are undetected (Caterino et al., 2013). These authors commented that older adults were especially more likely to be neglected. A review of suicide risk assessment tools revealed that, in general, there is a lack of evidence in supporting the use of traditional risk assessment tools, and most of the existing tools need further investigation and replication (Roos et al., 2013). More of these studies are required, not only among people presenting with self-harm but also including the broader group of ED psychiatric patients, because mental illness confers a heightened risk of suicide even in the absence of SA, and many suicides occur without previous attempt (Nordentoft et al., 2011). Because of the already backed-up and overcrowded ED, extensive suicide risk assessments have limited feasibility, and therefore, our current situation calls for further research to devise a quick and reliable evaluative instrument (Mitchell et al., 2005). So far, some of the most common risk factors identified in the literature include history of suicidal behavior (Nordentoft et al., 2011; Sokero et al., 2003; Suominen et al., 2004), psychiatric disorders (Bolton et al., 2008; Bolton and Robinson, 2010; Sareen, 2011; Sareen et al., 2005; Schneider et al., 2008; Tidemalm et al., 2008), and stressful life events (Wang et al., 2012), among many others. In the emergency setting specifically, factors such as access to means (Williams-Johnson et al., 2012), depression (Buzan and Weissberg, 1991), having suicidal plans (Cooper et al., 2003), and the female sex (Elisei et al., 2012) are correlated with an elevated risk for making an SA. Depression and a previous SA are among the strongest predictors for suicide completion (Choi et al., 2012). However, it is important to note that no single risk factor can fully predict later SAs
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015 Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
www.jonmd.com
507
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Wang et al.
(Hawton and van Heeringen, 2009; Logan et al., 2011), and we have yet to consistently identify the most reliable combinations of risk factors that accurately predict future attempts. Previous studies have identified persons with a mental illness who have made initial hospital contact as a robust and promising predictor for future suicidal behaviors (Sokero et al., 2003). Such evidence further bolsters the clinical importance of risk assessment in the ED because these people represent a cohort at substantially elevated risk of future death by suicide (Choi et al., 2012). The purpose of the current study was to test the ability of a variety of risk factors in predicting future SAs within 6 months of previous presentation to the ED. Previous studies in the area of risk assessment are bound by a number of limitations. First, most studies were conducted with restricted sample sizes. Second, many of the studies have been cross-sectional in nature, limiting their ability to generate causal inferences for studied risk factors. Perhaps most importantly, most of previous studies have examined existing suicide risk assessment scales and often examined outcomes other than future suicidal behavior (Bolton et al., 2012; Randall et al., 2011). Very few studies have empirically examined a range of clinical and sociodemographic candidate risk factors outside of a specific instrument, thus preventing the development of new, statistically derived risk assessment tools. In response to these limitations, we conducted a multisite study that collected a comprehensive list of potential risk factors among a large number of consecutive psychiatric referrals to the ED over 3 years. The primary objectives of the study were to examine the association between the collected risk factors and future SAs in a high-risk population and to develop a statistically derived risk assessment scale. A secondary objective compared the psychometric properties of this scale with the established SAD PERSONS scale (Patterson et al., 1983). On the basis of previous research, we hypothesized that although many potential risk factors will be individually associated with future SAs, only a limited number will independently account for the variance in future attempts. Finally, given the comprehensive list of candidate risk factors included in this study, we hypothesized that a statistically derived model would explain greater variance in predicting future SAs over the SAD PERSONS scale.
2005; Schneider et al., 2008; Sokero et al., 2003; Suominen et al., 2004; Tidemalm et al., 2008; Wang et al., 2012). The list of risk factors included in the study was not meant to be fully comprehensive, and for feasibility concerns, this list could capture information only on a limited number of risk factors. A physician with psychiatric training assessed each individual in the study. An attending psychiatrist supervised the assessment. The assessment included a comprehensive psychiatric interview (collection of demographic information, review of the presenting condition, assessment of psychiatric conditions, and review of previous psychiatric contact, medical history, and developmental issues). After assessment, the physician or senior medical student working with the physician completed the SAFE Database Study form. The form included individual candidate variables along with three clinician-assessed standardized scales: the SAD PERSONS scale (Tidemalm et al., 2008), the Modified SAD PERSONS scale (MSPS) (Gonzalez-Navarro et al., 2012), and the Columbia Classification Algorithm of Suicide Assessment (C-CASA; Cavanagh et al., 2003). The physicians involved in the study assessments received a yearly instructional seminar on how to complete the SAFE Database form.
Baseline Measures SAD PERSONS Scale
Ethical Approval
The 10-item scale is a mnemonic, with each letter corresponding to a potential risk factor for suicide: sex (male), age (>45 years), depression or hopelessness, previous attempts or psychiatric care, ethanol or substance abuse, rational thinking loss (psychosis), social supports lacking, organized plan or serious attempt, no spouse, and sickness (chronic pain or physical illness). Each item is scored as 1 if present, 0 if absent, based on the current presentation. Research showed that a modified version of the SAD PERSONS scale, which uses fewer items from the original scale, enhanced interrater reliability such that the concordance between psychiatrists and nonpsychiatrists on the outcome of the patients improved from 71% to 89% using the weighted MSPS. Therefore, we also tested the MSPS (Hockberger and Rothstein, 1988). Stated future suicidal intent was therefore recorded, as this replaces the “sickness” item on the SAD PERSONS. Appropriate weights were applied to calculate the MSPS. In this study, the individual scale items (as candidate predictors) were of interest rather than total scale scores.
The study was approved by the Research Ethics Board of the University of Manitoba.
Columbia Classification Algorithm of Suicide Assessment
METHODS
Setting Data came from the SAFE Database Study (Suicide Assessment Form in Emergency psychiatry), a large multisite study examining risk factors for suicide. Data were collected on psychiatric patients in the EDs of the two largest tertiary care hospitals in the province of Manitoba, Canada, in the city of Winnipeg. Psychiatric services for these two teaching hospitals are provided 24 hours daily by psychiatric residents and staff psychiatrists associated with the Department of Psychiatry at the University of Manitoba.
Study Population The study population included consecutive adult (>18 years) referrals to psychiatric services (N = 6919). There were no exclusionary criteria. The study period was 3 years (January 1, 2009, to December 31, 2011). This provided a recruitment time of 30 months, allowing for 6 months of follow-up for all participants who initially visited before July 1, 2011.
Baseline Patient Assessment The SAFE database collected information on a variety of risk factors based on existing literature (Bolton et al., 2008; Bolton and Robinson, 2010; Nordentoft et al., 2011; Sareen, 2011; Sareen et al., 508
www.jonmd.com
The C-CASA was developed as a standardized scale to classify suicidal behavior in eight mutually exclusive categories (Posner et al., 2007). It has been shown that the C-CASA has an excellent overall reliability, with a median intraclass correlation coefficient of 0.89 (Posner et al., 2007). This would be considered “almost perfect” by the Landis and Koch (1977) strength of agreement criteria. The classification of suicidal behavior is based on clinical judgment. A significant advantage of the scale is that it differentiates between SAs with intention to die from self-harm behaviors without intentions to die. Variables from the C-CASA examined as baseline predictors in this study included SAs, preparatory acts toward imminent suicidal behavior, and self-injurious behavior with no suicidal intent.
Other Baseline Measures The SAFE form also recorded the presence or absence of six additional candidate risk factors: childhood physical or sexual abuse (Hawton and van Heeringen, 2009; Ronquillo et al., 2012; Wang et al., 2012), current anxiety disorder (Bolton et al., 2008), an acute stressor (Wang et al., 2012), aggression or impulsivity (Hawton and van Heeringen, 2009; Wang et al., 2012), access to firearms (Crandall et al., 1997), and suicidal ideation (Sokero et al., 2003). Suicidal ideation was marked as present if the individual endorsed thoughts ranging from ambivalence about living to an active desire to die with a plan for suicide. © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Predictors of Future Suicide Attempts
analyses were also conducted to examine the association between the ROC-determined cutoff points and future SAs. The AUC scores of the three scales were directly compared using a z-test to determine if any of the scales were significantly better at predicting future attempts.
RESULTS Survival Analysis The survival curve revealed that of all patients who presented with future SA to the ED, half of these presentations occurred by day 78, and 66.5% of the presentations were within 6 months (Fig. 1).
Prevalence of Baseline Measures
FIGURE 1. Survival curve.
Outcome Measure The main outcome measure in the study was a future SA. Future SAs were defined as people who presented to one of the two examined EDs within the next 6 months, with an SA as assessed by C-CASA after their baseline presentation (with or without a SA). Recognizing that it would also be of interest to examine SA in a longer time frame, 6 months was chosen for this study based on its use in previous studies examining the prediction of future suicidal behavior (Cha et al., 2010; Nock et al., 2010; Olfson et al., 2012). This shorter time frame provides more information on short-term risk, which can, in turn, better inform treatment decisions at the time of clinical encounter.
Statistical Analysis Data were analyzed using SPSS 20.0 (IBM Corp, 2011). For individuals with multiple presentations, only the first was used to maintain the statistical assumption of independence of observation. A survival analysis was performed to examine time to future SA. For future SAs, analyses included only the records of people with a SA within 6 months of baseline presentation and records from the reference group (individuals with multiple visits but no future SA and individuals with only one presentation). Among individuals with a future SA, baseline measures from their ED presentation immediately before their SA were used. For individuals with multiple future SAs, only the first contributed to the analyses. Cross-tabulations yielded prevalence rates as well as missing rates for all baseline measures. Variables with high missing rates were excluded from further analyses. Binary logistic regression analyses were conducted to examine the association between future SAs and baseline risk factors. Backward stepwise logistic regression was used to determine the most parsimonious predictive model for future SAs by selecting baseline measures that independently explained the greatest proportion of variance in future SAs. All 19 baseline measures were entered into the backward regression. ROC curve analyses were conducted to determine optimum cutoff points for predicting future SA risk using the SAD PERSONS tool, the MSPS, as well as the model yielded by the stepwise regression. Each of these was scored such that the presence of an item on the scale was scored 1 and the absence of an item was scored 0. The sum of the scores for the items included on each of the scales was used for the ROC analysis. Area under the curve (AUC), as well as sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), was determined for the newly calculated cutoff points for each scale. AUC accuracy was interpreted as low (0.5 to 0.7), moderate (0.7 to 0.9), and high (>0.9) based on existing literature (Fischer et al., 2003). Binary logistic regression
Over the 3-year course of the study, there were 6919 referrals to psychiatric services in the ED. Our study included only patients who were identifiable with a correct personal health identifying number (the identifier used in the database), which yielded a total of 6555 referrals. After accounting for multiple referrals by the same individual, and removing initial referrals from the last 6 months of the study (because they could not be followed for 6 months), this left 3939 unique person baseline referrals. The prevalence rates of baseline measures for these referrals are presented in Table 1. Forty-eight percent of the patients were male, and 31.5% (n = 1242) of the baseline referrals featured suicidal ideation. A total of 179 referrals (4.5%) were classified as selfharm referrals without suicidal intent, and 561 (14.2%) were identified as having made a SA with intent to die. Of the 3939 baseline participants, 2.6% (n = 104) presented with a SA within the next 6 months (see Fig. 2).
