Suicide Risk Among Emergency Care Patients: Exploring Incidental Risk Detection During Routine Care

Abstract

Background:

Universal screening improves detection of suicide risk in emergency department (ED) patients presenting with a medical chief complaint, referred to as incidental risk. We sought to better understand characteristics of individuals who present with incidental suicide risk and evaluate their suicide-related outcomes after the ED visit.

Methods:

The Emergency Department Safety Assessment and Follow-up Evaluation 2 (ED-SAFE 2) trial involved eight geographically diverse EDs. We randomly sampled data from two samples (cross-sectional, longitudinal) to compare patients with incidental risk to those with either negligible (i.e., screened negative for suicide risk) or congruent risk (i.e., behavioral health patients with suicide risk), as well as evaluate suicide-related outcomes in the 24 weeks after the ED visit. Chi-square tests, Student’s t-tests, Fisher’s exact tests, and one-way ANOVAs were used.

Results:

The relative percentage of incidental versus congruent risk was largely similar across sex, race, and ethnicity groups. Those with congruent or incidental risk were younger than those with negligible risk. Compared to congruent risk, incidental risk was driven more frequently by a past suicide attempt without recent ideation and was less likely to result in a behavioral health evaluation or intervention during the visit. Both congruent and incidental risk groups had substantial post-visit suicidal behavior, though rates were higher with congruent risk.

Conclusions:

ED patients with suicide risk incidental to their chief complaint often have significant risk factors for suicide and exhibit elevated post-visit suicidal behavior but are less likely than those with congruent risk to receive a behavioral health evaluation or intervention during the visit.

1. INTRODUCTION

Suicide remains a leading cause of death in the United States, with emergency departments (EDs) serving as a critical point of care for individuals at risk for suicidal behavior.^1–4 While EDs frequently encounter patients with psychiatric complaints, a substantial proportion of individuals at risk of suicide present with non-psychiatric issues, making risk detection challenging in routine clinical practice.^1,5–8 Traditional approaches that rely on screening patients presenting with psychiatric chief complaints alone tend to miss patients presenting with non-psychiatric complaints who have suicide risk, a group known as having incidental risk.^1,9,10

The Emergency Department Safety Assessment and Follow-up Evaluation (ED-SAFE 1) study implemented universal suicide risk screening in EDs and revealed that it nearly doubled the identification of individuals at risk, in part by capturing patients with incidental risk.^1,9 ED-SAFE 1 demonstrated that universal suicide risk screening combined with a multi-component intervention consisting of improved discharge resources and post-visit follow-up telephone calls led to a 30% reduction in total suicide deaths and attempts.^1,9,11,12 Building on these findings, the ED-SAFE 2 trial evaluated the effectiveness of integrating continuous quality improvement strategies and collaborative safety planning into routine ED workflows.¹³ ED-SAFE 2 showed that these interventions led up to a 43% reduction in the risk of suicide-related outcomes, supporting the feasibility and impact of embedding evidence-based practices into routine emergency care.¹⁴

Despite these advances, there remains a limited understanding of the characteristics and suicide-related outcomes of patients with incidental risk. Using participant data from the ED-SAFE 2 trial, this paper seeks to replicate and extend the findings from the recently published analysis of incidental risk among participants in the ED-SAFE 1 study.¹⁵ The original paper found that patients with incidental risk are generally more comparable to patients with congruent risk (i.e., patients presenting for psychiatric reasons with suicide risk) than negligible risk with respect to their demographic and clinical characteristics. However, patients with incidental risk were significantly less likely to receive behavioral health assessments and interventions during the visit and found to share a similar propensity for suicidal behavior after the ED visit as those with congruent risk.

As studies have shown that mental health and suicide-related ED visits are associated with a differentially increased risk of suicidal behavior in the near-term following discharge^10,16–18, the current paper adds to the evidence by identifying this risk among incidental risk patients. We describe the demographic, clinical characteristics, and clinical management of patients with incidental suicide risk and compare their suicide-related outcomes to patients with congruent risk. Similar to ED-SAFE 1, we anticipated that, compared to patients with congruent risk, those with incidental risk would: 1) be less likely to receive behavioral health assessment and interventions during the visit, and 2) have similar rates of suicidal behavior after the ED visit.