Missing Data The rate of missing data for the six additional candidate risk factors ranged from just under 10% to close to 55%. The rate of missing data for the SAD PERSONS scale ranged from 1% to 6.8%. Childhood physical or sexual abuse, anxiety disorder, and access to firearms had the most concerning rates of missing data, which were 54.7%, 24.0%, and 39.0%, respectively. Hence, these three items were removed from further analyses. TABLE 1. Prevalence Rates of Candidate Risk Factors at Baseline Presentation Baseline Presentations (N = 3939), n (%) Acute stressor Aggression and impulsivity Suicidal ideation SAD PERSONS Sex (male) Age (>45 yrs) Depression or hopelessness Previous attempts or psychiatric care Ethanol or substance abuse Rational thinking loss (psychosis) Social support lacking Organized plan or serious attempt No spouse Sickness (chronic pain or physical illness) C-CASA Self-harm behavior, no suicidal intent Preparatory acts toward imminent suicidal behavior Suicide attempt
© 2015 Wolters Kluwer Health, Inc. All rights reserved. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
1917 (48.7) 1739 (44.1) 1242 (31.5) 1894 (48.1) 2195 (55.7) 1768 (44.9) 2364 (60.0) 1653 (42.0) 1192 (30.3) 1249 (31.7) 576 (14.6) 3408 (58.6) 955 (24.2) 179 (4.5) 127 (3.2) 561 (14.2)
www.jonmd.com
509
Wang et al.
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
FIGURE 2. Flowchart of the sample studied. Those with a suicide attempt within 6 months were compared with people who did not make a suicide attempt within 6 months.
Associations Between Baseline Measures and Future SA Presentations Table 2 shows the relationship between baseline measures and future SAs compared between individuals who made no known SA within 6 months and those who made an SA during follow-up. Only 9 of the examined 19 factors significantly predicted future SA. Suicidal ideation, previous attempts or psychiatric care, and SA were highly significant predictors of future SA with p < 0.001. Previous attempts or psychiatric care had an especially high odds ratio of 3.91 (95% confidence interval [CI], 2.22–6.89). Sex, depression and hopelessness, rational thinking loss (psychosis), and organized plan or serious attempt were also correlated with future attempt. Three of the items assessed on the SAD PERSONS scale were inversely associated with future SAs, including age older than 45 years, ethanol or substance abuse, and rational thinking loss (psychosis). The backward regression (n = 1433) revealed that of the initial 19 baseline measures, only 2 of these independently accounted for the variance in future SAs (and were thus retained in the model): suicidal ideation and previous attempts or psychiatric care. Their adjusted odds ratios were 1.95 (95% CI, 1.09–3.50; p < 0.05) and 5.54 (95% CI, 2.17–14.12; p < 0.001), respectively.
ROC Analysis The ROC analyses for SAD PERSONS, MSPS, and the stepwise regression model (Table 3) indicated that the two-item model yielded an AUC of 0.59 (95% CI, 0.54–0.64; p < 0.01), with an optimum cutoff point of 0. This model and its associated cutoff point offered the highest sensitivity (0.910), specificity (0.242), PPV (0.034), and NPV (0.989). Based on the clinical criteria proposed by Cichetti et al. (1995), whereby lower than 70% is poor, 70% to 79% is fair, 80% to 89% is good, and 90% and above is excellent, our results showed that sensitivity and NPV were excellent, whereas specificity and PPV were poor. Scores of 1 or 2 using the two-item model were associated with more than 3 times the likelihood of future SAs when compared with a score 510
www.jonmd.com
of 0. All scales showed low overall predictive ability, as evidenced by the AUC values ranging from 0.57 to 0.59. A comparison between the two-item model and the SAD PERSONS scale revealed that the two-item model’s AUC value was not significantly different from that of the SAD PERSONS and MSPS scales, with z-values of 0.60 and −0.05, respectively, based upon a one-tailed test. Of note, the ROC analysis selected lower risk cutoff points for the SAD PERSONS and MSPS scales than those that are characteristically used for the scales (categories for high risk of 7 to 10 and 9 to 14, respectively; Hockberger and Rothstein, 1988).
DISCUSSION Using a longitudinal design and robust statistical approaches, this study evaluated a comprehensive list of risk factors used by physicians at the ED to predict future SA within 6 months in a high-risk adult clinical population who presented to the ED and were referred to psychiatry. It is important to recognize that this study could assess future SA only among individuals who presented to the study hospitals; the control group of “no known suicide attempts” may have included people with SA who either died, did not seek care, or presented for care elsewhere. The results revealed that although many potential risk factors may be individually predictive, only two measures independently account for the variance in future attempts: suicidal ideation and previous attempts or psychiatric care. This study adds considerably to the literature in SA risk assessment, as it provides, for the first time, an extremely brief statistically derived clinical tool that assesses the likelihood of future SA among adults referred presenting to psychiatry in the ED, regardless of their reason for baseline presentation. The results from the survival analysis suggest that most patients who presented with a future SA did so within 6 months. This finding corroborates well with previous studies and further bolsters its clinical utility. The current results are very similar to the study of Horowitz et al., 2013, where a two-item assessment tool was effective as a screening measure for risks of suicide, although the focus of the study of Horowitz et al. was not in © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Predictors of Future Suicide Attempts
TABLE 2. Association Between Baseline Measures and Future Suicide Attempt Presentations No Known SA at 6 mos (n = 3835), n (%) SA at 6 mos (n = 104), n (%) Odds Ratio (95% CI)
Baseline Measures Acute stressor Aggression and impulsivity Suicidal ideation SAD PERSONS Sex (male) Age (>45 yrs) Depression or hopelessness Previous attempts or psychiatric care Ethanol or substance abuse Rational thinking loss (psychosis) Social support lacking Organized plan or serious attempt No spouse Sickness (chronic pain or physical illness) C-CASA Self-harm behavior, no suicidal intent Preparatory acts towards imminent suicidal behavior SA
1858 (55.4) 1690 (49.0) 1187 (35.3)
59 (59.6) 49 (50.0) 55 (56.7)
1.19 (0.79–1.78) 1.04 (0.70–1.56) 2.40 (1.59–3.60)***
1832 (48.3) 1554 (42.2) 1705 (47.7) 2274 (62.2) 1609 (45.0) 1174 (32.4) 1206 (33.7) 549 (15.4) 2243 (61.2) 924 (25.7)
62 (59.6) 39 (37.5) 63 (61.8) 90 (86.5) 45 (44.6) 18 (17.5) 43 (42.6) 27 (26.2) 65 (63.1) 31 (30.4)
1.58 (1.06–2.35)* 0.82 (0.55–1.23) 1.77 (1.18–2.65)** 3.91 (2.22–6.89)*** 0.98 (0.66–1.47) 0.44 (0.26–0.74)** 1.46 (0.77–2.17) 1.95 (1.25–3.06)** 1.09 (0.72–1.63) 1.26 (0.82–1.93)
178 (4.6) 123 (3.2) 533 (13.9)
– – 28 (26.9)
N/A N/A 2.28 (1.47–3.55)***
Bold items indicate those that are significant. –: Cell size fewer than 5, too small to perform regression. *p < 0.05. **p < 0.01. ***p < 0.001.
the ED. This is a meaningful contribution to the existing knowledge of risk assessment, which is based primarily on studies of numerous scales without evidence to support their use (Randall et al., 2011). Furthermore, this tool is clinically appealing as it would be easy and quick to conduct in clinical situations and is statistically superior to the longer and more time-consuming 10-item SAD PERSONS scale. However, our results also highlight the precision difficulties inherent in risk assessment. None of the examined instruments were able to accurately identify individuals who will make an attempt in the future, reinforcing the difficulty of suicide prediction and emphasizing the need for further study. It is important to observe that there were individual measures that were significantly associated with future SAs despite not being selected in the final model. Sex, depression and hopelessness, suicidal ideation, rational thinking loss, and having an organized plan or serious attempt did not significantly contribute to the final model in predicting future SA, despite being implicated as risk factors in previous studies (Bolton et al., 2010; Bruffaerts et al., 2010; Carli et al., 2010; Klonsky et al., 2012; O'Connor et al., 2012; Runeson et al., 2010). A possible reason may be that they are important factors in predicting completed suicide
rather than attempts. For example, in a study of 100 consecutive patients who had attempted suicide and were followed for 37 years, it was found that men are at greater risk for suicide, whereas women have higher rates of SAs (Suominen et al., 2004). Another possible explanation as to why these items were dropped for the final regression model is because they may be collinear with other items that accounted for a greater amount of variance. For example, depression was independently predictive of future SA, but its effects may have been accounted for by other variables such as suicidal ideation. Previous studies have repeatedly pointed toward depression as an important risk factor for both future SAs and suicide (Bolton et al., 2010; Harris and Barraclough, 1997; Hawton and van Heeringen, 2009). Therefore, despite depression not significantly contributing to the final model in this study, it is important not to disregard its association with suicidal behavior. Finally, the finding of psychosis being inversely related to future SA opposes several studies that show it to be a risk factor for suicide and suicidal behavior (Hor and Taylor, 2010; Siris, 2001). One explanation for this counterintuitive observation is that a presentation of psychosis may lead to more intensive care (i.e., hospitalization or case management) that alters their
TABLE 3. ROC Analysis for the Assessment of Future SA Presentations Within 6 Months Scale SAD PERSONS MSPS Two-item model
n
Cutoff Pointa
Odds Ratio (95% CI)b
AUC
Sensitivity
Specificity
PPV
NPV
3923 3224 3505
2 3 0
1.74 (0.95–3.19) 1.54 (0.93–2.57) 3.22 (1.62–6.42)**
0.57 (0.51, 0.62)* 0.59 (0.53, 0.65)** 0.59 (0.54, 0.64)**
0.885 0.802 0.910
0.185 0.276 0.242
0.029 0.033 0.034
0.983 0.978 0.989
Bold items indicate those that are significant. a Cutoff point is grouped in the low-risk category. For example, for SAD PERSONS, with a cutoff point of 2, low-risk scores are 0–2, and high-risk scores are 3–10. b Odds ratios show the association between high-risk scores and SAs, with the reference group being low-risk score. *p < 0.05. **p < 0.01.