2. MATERIALS & METHODS

2.1. Overview

The ED-SAFE 2 study was completed after the original ED-SAFE 1 and sought to expand and further improve the suicide-related care procedures used in those eight EDs.¹⁴ All eight sites had already implemented universal risk screening for all adult patients using the Patient Safety Screener-3^1,5,9,13,14 before ED-SAFE 2 started. ED-SAFE 2 had three phases, including Baseline (Phase 1), Implementation (Phase 2), and Maintenance (Phase 3). For this analysis, we collapsed the three phases of the study and evaluated the combined sample. ED-SAFE 2 was approved by the University of Massachusetts Chan Medical School Institutional Review Board, which acted as the central IRB for the eight sites. Patients were not required to sign informed consent given that ED-SAFE 2 was an implementation study.

We used data from two randomly selected samples: a cross-sectional sample of general population adult (≥18 years) patients and a longitudinal sample of suicide-positive adult patients who were positive for suicide risk on the Patient Safety Screener-3. For both samples, trained, supervised research assistants manually abstracted data from the electronic health record (EHR) using a structured protocol. For each sample, the eight sites randomly selected 25 encounters per month for 36 months for chart review. In addition to information about the index ED visit, we collected follow-up data via chart review of the health system's EHR for 24 weeks after the index visit. To identify suicide outcomes potentially not captured in the EHR, we obtained data from the National Death Index to identify deaths and deaths by suicide.

2.2. Participants

2.2.1. Cross-sectional sample.

The cross-sectional sample allowed comparison of negligible, congruent, and incidental risk rates across the general adult population and within individual sites, demographics, and clinical groups. We only included encounters that had a complete Patient Safety Screener-3, defined as a documented response to active ideation in the past two weeks and for a lifetime suicide attempt, including timeframe of most recent attempt. Those who were not screened, only partially screened (e.g., screened for active suicidal ideation but not lifetime suicide attempt), or otherwise screened but had incomplete responses recorded (e.g., screened for active suicidal ideation and lifetime attempt but timeframe of most recent attempt was unknown) were excluded. One site was not included because it used a slightly different screener, making its definition of positive screening different than the other sites. Another site closed during the final phase of the study, resulting in fewer than expected encounters and a total of 4,776 encounters included in analyses.

2.2.2. Longitudinal sample.

The longitudinal sample (all positive screens) allowed a comparison of the demographics, clinical characteristics, clinical management, and post-visit suicide-related outcomes between participants with congruent and incidental risk. These analyses included 4,504 encounters. Because this sample was random, there was a chance that some encounters in this sample were also included in the cross-sectional sample.

2.3. Measures

2.3.1. Study site.

Seven of eight ED-SAFE 2 sites were used for these analyses.

2.3.2. Chief complaint.

The chief complaint as documented in the chart was reviewed and categorized as either psychiatric or medical using a detailed coding manual. Psychiatric chief complaints included any of the following: agitation, altered mental status, anxiety, panic, delirium, depression, homicidal ideation, intoxication, mania, psychiatric problems, psychosis, substance abuse, suicidal thoughts or behavior, non-suicidal self-injurious thoughts or behavior, or threatening, violent, or aggressive behavior. Any presentation that was not considered psychiatric was considered medical. Patients with both medical and psychiatric complaints were classified as psychiatric for study purposes. Any vague presentations were also reviewed by a research supervisor who categorized the chief complaint.