© 2015 Wolters Kluwer Health, Inc. All rights reserved. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
www.jonmd.com
511
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Wang et al.
risk trajectory and protects against future attempts. Further study of this group is required to clarify this finding. The results of this study should be interpreted carefully while considering several limitations. First, missing data pose a major limitation to our data analysis; we could not examine certain variables because of high amounts of missing data. Thus, we were unable to determine the independent associations of these factors with future SA and how they may have influenced the other variables we examined. Although previous literature has repeatedly demonstrated a robust positive association between SAs and eventual death by suicide (Suominen et al., 2004), our findings cannot be extended to apply to the risk for suicide completion. In addition, those who died as a result of a SA were obviously not included in the examined sample yet represent an important group at risk. Given that some past research has indicated that suicide attempters and suicide completers are two very distinct groups of people who exhibit differential characteristics (DeJong et al., 2010; Fischer et al., 2003; Harris and Barraclough, 1997), it is possible, for example, that the items that did not contribute to our final model may actually predict eventual suicide. Nonetheless, SAs are recognized as one of the strongest predictors for future completed suicide (Beautrais, 2001), and thus, our findings may still contribute to the understanding of suicide risk. To confirm these possibilities, future studies are required to examine the association between these risk factors and suicide specifically. Another limitation pertains to the patients captured in the sample. It is important to recognize that patients who were seriously injured or had a more urgent physical health problem may have been directly transferred to surgical or intensive care units without being seen by the psychiatry department in the ED. Furthermore, as stated previously, the data collected were based on individuals who sought treatment at the ED at the two study hospitals, and it is likely that some people did not present to the ED after attempting suicide or may have presented to another ED in the city. However, our study attempted to reduce sampling bias by ensuring that every consecutive presentation was included and used data from the two main tertiary care hospitals in the province, which captures most emergency psychiatry presentations in the city of Winnipeg and nearby towns. Future studies are required to determine if these results remain consistent in hospital EDs located in other settings, and with potentially different populations selected for referral to psychiatry. Moreover, the reliability of the risk assessment was not directly measured in this study, and there is likely some degree of interrater differences; such measure was not available to us and an assessment of the interrater reliability is necessary. The use of standardized measures such as the C-CASA and the fact that physicians with psychiatric training conducted assessments both serve to partially address this concern, but it remains an important potential source of bias. We also recognize that certain items on the questionnaire may have been overly inclusive. For example, the item “previous attempts with psychiatric care” could entail psychiatric hospitalization and ongoing psychiatric outpatient treatment of various lengths. Finally, although the high NPV suggests that patients who undergo risk screening with a value of 0 on the proposed two-item model may be at minimal risk for future suicidal behaviors, there remain accuracy challenges with the low PPV. Further evaluation is required to determine if this two-item tool could help identify people requiring more comprehensive risk assessment referred to psychiatry in the ED.
CONCLUSIONS In sum, this study presents a two-item risk assessment model that could be used as a screening tool for future SA risk in a high-risk clinical population. This tool may help treatment providers determine which people require a more comprehensive assessment of SA risk. Limited PPVs suggest that prediction of future SAs is inaccurate, and a simple and concise checklist is unlikely to predict future SA on its own. However, our results suggest that the assessment of two-risk 512
www.jonmd.com
factors may serve as a useful clinical tool that could lead to more efficient health resource use. Future studies are needed to further explore these factors and possibly devise a better model for predicting future SAs and completed suicide. ACKNOWLEDGMENTS The authors acknowledge Daniel Palitsky, BSc, Cara Katz, BSc, and Will Husarewycz, BSc, for data entry. The authors also acknowledge the residents in the Department of Psychiatry at the University of Manitoba for their help with data collection. DISCLOSURES Author contributions: Dr Bolton and Ms Wang conceptualized the study. Ms Wang wrote the manuscript and analyzed the data. Ms Bhaskaran handled the data and assisted with data analysis. Dr Wang and Ms Spiwak provided statistical advice on study design. Dr Bolton and Dr Sareen supervised the writing of the manuscript, and all authors contributed substantially to its revision. Preparation of this article was supported by research grants from the Manitoba Health Research Council (Dr Bolton), Manitoba Health Research Council Chair Award (Dr Sareen), Canadian Institutes of Health Research New Investigator Awards (Dr Bolton, no. 113589; Dr Sareen, no. 152348), The Social Sciences and Humanities Research Council Joseph-Armand Bombardier Canada Graduate Scholarship (Ms Wang), The University of Manitoba Graduate Fellowship (Ms Bhaskaran), and the Manitoba Health Research Council Graduate Studentship (Ms Spiwak). The funding sources had no role in the design and conduct of the study; no role in the collection, management, analysis, and interpretation of data; and no role in the preparation, review, and approval of the manuscript. The authors declare no conflict of interest. REFERENCES Ballard ED, Horowitz LM, Jobes DA, Wanger BM, Pao M, Teach SJ (2013) Association of positive responses to suicide screening questions with hospital admission and repeated emergency department visits in children and adolescents. Pediatr Emerg Care. 29:1070–1074. Beautrais AL (2001) Suicides and serious suicide attempts: Two populations or one? Psychol Med. 31:837–845. Bolton JM, Cox BJ, Afifi TO, Enns MW, Bienvenu OJ, Sareen J (2008) Anxiety disorders and risk for suicide attempts: Findings from the Baltimore Epidemiologic Catchment area follow-up study. Depress Anxiety. 25:477–481. Bolton JM, Pagura J, Enns MW, Grant B, Sareen J (2010) A population-based longitudinal study of risk factors for suicide attempts in major depressive disorder. J Psychiatr Res. 44:817–826. Bolton JM, Robinson J (2010) Population-attributable fractions of axis I and axis II mental disorders for suicide attempts: Findings from a representative sample of the adult, noninstitutionalized US population. Am J Public Health. 100:2473–2480. Bolton JM, Spiwak R, Sareen J (2012) Predicting suicide attempts with the SAD PERSONS scale: A longitudinal analysis. J Clin Psychiatry. 73:e735–e741. Bruffaerts R, Demyttenaere K, Borges G, Haro JM, Chiu WT, Hwang I, Karam EG, Kessler RC, Sampson N, Alonso J, Andrade LH, Angermeyer M, Benjet C, Bromet E, de Girolamo G, de Graaf R, Florescu S, Gureje O, Horiguchi I, Hu C, Kovess V, Levinson D, Posada-Villa J, Sagar R, Scott K, Tsang A, Vassilev SM, William DR, Nock MK (2010) Childhood adversities as risk factors for onset and persistence of suicidal behavior. Br J Psychiatry. 197:20–27. Buzan RD, Weissberg MP (1991) Suicide: Risk factors and therapeutic considerations in the ED. J Emerg Med. 10:335–343. Carli V, Jovanovic N, Podlesek A, Roy A, Rihmer Z, Maggi S, Marusic D, Cesaro C, Marusic A, Sarchiapone M (2010) The role of impulsivity in self-mutilators, suicide ideators and suicide attempters—A study of 1265 male incarcerated individuals. J Affect Disord. 123:116–122. Caterino JM, Sullivan AF, Betz ME, Espinola JA, Miller I, Camargo CA, Boudreaux ED, Emergency Department Safety Assessment and Follow-up Evaluation (ED-SAFE)
© 2015 Wolters Kluwer Health, Inc. All rights reserved. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Predictors of Future Suicide Attempts
Investigators (2013) Evaluating current patterns of assessment for self-harm in emergency departments: A multicenter study. Acad Emerg Med. 20:807–815.
Typologies of suicidality based upon ambivalence about living and dying. Compr Psychiatry. 53:461–467.
Cavanagh J, Carson A, Sharpe M, Lawrie SM (2003) Psychological autopsy studies of suicide: A systematic review. Psychol Med. 33:395–405.
Olfson M, Marcus SC, Bridge JA (2012) Emergency treatment of deliberate self-harm. Arch Gen Psychiatry. 69:80–88.
Centers for Disease Control and Prevention (CDC) (2012) Injury prevention & control. National suicide statistics at a glance. Cited March 30, 2013. Retrieved from www.cdc.gov/violenceprevention/suicide/statistics. Accessed January 2015.
Olfson M, Marcus SC, Bridge JA (2013) Emergency department recognition of mental disorders and short-term outcome of deliberate self-harm. Am J Psychiatry. 170: 1442–1450.
Cha CB, Najmi S, Park JM, Finn CT, Nock MK (2010) Attentional bias toward suiciderelated stimuli predicts suicidal behavior. J Abnorm Psychol. 119:616–622.
Patterson WM, Dohn HH, Bird J, Patterson GA (1983) Evaluation of suicidal patients: The SAD PERSONS scale. Psychosomatics. 24:343–345, 348–349.
Choi JW, Park S, Yi KK, Hong JP (2012) Suicide mortality of suicide attempt patients discharged from emergency room, nonsuicidal psychiatric patients discharged from emergency room, admitted suicide attempt patients, and admitted nonsuicidal psychiatric patients. Suicide Life Threat Behav. 43:235–243.