2.3.3. Patient Safety Screener-3.

The Patient Safety Screener-3 screens for depressed mood and active suicidal ideation in the past two weeks, as well as presence of a lifetime suicide attempt. If a lifetime attempt is present, the time range for the most recent attempt is assessed as being in the past 24 hours (today), past month, between 1 and 6 months, or more than 6 months ago. Responses were abstracted from the chart. For the purposes of this study, a positive included anyone who screened positive for active suicidal ideation in the past two weeks or a lifetime suicide attempt, which is consistent with the original psychometric studies. In addition, detailed responses were defined by: (1) no ideation, but presence of a lifetime attempt that was not on the day of the visit; (2) ideation alone with no lifetime attempt; (3) ideation with a lifetime attempt that was not on the day of the visit; or (4) ideation and an attempt on the day of the visit.

2.3.4. Demographics.

Age, sex (male/female), race (White/non-White), and ethnicity (Hispanic/non-Hispanic) were documented for both samples.

2.3.5. Clinical characteristics.

Clinical characteristics were identified from the encounter record, including if there was any evidence of past or current (unless otherwise specified): alcohol abuse, alcohol intoxication during the visit, illicit drug abuse, illicit drugs present on toxicology testing, non-suicidal self-injury, domestic violence, and depressed mood. In addition, the individual's medical record for the six months prior to the visit was reviewed and the number of ED visits, medical hospitalizations, psychiatric hospitalizations, and substance abuse hospitalizations were documented.

2.3.6. Clinical management.

The behavioral health-related care received during the visit was identified, including whether the patient was evaluated by a mental health provider and ED disposition (discharged, admitted medically, admitted psychiatrically). Among those who were discharged, we documented whether they had received a personalized safety plan, an appointment scheduled with an outpatient mental health treatment provider prior to ED discharge, and printed mental health and crisis discharge resources. Length of ED stay was also calculated for the longitudinal sample.

2.3.7. Post-visit suicide-related acute care.

For the longitudinal sample only, the medical records for each participant were reviewed for the 24 weeks after the index visit. Every return ED or inpatient visit was identified and reviewed. The research assistant reviewed all documentation associated with the encounter and determined if the visit was related to suicidal ideation or a suicide attempt. In addition, the International Classification of Disease – 10 (ICD-10) codes associated with the visit as assigned by the treating clinician and the professional coder were collected.

2.3.8. Mortality.

The National Death Index, along with chart review, was used to identify all deaths within the 24 weeks after the index visit (all-cause mortality). Due to the lag in reporting to the National Death Index, we requested data 24 months after the last encounter date of the study to ensure all deaths would be accounted for, though, for analyses, we only report deaths within the 24-week follow-up window. ICD-10 codes and documented manner of death (natural, homicide, accident, pending investigation, suicide, could not be determined or otherwise not documented or collected) were collected.

2.4. Outcome Definition

We used a positive screen on the Patient Safety Screener-3 to identify positive risk and the chief complaint coding to identify medical versus psychiatric chief complaints (see above). Patients who screened negative were considered negligible risk; those who presented with a psychiatric chief complaint and screened positive were considered congruent risk; and those who presented with a medical complaint and screened positive were considered incidental risk.

For the longitudinal sample, our primary suicide outcome was a composite defined by (1) death by suicide, as ascertained by the National Death Index or chart review, or (2) any ED or inpatient visit that was related to suicidal ideation or a suicide attempt, as ascertained by chart review. We applied a standard definition that used ICD-10 codes associated with intentional self-injury to determine if the death or visit was related to suicide.¹⁹ As an exploratory outcome, we applied a less conservative definition that expanded to include self-injury with indeterminate or unknown intent, which can help account for under-coding.¹⁹ These were distinguished as traditional (intentional only) and expanded (intentional, indeterminate, unknown intent) definitions.

2.5. Data Analysis

To evaluate differences in predictors among participants with negligible risk, incidental risk, and congruent risk in the cross-sectional sample, we completed Chi-square tests, Student’s t-tests, Fisher’s exact tests, and one-way ANOVAs with post-hoc Chi-square tests with Bonferroni adjustment and Tukey HSD tests, respectively. In the longitudinal sample, we used Chi-square tests to compare the cumulative proportions of participants who had a traditional or expanded suicide composite outcome within 4-, 12-, and 24-weeks after the index ED visit. All p-values were two-tailed, with p < .05 considered statistically significant. All analyses were conducted using SAS version 9.4 (SAS Institute Inc).

3. RESULTS

3.1. Cross-sectional sample

Table 1 summarizes the findings for the cross-sectional sample. Among the 4,776 encounters, 4,548 (95.2%) had negligible risk, 100 (2.1%) had congruent risk, and 128 (2.7%) had incidental risk.

Table 1.

Comparison of Patient Characteristics Across Negligible, Congruent, and Incidental Risk Detection in ED-SAFE 2 Cross-Sectional Sample

	Overall1	Risk Negative2	Congruent Risk Positive	Incidental Risk Positive	Statistic, p-value
Total	n=4,776 (100%)	n=4,548 (95.2%)	n=100 (2.1%)	n=128 (2.7%)
Site and Screening Characteristics
Study site					X2(12) = 66.26, p < .001
Site 100	470 (9.8%)	466 (99.2%)	3 (0.6%)	1 (0.2%)
Site 200	687 (14.4%)	635 (92.4%)	25 (3.6%)	27 (3.9%)
Site 300	729 (15.3%)	704 (96.6%)	9 (1.2%)	16 (2.2%)
Site 400	-	-	-	-
Site 500	775 (16.2%)	751 (96.9%)	13 (1.7%)	11 (1.4%)
Site 600	809 (16.9%)	742 (91.7%)	23 (2.8%)	44 (5.4%)
Site 700	605 (12.7%)	579 (95.7%)	15 (2.5%)	11 (1.8%)
Site 800	701 (14.7%)	671 (95.7%)	12 (1.7%)	18 (2.6%)
PSS-3 Cat					Fisher’s Exact Test, p < .001
No Ideation/No LA	4,548 (95.2%)	4,548 (100%)	0 (0%)	0 (0%)
No Ideation/LA but Not Current	118 (2.5%)	0 (0%)a	13 (13.0%)b	105 (82.0%)c
Ideation Only (No LA)	50 (1.1%)	0 (0%)	34 (34.0%)	16 (12.5%)
Ideation/LA but Not Current	43 (0.9%)	0 (0%)	37 (37.0%)	6 (4.7%)
Ideation/Current Attempt	17 (0.4%)	0 (0%)	16 (16.0%)	1 (0.8%)
Demographic Characteristics
Age					F(2, 4769) = 17.27, p < .001
Mean (SD)	45.86 (18.8)	46.19 (18.9)a	36.11 (13.7)b	41.84 (15.7)b
Sex					X2(2) = 4.62, p = .0995
Male	2,145 (44.9%)	2,051 (45.1%)	48 (48.0%)	46 (35.9%)
Female	2,631 (55.1%)	2,497 (54.9%)	52 (52.0%)	82 (60.1%)
Race					X2(2) = 2.97, p = .2265
White	3,338 (69.9%)	3,167 (70.0%)	75 (75.0%)	96 (75.0%)
Non-white	1,438 (30.1%)	1,381 (30.0%)	25 (25.0%)	32 (25.0%)
Ethnicity					X2(2) = 8.86, p = .0119
Hispanic	779 (17.0%)	758 (17.3%)	9 (5.0%)	12 (9.7%)
Non-Hispanic	3,816 (83.1%)	3,618 (82.7%)	86 (95.0%)	112 (90.3%)
Clinical Characteristics
Alcohol Use Problems					X2(2) = 94.70, p < .001
No/Not Documented	4,481 (93.8%)	4,293 (94.4%)	70 (70.7%)	118 (92.2%)
Current or History of Alcohol Use Problems	294 (6.2%)	255 (5.6%)a	29 (29.3%)b	10 (7.8%)a
Intoxicated					Fisher’s Exact Test, p < .001
No/Not Documented	4,682 (98.1%)	4,480 (98.5%)	78 (78.0%)	124 (97.6%)
Patient Intoxicated During Visit	93 (2.0%)	68 (1.5%)a	22 (22.0%)b	3 (2.4%)a
Drug Use Problems					X2(2) = 124.03, p < .001
No/Not Documented	4,269 (89.4%)	4,110 (90.4%)	58 (58.0%)	101 (78.9%)
Current or History of Drug Use Problems	505 (10.6%)	436 (9.6%)a	42 (42.0%)b	27 (21.1%)c
Toxicology screening positive					Fisher’s Exact Test, p < .001
No/Not Documented	4,616 (96.7%)	4,434 (97.5%)	59 (59.6%)	123 (96.1%)