Posner K, Oquendo MA, Gould M, Stanley B, Davies M (2007) Columbia Classification Algorithm of Suicide Assessment (C-CASA): Classification of suicidal events in the FDA’s pediatric suicidal risk analysis of antidepressants. Am J Psychiatry. 164:1035–1043.
Cichetti DV, Volkmar F, Klin A, Showalter D (1995) Diagnosing autism using ICD-10 criteria: A comparison of neural networks and standard multivariate procedures. Child Neuropsychol. 1:26–37.
Randall JR, Colman I, Rowe BH (2011) A systematic review of psychometric assessment of self-harm risk in the emergency department. J Affect Disord. 134:348–355.
Cooper JB, Lawlor MP, Hiroeh U, Kapur N, Appleby L (2003) Factors that influence emergency department doctors’ assessment of suicide risk in deliberate self-harm patients. Eur J Emerg Med. 10:283–287. Crandall C, Olson L, Fullerton L, Sklar Z, Zumwalt R (1997) Guns and knives in New Mexico: Patterns of penetrating trauma, 1978–1993. Acad Emerg Med. 4:263–267. DeJong TM, Overholser JC, Stockmeier CA (2010) Apples to oranges?: A direct comparison between suicide attempts and suicide completers. J Affect Disord. 124:90–97. Elisei S, Verdolini N, Anastasi S (2012) Suicidal attempts among emergency department patients: One-year of clinical experience. Psychiatr Danub. 24: S140–S142. Fischer JE, Bachman LM, Jaeschke R (2003) A readers’ guide to the interpretation of diagnostic test properties: Clinical example of sepsis. Intensive Care Med. 29: 1043–1051. Gonzalez-Navarro MD, Lorenzo-Roman MI, Luna-Maldonado A, Gomez-Zapata M, Imbernon-Pardo E, Ruiz-Riquelme J (2012) Study of sociodemographic and psychopathological risk factors in suicide attempts. 2008–2010. Semergen. 38:439–444. Harris EC, Barraclough B (1997) Suicide as an outcome for mental disorders. A metaanalysis. Br J Psychiatry. 170:205–228. Hawton K, van Heeringen K (2009) Suicide. Lancet. 373:1372–1381. Hockberger RS, Rothstein RJ (1988) Assessment of suicide potential by nonpsychiatrists using the SAD PERSONS score. J Emerg Med. 6:99–107. Hor K, Taylor M (2010) Suicide and schizophrenia: A systematic review of rates and risk factors. J Psychopharmacol. 4(Suppl.):81–90. Horowitz LM, Snyder D, Ludi E, Rosenstein DL, Kohn-Godbout J, Lee L, Cartledge T, Farrar A, Pao M (2013) Ask suicide-screening questions to everyone in medical settings: The asQ’em Quality Improvement Project. Psychosomatics. 54:239–247. IBM Corp (2011) IBM SPSS statistics for Windows, Version 20.0. Armonk, NY: IBM Corp. Klonsky ED, May AM, Glenn CR (2012) The relationship between nonsuicidal self-injury and attempted suicide: Converging evidence from four samples. J Abnormal Psychol. 122:231–237. Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics. 33:159–174.
Ronquillo L, Minassian A, Vilke GM, Wilson MP (2012) Literature-based recommendations for suicide assessment in the emergency department: A review. J Emerg Med. 43:836–842. Roos L, Sareen J, Bolton JM (2013) Suicide risk assessment tools, predictive validity and utility today: Time for a revamp? Neuropsychiatry. 3:483–495. Runeson B, Tidemalm D, Dahlin M, Lichtenstein P, Langstrom N (2010) Method of attempted suicide as predictor of subsequent successful suicide: National long term cohort study. BMJ. 341:c3222. Sareen J (2011) Anxiety disorders and risk for suicide: Why such controversy? Depress Anxiety. 28:941–945. Sareen J, Cox BJ, Afifi TO, de Graaf R, Asmundson GJ, ten Have M, Stein MB (2005) Anxiety disorders and risk for suicidal ideation and suicide attempts: A populationbased longitudinal study of adults. Arch Gen Psychiatry. 62:1249–1257. Schneider B, Schnabel A, Wetterling T, Bartusch B, Weber B, Georgi K (2008) How do personality disorders modify suicide risk? J Pers Disord. 22:233–245. Siris SG (2001) Suicide and schizophrenia. J Psychopharmacol. 15:127–135. Sokero TP, Melartin TK, Rytsala HJ, Leskela US, Lestela-Mielonen PS, Isometsa ET (2003) Suicidal ideation and attempts among psychiatric patients with major depressive disorder. J Clin Psychiatry. 64:1094–1100. Steeg S, Kapur N, Webb R, Applegate E, Stewart SL, Hawton K, Bergen H, Waters K, Cooper J (2012) The development of a population-level clinical screening tool for self-harm repetition and suicide: The ReACT Self-Harm Rule. Psychol Med. 42: 2383–2394. Suominen K, Isometsa E, Suokas J, Haukka J, Achte K, Lonngvist J (2004) Completed suicide after a suicide attempt: A 37-year follow-up study. Am J Psychiatry. 161: 562–563. Tidemalm D, Langstrom N, Lichtenstein P, Runeson B (2008) Risk of suicide after suicide attempt according to coexisting psychiatric disorder: Swedish cohort study with long term follow-up. BMJ. 337:1–6. Ting SA, Sullivan AF, Miller I, Espinola JA, Allen MH, Camargo CA, Boudreaux ED, Emergency Department Safety and Follow-up Evaluation (ED-SAFE) Investigators (2012) Multicenter study of predictors of suicide screening in emergency departments. Acad Emerg Med. 19:239–243.
Logan J, Hall J, Karch D (2011) Suicide categories by patterns of known risk factors. Arch Gen Psychiatry. 68:935–941.
US Department of Health and Human Services (HHS) Office of the Surgeon General and National Action Alliance for Suicide Prevention (2012, Sept.) National strategy for suicide prevention: Goals and objectives for action. Washington, DC: HHS.
Mitchell A, Garand D, Dean L, Panzak G, Taylor M (2005) Suicide assessment in hospital emergency departments: Implications for patient satisfaction and compliance. Top Emerg Med. 27:302–312.
Wang Y, Sareen J, Afifi T, Bolton S-L, Johnson EA, Bolton JM (2012) Recent stressful life events and suicide attempt: Results from a nationally representative sample. Psychiatr Ann. 42:101–108.
Nock MK, Park JM, Finn CT, Deliberto TL, Dour HJ, Banaji MR (2010) Measuring the suicidal mind: Implicit cognition predicts suicidal behavior. Psychol Sci. 21: 511–517.
Williams-Johnson J, Williams E, Gossel-Williams M, Sewell CA, Abel WD, Whitehorne-Smith PA (2012) Suicide attempt by self-poisoning: Characteristics of suicide attempters seen at the Emergency Room at the University Hospital of the West Indies. West Indian Med J. 61:526–531.
Nordentoft M, Mortensen PB, Pedersen CB (2011) Absolute risk of suicide after first hospital contact in mental disorder. Arch Gen Psychiatry. 68:1058–1064. O’Connor SS, Jobes DA, Yeargin MK, Fitzgerald ME, Rodriguez VM, Conrad AK, Linebrry TW (2012) A cross sectional investigation of the suicidal spectrum:
World Health Organization (WHO) (2010) Suicide prevention (SUPRE). Cited January 24, 2013. World Health Organization. Retrieved from www.who.int/ mental_health/prevention/suicide/suicideprevent/en. Accessed January 2015.