Positive Tox Screen	159 (3.3%)	114 (2.5%)a	40 (40.4%)b	5 (3.9%)a
Non-suicidal self-harm behavior					Fisher’s Exact Test, p < .001
No/Not Documented	4,679 (98.0%)	4,531 (99.7%)	63 (63.0%)	85 (66.4%)
Present in Past Week	77 (1.6%)	7 (0.2%)	28 (28.0%)	42 (32.8%)
Present >1 Week Ago or Time Not Specified	19 (0.4%)	9 (0.2%)a	9 (9.0)b	1 (0.8%)b
Domestic violence					X2(4) = 46.22, p < .001
No/Not Documented	4,721 (98.9%)	4,503 (99.0%)	92 (92.0%)	126 (98.4%)
Present in Past Week	29 (0.6%)	23 (0.5%)	4 (4.0%)	2 (1.6%)
Present >1 Week Ago or Time Not Specified	25 (0.5%)	21 (0.5%)a	4 (4.0%)b	0 (0%)a,b
History of depressed mood					X2(4) = 840.43, p < .001
No/Not Documented	4,180 (87.5%)	4,106 (90.3%)	8 (8.0%)	66 (51.6%)
Present in Past Week	304 (6.4%)	232 (5.1%)	33 (33.0%)	39 (30.5%)
Present >1 Week Ago or Time Not Specified	291 (6.1%)	209 (4.6%)a	59 (59.0%)b	23 (18.0%)c
Number of ED visits in past 6 months					F(2, 4772) = 12.72, p < .001
Mean (SD)	1.15 (3.0)	1.11 (2.9)a	1.83 (3.2)b	2.30 (4.1)b
Number of medical-related hospital visits					F(2, 4771) = 1.51, p = .22
Mean (SD)	0.38 (1.4)	0.38 (1.4)	0.14 (0.6)	0.39 (0.9)
Number of psychological-related hospital visits					F(2, 4761) = 111.40, p < .001
Mean (SD)	0.02 (0.2)	0.01 (0.2)a	0.26 (0.6)b	0.17 (0.6)c
Number of substance abuse-related hospital visits					F(2, 4761) = 2.71, p = .0668
Mean (SD)	0.01 (0.1)	0.01 (0.1)	0.03 (0.2)	0.01 (0.1)
ED Clinical Management Characteristics
Patient evaluated by mental health professional during visit					X2(2) = 868.30, p < .001
Patient Not Evaluated	2,960 (93.6%)	2,863 (96.1%)	20 (21.1%)	77 (87.5%)
Patient Evaluated	203 (6.4%)	117 (3.9%)a	75 (79.0%)b	11 (12.5%)c
ED disposition					Fisher’s Exact Test, p < .001
Discharged	3,580 (75.9%)	3,455 (76.9%)a	31 (33.0%)b	94 (74.6%)a
Admitted (Medical)	1,027 (21.8%)	992 (22.1%)a	7 (7.5%)b	28 (22.2%)a
Admitted (Psych/ Substance Use-Related Admission)	109 (2.3%)	49 (1.1%)a	56 (59.6%)b	4 (3.2%)a
Referral resources for mental health services provided to patient (e.g., provider lists, hotline numbers) *					X2(2) = 676.11, p < .001
No Resources Provided	3,550 (99.2%)	3,444 (99.7%)	18 (58.1%)	88 (93.6%)
Resources Provided	30 (0.8%)	11 (0.3%)a	13 (41.9%)b	6 (6.4%)c
Appointment made with outpatient mental health treatment provider before patient discharge *					Fisher’s Exact Test, p < .001
No Appointment Made	3,575 (99.9%)	3,454 (>99.9%)	27 (87.1%)	94 (100%)
Appointment Made	5 (0.1%)	1 (<0.1%)a	4 (12.9%)b	0 (0%)a
Safety plan created for patient *					Fisher’s Exact Test, p = .0242
No Safety Plan	3,577 (99.9%)	3,453 (99.9%)	30 (96.8%)	94 (100%)
Safety Plan Present	3 (0.1%)	2 (0.1%)a	1 (3.2%)b	0 (0%)a,b

Abbreviations: ED, emergency department; NA, not applicable; LA = Lifetime attempt; PSS-3 = Patient Safety Screener-3

^*Among several ED Management Characteristics (safety planning, appointment made with outpatient mental health provider, and referral resources), the sample was restricted to individuals who were discharged as these factors are most pertinent to this group.

Values not sharing superscripts differ significantly from each other at p < .05 for post-hoc tests. Post hoc tests were Bonferroniadjusted X² test (following significant overall X²) or Tukey HSD test (following significant overall one-way ANOVA).

For cross-sectional ethnicity analyses, though the overall chi square was significant, pairwise differences were not significant. For post-hoc tests for PSS-3 Cat, comparisons were made between “No Ideation/LA but Not Current” and a collapsed category comprising all other PSS-3 Cat categories. For post-hoc tests for non-suicidal self-harm behavior, domestic violence, and history of depressed mood, “Present in Past Week” and “Present >1 Week Ago or Time Not Specified” were combined into a “Yes” category. For post-hoc tests for ED disposition, the following comparisons of disposition were made for each of the three comparisons of risk group (i.e., Congruent vs. Incidental, Congruent vs. Negligible, and Incidental vs. Negligible: i) discharged vs. any admission (medical or psych/substance use-related) and ii) medical admission vs. psych/substance use-related admission.

Among a total of 6,224 eligible encounters, 4,776 with complete PSS-3 screens were included and 1,448 with incomplete/missing screens were excluded. Complete screens were defined as screening for and recording complete responses to active ideation in the past two weeks and for a lifetime suicide attempt, including timeframe of most recent attempt. Incomplete/missing screens were defined as those who were not screened, only partially screened (e.g., screened for active suicidal ideation but not lifetime suicide attempt), or otherwise screened but had incomplete responses recorded (e.g., screened for active suicidal ideation and lifetime attempt but timeframe of most recent attempt was unknown).

A sensitivity analysis comparing those with complete PSS-3 screens vs. incomplete/missing PSS-3 screens revealed no significant differences in mean age, [F(1, 6217)=0.00, p=.9614], sex [X²(1)=0.91, p=.3390], or race [X²(1)=0.49, p=.4821]. Those with complete vs. incomplete/missing PSS-3 screens differed significantly with respect to ethnicity, where Hispanic individuals were more likely to have an incomplete/missing PSS-3 screen than non-Hispanic individuals (32.4% vs. 20.9%, respectively), [X²(1)=69.07, p<.001], and chief complaint, where individuals with a psychiatric chief complaint were more likely to have an incomplete/missing PSS-3 screen than individuals with a medical chief complaint (33.5% vs. 22.3%, respectively), [X²(1)=34.82, p<.001]. However, since the majority of individuals in the sample had a medical complaint, estimates of incidental risk are reasonably underestimated.

^1:Percentages are based on column totals.

^2:Percentages are based on row totals for study site and demographic characteristics and column totals for all others.

3.1.1. Study site.

Sites had markedly different proportions of negligible, congruent risk, and incidental risk, with prevalence of congruent risk ranging from 0.6% to 3.6% and incidental risk ranging from 0.2% to 5.4% [X² (12, N=4,776)=66.26, p<.001].