© 2015 Wolters Kluwer Health, Inc. All rights reserved. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
www.jonmd.com
513
Predictors of Future Suicide Attempts Among Individuals Referred to Psychiatric Services in the Emergency Department A Longitudinal Study Yunqiao Wang, MA,* Joanna Bhaskaran, MA,* Jitender Sareen, MD,*†‡ JianLi Wang, PhD,§k Rae Spiwak, MSc,‡ and James M. Bolton, MD*†‡
Abstract: This study examined which factors predict future suicide attempts (SAs) among people referred to psychiatric services in the emergency department (ED). It included consecutive adult (age >18 years) presentations (N = 6919) over a 3-year period to the two tertiary care hospitals in Manitoba, Canada. Medical professionals assessed each individual on 19 candidate risk factors. Stepwise logistic regression and receiver operating characteristic curves examined the association between the baseline variables and future SAs within the next 6 months. A total of 104 individuals re-presented to the ED with future SAs. Of the 19 baseline variables, only two independently accounted for the variance in future attempts. High-risk scores using this two-item model were associated with elevated odds of future SA (odds ratio, 3.22; 95% confidence interval, 1.62–6.42; p < 0.01), but this was tempered by a low positive predictive value. Further evaluation is required to determine if this two-item tool could help identify people requiring more comprehensive risk assessment referred to psychiatry in the ED. Key Words: Suicide attempt, suicidal ideation, emergency department, risk factors, assessment, SAD PERSONS (J Nerv Ment Dis 2015;203: 507–513)
S
uicide is a leading cause of death worldwide (Centers for Disease Control and Prevention, 2012). Suicide attempts (SAs) are believed to be 20 times more frequent than completed suicide, and in the United States there are 2.5 million SAs each year (US Department of Health and Human Services (HHS) Office of the Surgeon General and National Action Alliance for Suicide Prevention, 2012; World Health Organization, 2010). SAs are among the strongest risk factors for suicide completion (Nordentoft et al., 2011; Sokero et al., 2003), and those who attempt suicide exhibit great similarities with individuals who die by suicide (Suominen et al., 2004). Every year, there are thousands of patients who visit the emergency department (ED) with suicidal ideation and SA as their major complaints (Olfson et al., 2012). Presenting to the ED with self-harm behavior increases an individual’s risk of suicide 6-fold (Crandall et al., 1997), and those who visited the ED for deliberate self-harm have shown to be at higher short-term risk for revisiting the ED with the same complaint (Olfson et al., 2013; Steeg et al., 2012). The recently released US Surgeon General’s 2012 National Strategy for Suicide Prevention highlights specific goals of evaluating research on screening of high-risk individuals and emphasizes the ED as a key clinical area to managing suicide risk (US Department of Health and Human Services (HHS) Office of the Surgeon General and National Action Alliance for Suicide Prevention, 2012). As such,
Departments of *Psychology, †Psychiatry, and ‡Community Health Sciences, University of Manitoba, Winnipeg, Manitoba; and Departments of §Psychiatry and kCommunity Health Sciences, University of Calgary, Calgary, Alberta, Canada. Send reprint requests to James M. Bolton, MD, PZ430-771 Bannatyne Avenue, Winnipeg, Manitoba, Canada R3E 3N4. E-mail: [email protected]. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved. ISSN: 0022-3018/15/20307–0507 DOI: 10.1097/NMD.0000000000000320
the ED is a critical component of the broader public health priority to reduce the rate of suicide. Most of existing risk assessment scales used in the ED are unable to predict future suicidal behaviors (Bolton et al., 2012; Randall et al., 2011). For example, Bolton et al. used the receiver operating characteristic (ROC) curves in a sample of 4019 ED presentations and demonstrated that the SAD PERSONS scale, a widely used instrument in assessing one’s risk of suicide, was no better than chance in predicting future SAs (Bolton et al., 2012). This often leaves many physicians doubting the results of their assessments and wondering what factors would dependably contribute to future risk (Ronquillo et al., 2012). Furthermore, the systematic review of Randall et al., which examined 12 studies on the predictability of future suicidal behavior, revealed that although these instruments have impressive psychometric properties, there is very little clinical utility. Unfortunately, the study did not perform a meta-analysis, and the presence of publication or selection biases is unknown. A recent study showed improved accuracy using a statistically derived tool among people presenting with self-harm (Steeg et al., 2012). Several other studies have found that a positive SA screening predicts future ED presentations among various age groups in a number of settings (Ballard et al., 2013; Ting et al., 2012). Ting et al. specifically found that alcohol and other substance use was the most significant predictor for future SA and noted that many assessment tools do not assess this issue. However, the screening process may vary substantially between EDs and many self-harm and suicidal presentations are undetected (Caterino et al., 2013). These authors commented that older adults were especially more likely to be neglected. A review of suicide risk assessment tools revealed that, in general, there is a lack of evidence in supporting the use of traditional risk assessment tools, and most of the existing tools need further investigation and replication (Roos et al., 2013). More of these studies are required, not only among people presenting with self-harm but also including the broader group of ED psychiatric patients, because mental illness confers a heightened risk of suicide even in the absence of SA, and many suicides occur without previous attempt (Nordentoft et al., 2011). Because of the already backed-up and overcrowded ED, extensive suicide risk assessments have limited feasibility, and therefore, our current situation calls for further research to devise a quick and reliable evaluative instrument (Mitchell et al., 2005). So far, some of the most common risk factors identified in the literature include history of suicidal behavior (Nordentoft et al., 2011; Sokero et al., 2003; Suominen et al., 2004), psychiatric disorders (Bolton et al., 2008; Bolton and Robinson, 2010; Sareen, 2011; Sareen et al., 2005; Schneider et al., 2008; Tidemalm et al., 2008), and stressful life events (Wang et al., 2012), among many others. In the emergency setting specifically, factors such as access to means (Williams-Johnson et al., 2012), depression (Buzan and Weissberg, 1991), having suicidal plans (Cooper et al., 2003), and the female sex (Elisei et al., 2012) are correlated with an elevated risk for making an SA. Depression and a previous SA are among the strongest predictors for suicide completion (Choi et al., 2012). However, it is important to note that no single risk factor can fully predict later SAs
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015 Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
www.jonmd.com
507
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Wang et al.
(Hawton and van Heeringen, 2009; Logan et al., 2011), and we have yet to consistently identify the most reliable combinations of risk factors that accurately predict future attempts. Previous studies have identified persons with a mental illness who have made initial hospital contact as a robust and promising predictor for future suicidal behaviors (Sokero et al., 2003). Such evidence further bolsters the clinical importance of risk assessment in the ED because these people represent a cohort at substantially elevated risk of future death by suicide (Choi et al., 2012). The purpose of the current study was to test the ability of a variety of risk factors in predicting future SAs within 6 months of previous presentation to the ED. Previous studies in the area of risk assessment are bound by a number of limitations. First, most studies were conducted with restricted sample sizes. Second, many of the studies have been cross-sectional in nature, limiting their ability to generate causal inferences for studied risk factors. Perhaps most importantly, most of previous studies have examined existing suicide risk assessment scales and often examined outcomes other than future suicidal behavior (Bolton et al., 2012; Randall et al., 2011). Very few studies have empirically examined a range of clinical and sociodemographic candidate risk factors outside of a specific instrument, thus preventing the development of new, statistically derived risk assessment tools. In response to these limitations, we conducted a multisite study that collected a comprehensive list of potential risk factors among a large number of consecutive psychiatric referrals to the ED over 3 years. The primary objectives of the study were to examine the association between the collected risk factors and future SAs in a high-risk population and to develop a statistically derived risk assessment scale. A secondary objective compared the psychometric properties of this scale with the established SAD PERSONS scale (Patterson et al., 1983). On the basis of previous research, we hypothesized that although many potential risk factors will be individually associated with future SAs, only a limited number will independently account for the variance in future attempts. Finally, given the comprehensive list of candidate risk factors included in this study, we hypothesized that a statistically derived model would explain greater variance in predicting future SAs over the SAD PERSONS scale.
2005; Schneider et al., 2008; Sokero et al., 2003; Suominen et al., 2004; Tidemalm et al., 2008; Wang et al., 2012). The list of risk factors included in the study was not meant to be fully comprehensive, and for feasibility concerns, this list could capture information only on a limited number of risk factors. A physician with psychiatric training assessed each individual in the study. An attending psychiatrist supervised the assessment. The assessment included a comprehensive psychiatric interview (collection of demographic information, review of the presenting condition, assessment of psychiatric conditions, and review of previous psychiatric contact, medical history, and developmental issues). After assessment, the physician or senior medical student working with the physician completed the SAFE Database Study form. The form included individual candidate variables along with three clinician-assessed standardized scales: the SAD PERSONS scale (Tidemalm et al., 2008), the Modified SAD PERSONS scale (MSPS) (Gonzalez-Navarro et al., 2012), and the Columbia Classification Algorithm of Suicide Assessment (C-CASA; Cavanagh et al., 2003). The physicians involved in the study assessments received a yearly instructional seminar on how to complete the SAFE Database form.
Baseline Measures SAD PERSONS Scale
Ethical Approval
The 10-item scale is a mnemonic, with each letter corresponding to a potential risk factor for suicide: sex (male), age (>45 years), depression or hopelessness, previous attempts or psychiatric care, ethanol or substance abuse, rational thinking loss (psychosis), social supports lacking, organized plan or serious attempt, no spouse, and sickness (chronic pain or physical illness). Each item is scored as 1 if present, 0 if absent, based on the current presentation. Research showed that a modified version of the SAD PERSONS scale, which uses fewer items from the original scale, enhanced interrater reliability such that the concordance between psychiatrists and nonpsychiatrists on the outcome of the patients improved from 71% to 89% using the weighted MSPS. Therefore, we also tested the MSPS (Hockberger and Rothstein, 1988). Stated future suicidal intent was therefore recorded, as this replaces the “sickness” item on the SAD PERSONS. Appropriate weights were applied to calculate the MSPS. In this study, the individual scale items (as candidate predictors) were of interest rather than total scale scores.
The study was approved by the Research Ethics Board of the University of Manitoba.
Columbia Classification Algorithm of Suicide Assessment
METHODS
Setting Data came from the SAFE Database Study (Suicide Assessment Form in Emergency psychiatry), a large multisite study examining risk factors for suicide. Data were collected on psychiatric patients in the EDs of the two largest tertiary care hospitals in the province of Manitoba, Canada, in the city of Winnipeg. Psychiatric services for these two teaching hospitals are provided 24 hours daily by psychiatric residents and staff psychiatrists associated with the Department of Psychiatry at the University of Manitoba.
Study Population The study population included consecutive adult (>18 years) referrals to psychiatric services (N = 6919). There were no exclusionary criteria. The study period was 3 years (January 1, 2009, to December 31, 2011). This provided a recruitment time of 30 months, allowing for 6 months of follow-up for all participants who initially visited before July 1, 2011.
Baseline Patient Assessment The SAFE database collected information on a variety of risk factors based on existing literature (Bolton et al., 2008; Bolton and Robinson, 2010; Nordentoft et al., 2011; Sareen, 2011; Sareen et al., 508
www.jonmd.com
The C-CASA was developed as a standardized scale to classify suicidal behavior in eight mutually exclusive categories (Posner et al., 2007). It has been shown that the C-CASA has an excellent overall reliability, with a median intraclass correlation coefficient of 0.89 (Posner et al., 2007). This would be considered “almost perfect” by the Landis and Koch (1977) strength of agreement criteria. The classification of suicidal behavior is based on clinical judgment. A significant advantage of the scale is that it differentiates between SAs with intention to die from self-harm behaviors without intentions to die. Variables from the C-CASA examined as baseline predictors in this study included SAs, preparatory acts toward imminent suicidal behavior, and self-injurious behavior with no suicidal intent.
Other Baseline Measures The SAFE form also recorded the presence or absence of six additional candidate risk factors: childhood physical or sexual abuse (Hawton and van Heeringen, 2009; Ronquillo et al., 2012; Wang et al., 2012), current anxiety disorder (Bolton et al., 2008), an acute stressor (Wang et al., 2012), aggression or impulsivity (Hawton and van Heeringen, 2009; Wang et al., 2012), access to firearms (Crandall et al., 1997), and suicidal ideation (Sokero et al., 2003). Suicidal ideation was marked as present if the individual endorsed thoughts ranging from ambivalence about living to an active desire to die with a plan for suicide. © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Predictors of Future Suicide Attempts
analyses were also conducted to examine the association between the ROC-determined cutoff points and future SAs. The AUC scores of the three scales were directly compared using a z-test to determine if any of the scales were significantly better at predicting future attempts.
RESULTS Survival Analysis The survival curve revealed that of all patients who presented with future SA to the ED, half of these presentations occurred by day 78, and 66.5% of the presentations were within 6 months (Fig. 1).
Prevalence of Baseline Measures
FIGURE 1. Survival curve.
Outcome Measure The main outcome measure in the study was a future SA. Future SAs were defined as people who presented to one of the two examined EDs within the next 6 months, with an SA as assessed by C-CASA after their baseline presentation (with or without a SA). Recognizing that it would also be of interest to examine SA in a longer time frame, 6 months was chosen for this study based on its use in previous studies examining the prediction of future suicidal behavior (Cha et al., 2010; Nock et al., 2010; Olfson et al., 2012). This shorter time frame provides more information on short-term risk, which can, in turn, better inform treatment decisions at the time of clinical encounter.
Statistical Analysis Data were analyzed using SPSS 20.0 (IBM Corp, 2011). For individuals with multiple presentations, only the first was used to maintain the statistical assumption of independence of observation. A survival analysis was performed to examine time to future SA. For future SAs, analyses included only the records of people with a SA within 6 months of baseline presentation and records from the reference group (individuals with multiple visits but no future SA and individuals with only one presentation). Among individuals with a future SA, baseline measures from their ED presentation immediately before their SA were used. For individuals with multiple future SAs, only the first contributed to the analyses. Cross-tabulations yielded prevalence rates as well as missing rates for all baseline measures. Variables with high missing rates were excluded from further analyses. Binary logistic regression analyses were conducted to examine the association between future SAs and baseline risk factors. Backward stepwise logistic regression was used to determine the most parsimonious predictive model for future SAs by selecting baseline measures that independently explained the greatest proportion of variance in future SAs. All 19 baseline measures were entered into the backward regression. ROC curve analyses were conducted to determine optimum cutoff points for predicting future SA risk using the SAD PERSONS tool, the MSPS, as well as the model yielded by the stepwise regression. Each of these was scored such that the presence of an item on the scale was scored 1 and the absence of an item was scored 0. The sum of the scores for the items included on each of the scales was used for the ROC analysis. Area under the curve (AUC), as well as sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV), was determined for the newly calculated cutoff points for each scale. AUC accuracy was interpreted as low (0.5 to 0.7), moderate (0.7 to 0.9), and high (>0.9) based on existing literature (Fischer et al., 2003). Binary logistic regression
Over the 3-year course of the study, there were 6919 referrals to psychiatric services in the ED. Our study included only patients who were identifiable with a correct personal health identifying number (the identifier used in the database), which yielded a total of 6555 referrals. After accounting for multiple referrals by the same individual, and removing initial referrals from the last 6 months of the study (because they could not be followed for 6 months), this left 3939 unique person baseline referrals. The prevalence rates of baseline measures for these referrals are presented in Table 1. Forty-eight percent of the patients were male, and 31.5% (n = 1242) of the baseline referrals featured suicidal ideation. A total of 179 referrals (4.5%) were classified as selfharm referrals without suicidal intent, and 561 (14.2%) were identified as having made a SA with intent to die. Of the 3939 baseline participants, 2.6% (n = 104) presented with a SA within the next 6 months (see Fig. 2).
Missing Data The rate of missing data for the six additional candidate risk factors ranged from just under 10% to close to 55%. The rate of missing data for the SAD PERSONS scale ranged from 1% to 6.8%. Childhood physical or sexual abuse, anxiety disorder, and access to firearms had the most concerning rates of missing data, which were 54.7%, 24.0%, and 39.0%, respectively. Hence, these three items were removed from further analyses. TABLE 1. Prevalence Rates of Candidate Risk Factors at Baseline Presentation Baseline Presentations (N = 3939), n (%) Acute stressor Aggression and impulsivity Suicidal ideation SAD PERSONS Sex (male) Age (>45 yrs) Depression or hopelessness Previous attempts or psychiatric care Ethanol or substance abuse Rational thinking loss (psychosis) Social support lacking Organized plan or serious attempt No spouse Sickness (chronic pain or physical illness) C-CASA Self-harm behavior, no suicidal intent Preparatory acts toward imminent suicidal behavior Suicide attempt
© 2015 Wolters Kluwer Health, Inc. All rights reserved. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
1917 (48.7) 1739 (44.1) 1242 (31.5) 1894 (48.1) 2195 (55.7) 1768 (44.9) 2364 (60.0) 1653 (42.0) 1192 (30.3) 1249 (31.7) 576 (14.6) 3408 (58.6) 955 (24.2) 179 (4.5) 127 (3.2) 561 (14.2)
www.jonmd.com
509
Wang et al.
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
FIGURE 2. Flowchart of the sample studied. Those with a suicide attempt within 6 months were compared with people who did not make a suicide attempt within 6 months.
Associations Between Baseline Measures and Future SA Presentations Table 2 shows the relationship between baseline measures and future SAs compared between individuals who made no known SA within 6 months and those who made an SA during follow-up. Only 9 of the examined 19 factors significantly predicted future SA. Suicidal ideation, previous attempts or psychiatric care, and SA were highly significant predictors of future SA with p < 0.001. Previous attempts or psychiatric care had an especially high odds ratio of 3.91 (95% confidence interval [CI], 2.22–6.89). Sex, depression and hopelessness, rational thinking loss (psychosis), and organized plan or serious attempt were also correlated with future attempt. Three of the items assessed on the SAD PERSONS scale were inversely associated with future SAs, including age older than 45 years, ethanol or substance abuse, and rational thinking loss (psychosis). The backward regression (n = 1433) revealed that of the initial 19 baseline measures, only 2 of these independently accounted for the variance in future SAs (and were thus retained in the model): suicidal ideation and previous attempts or psychiatric care. Their adjusted odds ratios were 1.95 (95% CI, 1.09–3.50; p < 0.05) and 5.54 (95% CI, 2.17–14.12; p < 0.001), respectively.
ROC Analysis The ROC analyses for SAD PERSONS, MSPS, and the stepwise regression model (Table 3) indicated that the two-item model yielded an AUC of 0.59 (95% CI, 0.54–0.64; p < 0.01), with an optimum cutoff point of 0. This model and its associated cutoff point offered the highest sensitivity (0.910), specificity (0.242), PPV (0.034), and NPV (0.989). Based on the clinical criteria proposed by Cichetti et al. (1995), whereby lower than 70% is poor, 70% to 79% is fair, 80% to 89% is good, and 90% and above is excellent, our results showed that sensitivity and NPV were excellent, whereas specificity and PPV were poor. Scores of 1 or 2 using the two-item model were associated with more than 3 times the likelihood of future SAs when compared with a score 510
www.jonmd.com
of 0. All scales showed low overall predictive ability, as evidenced by the AUC values ranging from 0.57 to 0.59. A comparison between the two-item model and the SAD PERSONS scale revealed that the two-item model’s AUC value was not significantly different from that of the SAD PERSONS and MSPS scales, with z-values of 0.60 and −0.05, respectively, based upon a one-tailed test. Of note, the ROC analysis selected lower risk cutoff points for the SAD PERSONS and MSPS scales than those that are characteristically used for the scales (categories for high risk of 7 to 10 and 9 to 14, respectively; Hockberger and Rothstein, 1988).
DISCUSSION Using a longitudinal design and robust statistical approaches, this study evaluated a comprehensive list of risk factors used by physicians at the ED to predict future SA within 6 months in a high-risk adult clinical population who presented to the ED and were referred to psychiatry. It is important to recognize that this study could assess future SA only among individuals who presented to the study hospitals; the control group of “no known suicide attempts” may have included people with SA who either died, did not seek care, or presented for care elsewhere. The results revealed that although many potential risk factors may be individually predictive, only two measures independently account for the variance in future attempts: suicidal ideation and previous attempts or psychiatric care. This study adds considerably to the literature in SA risk assessment, as it provides, for the first time, an extremely brief statistically derived clinical tool that assesses the likelihood of future SA among adults referred presenting to psychiatry in the ED, regardless of their reason for baseline presentation. The results from the survival analysis suggest that most patients who presented with a future SA did so within 6 months. This finding corroborates well with previous studies and further bolsters its clinical utility. The current results are very similar to the study of Horowitz et al., 2013, where a two-item assessment tool was effective as a screening measure for risks of suicide, although the focus of the study of Horowitz et al. was not in © 2015 Wolters Kluwer Health, Inc. All rights reserved.
Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Predictors of Future Suicide Attempts
TABLE 2. Association Between Baseline Measures and Future Suicide Attempt Presentations No Known SA at 6 mos (n = 3835), n (%) SA at 6 mos (n = 104), n (%) Odds Ratio (95% CI)
Baseline Measures Acute stressor Aggression and impulsivity Suicidal ideation SAD PERSONS Sex (male) Age (>45 yrs) Depression or hopelessness Previous attempts or psychiatric care Ethanol or substance abuse Rational thinking loss (psychosis) Social support lacking Organized plan or serious attempt No spouse Sickness (chronic pain or physical illness) C-CASA Self-harm behavior, no suicidal intent Preparatory acts towards imminent suicidal behavior SA
1858 (55.4) 1690 (49.0) 1187 (35.3)
59 (59.6) 49 (50.0) 55 (56.7)
1.19 (0.79–1.78) 1.04 (0.70–1.56) 2.40 (1.59–3.60)***
1832 (48.3) 1554 (42.2) 1705 (47.7) 2274 (62.2) 1609 (45.0) 1174 (32.4) 1206 (33.7) 549 (15.4) 2243 (61.2) 924 (25.7)
62 (59.6) 39 (37.5) 63 (61.8) 90 (86.5) 45 (44.6) 18 (17.5) 43 (42.6) 27 (26.2) 65 (63.1) 31 (30.4)
1.58 (1.06–2.35)* 0.82 (0.55–1.23) 1.77 (1.18–2.65)** 3.91 (2.22–6.89)*** 0.98 (0.66–1.47) 0.44 (0.26–0.74)** 1.46 (0.77–2.17) 1.95 (1.25–3.06)** 1.09 (0.72–1.63) 1.26 (0.82–1.93)
178 (4.6) 123 (3.2) 533 (13.9)
– – 28 (26.9)
N/A N/A 2.28 (1.47–3.55)***
Bold items indicate those that are significant. –: Cell size fewer than 5, too small to perform regression. *p < 0.05. **p < 0.01. ***p < 0.001.