3.1.2. Patient Safety Screener-3.

Patterns of responses to screener items differed significantly overall [Fisher’s exact test, p<.001]. A majority of incidental risk positives had a lifetime suicide attempt with no active ideation in the past two weeks: 105 (82%) of the incidental positives versus 13 (13%) of the congruent positives [X² (1, N=228)=107.14, p<.001]. Most incidental risk was related to an attempt that was not on the day of the index ED visit.

3.1.3. Demographics.

The only demographic difference found was age, with individuals with congruent risk (M=36.11, SD=13.70) and incidental risk (M=41.84, SD=15.70), being younger than those with negligible risk, (M=46.19, SD=18.90) [F (2, 4769)=17.27, p<.001].

3.1.4. Clinical characteristics.

The likelihood of having a past or current drug use problem was highest among individuals with congruent risk, followed by those with incidental risk, and lowest among those with negligible risk [X² (2, N=4,774)=124.03 p<.001]. The same pattern was observed for depressed mood [X² (4, N=4,775)=840.43 p<.001]. Compared to negligible risk participants, those with incidental or congruent risk were more likely to have a past or current history of non-suicidal self-injury [Fisher’s exact test, p< .001]. However, compared to patients with incidental or negligible risk, congruent risk patients were more likely to have a past or current history of alcohol problems [X² (2, N=4,775)=94.70, p<.001], to be intoxicated during the visit [Fisher’s exact test, p< .001], and have a positive toxicology screen [Fisher’s exact test, p< .001] as well as were more likely than those with negligible risk to have a past or current history of domestic violence [X² (4, N=4,775)=46.22 p<.001].

Incidental (M=2.30, SD=4.1) and congruent risk (M=1.83, SD=3.2) patients had more ED visits in the past 6 months compared to negligible risk patients (M=1.11, SD=2.9) [F(2, 4772)=12.72, p<.001], but differences between the incidental and congruent groups were not significant. Congruent risk patients (M=0.26, SD=0.6) had more psychiatric inpatient hospitalizations compared to those with incidental risk (M=0.17, SD=0.6), who had more psychiatric inpatient hospitalizations than those with negligible risk, (M=0.01, SD=0.2), respectively, [F(2, 4761)=111.40, p<.001].

3.1.5. Clinical management.

During the index ED visit, incidental risk and congruent risk patients were both more likely to be evaluated by a mental health provider than negligible risk patients, but incidental risk patients were less likely to be evaluated than congruent risk patients [X² (2, N=3,163)=868.30, p<.001]. For disposition, this same pattern was true for being admitted to a psychiatric unit. For being admitted to a medical unit and being discharged, incidental risk and negligible risk patients had comparable rates and both were more likely than with congruent risk patients [Fisher’s exact test, p<.001].

Among discharged patients, those with congruent risk were more likely to receive mental health and crisis referral information than those with incidental risk, who in turn were more likely to receive referral information than those with negligible risk [X² (2, N=3,580)=676.11, p<.001]. Congruent risk patients were more likely than either incidental or negligible risk to have an appointment scheduled with an outpatient mental health treatment provider prior to discharge [Fisher’s exact test, p<.001]. All risk groups had low rates of receiving a safety plan. Among 3 total safety plans, 2 were created for patients with negligible risk, 1 for congruent risk, and none for incidental risk.

3.2. Longitudinal sample

Table 2 summarizes the findings for the baseline characteristics of the longitudinal sample, while Table 3 summarizes the 24-week suicide-related outcomes. Among the 4,504 suicide risk positive encounters, 3,451 (76.6%) were congruent risk and 1,053 (23.4%) were incidental risk.

Table 2.