the ED. This is a meaningful contribution to the existing knowledge of risk assessment, which is based primarily on studies of numerous scales without evidence to support their use (Randall et al., 2011). Furthermore, this tool is clinically appealing as it would be easy and quick to conduct in clinical situations and is statistically superior to the longer and more time-consuming 10-item SAD PERSONS scale. However, our results also highlight the precision difficulties inherent in risk assessment. None of the examined instruments were able to accurately identify individuals who will make an attempt in the future, reinforcing the difficulty of suicide prediction and emphasizing the need for further study. It is important to observe that there were individual measures that were significantly associated with future SAs despite not being selected in the final model. Sex, depression and hopelessness, suicidal ideation, rational thinking loss, and having an organized plan or serious attempt did not significantly contribute to the final model in predicting future SA, despite being implicated as risk factors in previous studies (Bolton et al., 2010; Bruffaerts et al., 2010; Carli et al., 2010; Klonsky et al., 2012; O'Connor et al., 2012; Runeson et al., 2010). A possible reason may be that they are important factors in predicting completed suicide
rather than attempts. For example, in a study of 100 consecutive patients who had attempted suicide and were followed for 37 years, it was found that men are at greater risk for suicide, whereas women have higher rates of SAs (Suominen et al., 2004). Another possible explanation as to why these items were dropped for the final regression model is because they may be collinear with other items that accounted for a greater amount of variance. For example, depression was independently predictive of future SA, but its effects may have been accounted for by other variables such as suicidal ideation. Previous studies have repeatedly pointed toward depression as an important risk factor for both future SAs and suicide (Bolton et al., 2010; Harris and Barraclough, 1997; Hawton and van Heeringen, 2009). Therefore, despite depression not significantly contributing to the final model in this study, it is important not to disregard its association with suicidal behavior. Finally, the finding of psychosis being inversely related to future SA opposes several studies that show it to be a risk factor for suicide and suicidal behavior (Hor and Taylor, 2010; Siris, 2001). One explanation for this counterintuitive observation is that a presentation of psychosis may lead to more intensive care (i.e., hospitalization or case management) that alters their
TABLE 3. ROC Analysis for the Assessment of Future SA Presentations Within 6 Months Scale SAD PERSONS MSPS Two-item model
n
Cutoff Pointa
Odds Ratio (95% CI)b
AUC
Sensitivity
Specificity
PPV
NPV
3923 3224 3505
2 3 0
1.74 (0.95–3.19) 1.54 (0.93–2.57) 3.22 (1.62–6.42)**
0.57 (0.51, 0.62)* 0.59 (0.53, 0.65)** 0.59 (0.54, 0.64)**
0.885 0.802 0.910
0.185 0.276 0.242
0.029 0.033 0.034
0.983 0.978 0.989
Bold items indicate those that are significant. a Cutoff point is grouped in the low-risk category. For example, for SAD PERSONS, with a cutoff point of 2, low-risk scores are 0–2, and high-risk scores are 3–10. b Odds ratios show the association between high-risk scores and SAs, with the reference group being low-risk score. *p < 0.05. **p < 0.01.
© 2015 Wolters Kluwer Health, Inc. All rights reserved. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
www.jonmd.com
511
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Wang et al.
risk trajectory and protects against future attempts. Further study of this group is required to clarify this finding. The results of this study should be interpreted carefully while considering several limitations. First, missing data pose a major limitation to our data analysis; we could not examine certain variables because of high amounts of missing data. Thus, we were unable to determine the independent associations of these factors with future SA and how they may have influenced the other variables we examined. Although previous literature has repeatedly demonstrated a robust positive association between SAs and eventual death by suicide (Suominen et al., 2004), our findings cannot be extended to apply to the risk for suicide completion. In addition, those who died as a result of a SA were obviously not included in the examined sample yet represent an important group at risk. Given that some past research has indicated that suicide attempters and suicide completers are two very distinct groups of people who exhibit differential characteristics (DeJong et al., 2010; Fischer et al., 2003; Harris and Barraclough, 1997), it is possible, for example, that the items that did not contribute to our final model may actually predict eventual suicide. Nonetheless, SAs are recognized as one of the strongest predictors for future completed suicide (Beautrais, 2001), and thus, our findings may still contribute to the understanding of suicide risk. To confirm these possibilities, future studies are required to examine the association between these risk factors and suicide specifically. Another limitation pertains to the patients captured in the sample. It is important to recognize that patients who were seriously injured or had a more urgent physical health problem may have been directly transferred to surgical or intensive care units without being seen by the psychiatry department in the ED. Furthermore, as stated previously, the data collected were based on individuals who sought treatment at the ED at the two study hospitals, and it is likely that some people did not present to the ED after attempting suicide or may have presented to another ED in the city. However, our study attempted to reduce sampling bias by ensuring that every consecutive presentation was included and used data from the two main tertiary care hospitals in the province, which captures most emergency psychiatry presentations in the city of Winnipeg and nearby towns. Future studies are required to determine if these results remain consistent in hospital EDs located in other settings, and with potentially different populations selected for referral to psychiatry. Moreover, the reliability of the risk assessment was not directly measured in this study, and there is likely some degree of interrater differences; such measure was not available to us and an assessment of the interrater reliability is necessary. The use of standardized measures such as the C-CASA and the fact that physicians with psychiatric training conducted assessments both serve to partially address this concern, but it remains an important potential source of bias. We also recognize that certain items on the questionnaire may have been overly inclusive. For example, the item “previous attempts with psychiatric care” could entail psychiatric hospitalization and ongoing psychiatric outpatient treatment of various lengths. Finally, although the high NPV suggests that patients who undergo risk screening with a value of 0 on the proposed two-item model may be at minimal risk for future suicidal behaviors, there remain accuracy challenges with the low PPV. Further evaluation is required to determine if this two-item tool could help identify people requiring more comprehensive risk assessment referred to psychiatry in the ED.
CONCLUSIONS In sum, this study presents a two-item risk assessment model that could be used as a screening tool for future SA risk in a high-risk clinical population. This tool may help treatment providers determine which people require a more comprehensive assessment of SA risk. Limited PPVs suggest that prediction of future SAs is inaccurate, and a simple and concise checklist is unlikely to predict future SA on its own. However, our results suggest that the assessment of two-risk 512
www.jonmd.com
factors may serve as a useful clinical tool that could lead to more efficient health resource use. Future studies are needed to further explore these factors and possibly devise a better model for predicting future SAs and completed suicide. ACKNOWLEDGMENTS The authors acknowledge Daniel Palitsky, BSc, Cara Katz, BSc, and Will Husarewycz, BSc, for data entry. The authors also acknowledge the residents in the Department of Psychiatry at the University of Manitoba for their help with data collection. DISCLOSURES Author contributions: Dr Bolton and Ms Wang conceptualized the study. Ms Wang wrote the manuscript and analyzed the data. Ms Bhaskaran handled the data and assisted with data analysis. Dr Wang and Ms Spiwak provided statistical advice on study design. Dr Bolton and Dr Sareen supervised the writing of the manuscript, and all authors contributed substantially to its revision. Preparation of this article was supported by research grants from the Manitoba Health Research Council (Dr Bolton), Manitoba Health Research Council Chair Award (Dr Sareen), Canadian Institutes of Health Research New Investigator Awards (Dr Bolton, no. 113589; Dr Sareen, no. 152348), The Social Sciences and Humanities Research Council Joseph-Armand Bombardier Canada Graduate Scholarship (Ms Wang), The University of Manitoba Graduate Fellowship (Ms Bhaskaran), and the Manitoba Health Research Council Graduate Studentship (Ms Spiwak). The funding sources had no role in the design and conduct of the study; no role in the collection, management, analysis, and interpretation of data; and no role in the preparation, review, and approval of the manuscript. The authors declare no conflict of interest. REFERENCES Ballard ED, Horowitz LM, Jobes DA, Wanger BM, Pao M, Teach SJ (2013) Association of positive responses to suicide screening questions with hospital admission and repeated emergency department visits in children and adolescents. Pediatr Emerg Care. 29:1070–1074. Beautrais AL (2001) Suicides and serious suicide attempts: Two populations or one? Psychol Med. 31:837–845. Bolton JM, Cox BJ, Afifi TO, Enns MW, Bienvenu OJ, Sareen J (2008) Anxiety disorders and risk for suicide attempts: Findings from the Baltimore Epidemiologic Catchment area follow-up study. Depress Anxiety. 25:477–481. Bolton JM, Pagura J, Enns MW, Grant B, Sareen J (2010) A population-based longitudinal study of risk factors for suicide attempts in major depressive disorder. J Psychiatr Res. 44:817–826. Bolton JM, Robinson J (2010) Population-attributable fractions of axis I and axis II mental disorders for suicide attempts: Findings from a representative sample of the adult, noninstitutionalized US population. Am J Public Health. 100:2473–2480. Bolton JM, Spiwak R, Sareen J (2012) Predicting suicide attempts with the SAD PERSONS scale: A longitudinal analysis. J Clin Psychiatry. 73:e735–e741. Bruffaerts R, Demyttenaere K, Borges G, Haro JM, Chiu WT, Hwang I, Karam EG, Kessler RC, Sampson N, Alonso J, Andrade LH, Angermeyer M, Benjet C, Bromet E, de Girolamo G, de Graaf R, Florescu S, Gureje O, Horiguchi I, Hu C, Kovess V, Levinson D, Posada-Villa J, Sagar R, Scott K, Tsang A, Vassilev SM, William DR, Nock MK (2010) Childhood adversities as risk factors for onset and persistence of suicidal behavior. Br J Psychiatry. 197:20–27. Buzan RD, Weissberg MP (1991) Suicide: Risk factors and therapeutic considerations in the ED. J Emerg Med. 10:335–343. Carli V, Jovanovic N, Podlesek A, Roy A, Rihmer Z, Maggi S, Marusic D, Cesaro C, Marusic A, Sarchiapone M (2010) The role of impulsivity in self-mutilators, suicide ideators and suicide attempters—A study of 1265 male incarcerated individuals. J Affect Disord. 123:116–122. Caterino JM, Sullivan AF, Betz ME, Espinola JA, Miller I, Camargo CA, Boudreaux ED, Emergency Department Safety Assessment and Follow-up Evaluation (ED-SAFE)
© 2015 Wolters Kluwer Health, Inc. All rights reserved. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
The Journal of Nervous and Mental Disease • Volume 203, Number 7, July 2015
Predictors of Future Suicide Attempts
Investigators (2013) Evaluating current patterns of assessment for self-harm in emergency departments: A multicenter study. Acad Emerg Med. 20:807–815.
Typologies of suicidality based upon ambivalence about living and dying. Compr Psychiatry. 53:461–467.
Cavanagh J, Carson A, Sharpe M, Lawrie SM (2003) Psychological autopsy studies of suicide: A systematic review. Psychol Med. 33:395–405.
Olfson M, Marcus SC, Bridge JA (2012) Emergency treatment of deliberate self-harm. Arch Gen Psychiatry. 69:80–88.
Centers for Disease Control and Prevention (CDC) (2012) Injury prevention & control. National suicide statistics at a glance. Cited March 30, 2013. Retrieved from www.cdc.gov/violenceprevention/suicide/statistics. Accessed January 2015.
Olfson M, Marcus SC, Bridge JA (2013) Emergency department recognition of mental disorders and short-term outcome of deliberate self-harm. Am J Psychiatry. 170: 1442–1450.
Cha CB, Najmi S, Park JM, Finn CT, Nock MK (2010) Attentional bias toward suiciderelated stimuli predicts suicidal behavior. J Abnorm Psychol. 119:616–622.
Patterson WM, Dohn HH, Bird J, Patterson GA (1983) Evaluation of suicidal patients: The SAD PERSONS scale. Psychosomatics. 24:343–345, 348–349.
Choi JW, Park S, Yi KK, Hong JP (2012) Suicide mortality of suicide attempt patients discharged from emergency room, nonsuicidal psychiatric patients discharged from emergency room, admitted suicide attempt patients, and admitted nonsuicidal psychiatric patients. Suicide Life Threat Behav. 43:235–243.
Posner K, Oquendo MA, Gould M, Stanley B, Davies M (2007) Columbia Classification Algorithm of Suicide Assessment (C-CASA): Classification of suicidal events in the FDA’s pediatric suicidal risk analysis of antidepressants. Am J Psychiatry. 164:1035–1043.
Cichetti DV, Volkmar F, Klin A, Showalter D (1995) Diagnosing autism using ICD-10 criteria: A comparison of neural networks and standard multivariate procedures. Child Neuropsychol. 1:26–37.
Randall JR, Colman I, Rowe BH (2011) A systematic review of psychometric assessment of self-harm risk in the emergency department. J Affect Disord. 134:348–355.
Cooper JB, Lawlor MP, Hiroeh U, Kapur N, Appleby L (2003) Factors that influence emergency department doctors’ assessment of suicide risk in deliberate self-harm patients. Eur J Emerg Med. 10:283–287. Crandall C, Olson L, Fullerton L, Sklar Z, Zumwalt R (1997) Guns and knives in New Mexico: Patterns of penetrating trauma, 1978–1993. Acad Emerg Med. 4:263–267. DeJong TM, Overholser JC, Stockmeier CA (2010) Apples to oranges?: A direct comparison between suicide attempts and suicide completers. J Affect Disord. 124:90–97. Elisei S, Verdolini N, Anastasi S (2012) Suicidal attempts among emergency department patients: One-year of clinical experience. Psychiatr Danub. 24: S140–S142. Fischer JE, Bachman LM, Jaeschke R (2003) A readers’ guide to the interpretation of diagnostic test properties: Clinical example of sepsis. Intensive Care Med. 29: 1043–1051. Gonzalez-Navarro MD, Lorenzo-Roman MI, Luna-Maldonado A, Gomez-Zapata M, Imbernon-Pardo E, Ruiz-Riquelme J (2012) Study of sociodemographic and psychopathological risk factors in suicide attempts. 2008–2010. Semergen. 38:439–444. Harris EC, Barraclough B (1997) Suicide as an outcome for mental disorders. A metaanalysis. Br J Psychiatry. 170:205–228. Hawton K, van Heeringen K (2009) Suicide. Lancet. 373:1372–1381. Hockberger RS, Rothstein RJ (1988) Assessment of suicide potential by nonpsychiatrists using the SAD PERSONS score. J Emerg Med. 6:99–107. Hor K, Taylor M (2010) Suicide and schizophrenia: A systematic review of rates and risk factors. J Psychopharmacol. 4(Suppl.):81–90. Horowitz LM, Snyder D, Ludi E, Rosenstein DL, Kohn-Godbout J, Lee L, Cartledge T, Farrar A, Pao M (2013) Ask suicide-screening questions to everyone in medical settings: The asQ’em Quality Improvement Project. Psychosomatics. 54:239–247. IBM Corp (2011) IBM SPSS statistics for Windows, Version 20.0. Armonk, NY: IBM Corp. Klonsky ED, May AM, Glenn CR (2012) The relationship between nonsuicidal self-injury and attempted suicide: Converging evidence from four samples. J Abnormal Psychol. 122:231–237. Landis JR, Koch GG (1977) The measurement of observer agreement for categorical data. Biometrics. 33:159–174.
Ronquillo L, Minassian A, Vilke GM, Wilson MP (2012) Literature-based recommendations for suicide assessment in the emergency department: A review. J Emerg Med. 43:836–842. Roos L, Sareen J, Bolton JM (2013) Suicide risk assessment tools, predictive validity and utility today: Time for a revamp? Neuropsychiatry. 3:483–495. Runeson B, Tidemalm D, Dahlin M, Lichtenstein P, Langstrom N (2010) Method of attempted suicide as predictor of subsequent successful suicide: National long term cohort study. BMJ. 341:c3222. Sareen J (2011) Anxiety disorders and risk for suicide: Why such controversy? Depress Anxiety. 28:941–945. Sareen J, Cox BJ, Afifi TO, de Graaf R, Asmundson GJ, ten Have M, Stein MB (2005) Anxiety disorders and risk for suicidal ideation and suicide attempts: A populationbased longitudinal study of adults. Arch Gen Psychiatry. 62:1249–1257. Schneider B, Schnabel A, Wetterling T, Bartusch B, Weber B, Georgi K (2008) How do personality disorders modify suicide risk? J Pers Disord. 22:233–245. Siris SG (2001) Suicide and schizophrenia. J Psychopharmacol. 15:127–135. Sokero TP, Melartin TK, Rytsala HJ, Leskela US, Lestela-Mielonen PS, Isometsa ET (2003) Suicidal ideation and attempts among psychiatric patients with major depressive disorder. J Clin Psychiatry. 64:1094–1100. Steeg S, Kapur N, Webb R, Applegate E, Stewart SL, Hawton K, Bergen H, Waters K, Cooper J (2012) The development of a population-level clinical screening tool for self-harm repetition and suicide: The ReACT Self-Harm Rule. Psychol Med. 42: 2383–2394. Suominen K, Isometsa E, Suokas J, Haukka J, Achte K, Lonngvist J (2004) Completed suicide after a suicide attempt: A 37-year follow-up study. Am J Psychiatry. 161: 562–563. Tidemalm D, Langstrom N, Lichtenstein P, Runeson B (2008) Risk of suicide after suicide attempt according to coexisting psychiatric disorder: Swedish cohort study with long term follow-up. BMJ. 337:1–6. Ting SA, Sullivan AF, Miller I, Espinola JA, Allen MH, Camargo CA, Boudreaux ED, Emergency Department Safety and Follow-up Evaluation (ED-SAFE) Investigators (2012) Multicenter study of predictors of suicide screening in emergency departments. Acad Emerg Med. 19:239–243.
Logan J, Hall J, Karch D (2011) Suicide categories by patterns of known risk factors. Arch Gen Psychiatry. 68:935–941.
US Department of Health and Human Services (HHS) Office of the Surgeon General and National Action Alliance for Suicide Prevention (2012, Sept.) National strategy for suicide prevention: Goals and objectives for action. Washington, DC: HHS.
Mitchell A, Garand D, Dean L, Panzak G, Taylor M (2005) Suicide assessment in hospital emergency departments: Implications for patient satisfaction and compliance. Top Emerg Med. 27:302–312.
Wang Y, Sareen J, Afifi T, Bolton S-L, Johnson EA, Bolton JM (2012) Recent stressful life events and suicide attempt: Results from a nationally representative sample. Psychiatr Ann. 42:101–108.
Nock MK, Park JM, Finn CT, Deliberto TL, Dour HJ, Banaji MR (2010) Measuring the suicidal mind: Implicit cognition predicts suicidal behavior. Psychol Sci. 21: 511–517.
Williams-Johnson J, Williams E, Gossel-Williams M, Sewell CA, Abel WD, Whitehorne-Smith PA (2012) Suicide attempt by self-poisoning: Characteristics of suicide attempters seen at the Emergency Room at the University Hospital of the West Indies. West Indian Med J. 61:526–531.
Nordentoft M, Mortensen PB, Pedersen CB (2011) Absolute risk of suicide after first hospital contact in mental disorder. Arch Gen Psychiatry. 68:1058–1064. O’Connor SS, Jobes DA, Yeargin MK, Fitzgerald ME, Rodriguez VM, Conrad AK, Linebrry TW (2012) A cross sectional investigation of the suicidal spectrum:
World Health Organization (WHO) (2010) Suicide prevention (SUPRE). Cited January 24, 2013. World Health Organization. Retrieved from www.who.int/ mental_health/prevention/suicide/suicideprevent/en. Accessed January 2015.
© 2015 Wolters Kluwer Health, Inc. All rights reserved. Copyright © 2015 Wolters Kluwer Health, Inc. All rights reserved.
www.jonmd.com
513