Comparison of Patient Characteristics Across Incidental and Congruent Risk Detection in ED-SAFE 2 Longitudinal Sample

	Overall1	Congruent Risk Positive2	Incidental Risk Positive	Statistic, p-value
Total	n=4,504 (100%)	n=3,451 (76.6%)	n=1,053 (23.4%)
Site and Screening Characteristics
Study site				X2(6) = 293.98, p < .001
Site 100	305 (6.8%)	290 (95.1%)	15 (4.9%)
Site 200	640 (14.2%)	436 (68.1%)	204 (31.9%)
Site 300	737 (16.4%)	648 (87.9%)	89 (12.1%)
Site 400	-	-	-
Site 500	722 (16.0%)	606 (83.9%)	116 (16.1%)
Site 600	826 (18.3%)	502 (60.8%)	324 (39.2%)
Site 700	682 (15.1%)	485 (71.1%)	197 (28.9%)
Site 800	592 (13.1%)	484 (81.8%)	108 (18.2%)
PSS-3 Cat				X2(3) = 617.25, p < .001
No Ideation/LA but Not Current	300 (6.7%)	75 (2.2%)	225 (21.4%)
Ideation Only (No LA)	1,726 (38.3%)	1,243 (36.0%)	483 (45.9%)
Ideation/LA but Not Current	1,876 (41.7%)	1,568 (45.4%)	308 (29.3%)
Ideation/Current Attempt	602 (13.4%)	565 (16.4%)	37 (3.5%)
Demographic Characteristics
Age				t(1560.2) = 8.56, p < .001
Mean (SD)	38.25 (14.7)	37.14 (14.0)	41.88 (16.2)
Sex				X2(1) = 17.86, p < .001
Male	2,314 (51.4%)	1,833 (53.1%)	481 (45.7%)
Female	2,190 (48.6%)	1,618 (46.9%)	572 (54.3%)
Race				X2(1) = 4.58, p = .0323
White	3,339 (74.1%)	2,585 (74.9%)	754 (71.6%)
Non-white	1,165 (25.9%)	866 (25.3%)	299 (28.4%)
Ethnicity				X2(1) = 0.14, p = .71
Hispanic	465 (10.8%)	354 (10.7%)	111 (11.1%)
Non-Hispanic	3,855 (89.2%)	2,965 (89.3%)	890 (89.9%)
Clinical Characteristics
Alcohol Use Problems				X2(1) = 125.55 p < .001
No/Not Documented	3,339 (74.1%)	2,419 (70.1%)	920 (87.4%)
Current or History of Alcohol Use Problems	1,165 (25.9%)	1,032 (29.9%)	133 (12.6%)
Intoxicated				X2(1) = 125.70, p < .001
No/Not Documented	3,764 (83.6%)	2,766 (80.2%)	998 (94.8%)
Patient Intoxicated During Visit	740 (16.4%)	685 (19.9%)	55 (5.2%)
Drug Use Problems				X2(1) = 102.70, p < .001
No/Not Documented	2,735 (60.7%)	1,955 (56.7%)	780 (74.1%)
Current or History of Drug Use Problems	1,769 (39.3%)	1,496 (43.4%)	273 (25.9%)
Toxicology screening positive				X2(1) = 215.82, p < .001
No/Not Documented	3,110 (69.1%)	2,190 (63.5%)	920 (87.4%)
Positive Tox Screen	1,394 (31.0%)	1,261 (36.5%)	133 (12.6%)
Non-suicidal self-harm behavior				X2(2) = 50.82, p < .001
No/Not Documented	3,222 (71.6%)	2,420 (70.1%)	802 (76.2%)
Present in Past Week	988 (21.9%)	757 (21.9%)	231 (22.0%)
Present >1 Week Ago or Time Not Specified	293 (6.5%)	274 (7.9%)	19 (1.8%)
Domestic violence				X2(2) = 25.04, p < .001
No/Not Documented	4,140 (91.9%)	3,139 (90.7%)	1,001 (95.1%)
Present in Past Week	262 (5.8%)	234 (6.8%)	28 (2.7%)
Present >1 Week Ago or Time Not Specified	102 (2.3%)	78 (2.3%)	24 (2.3%)
History of depressed mood				X2(2) = 361.98, p < .001
No/Not Documented	340 (7.6%)	155 (4.5%)	185 (17.6%)
Present in Past Week	1,031 (22.9%)	665 (19.3%)	366 (34.8%)
Present >1 Week Ago or Time Not Specified	3,133 (69.6%)	2,631 (76.2%)	502 (47.7%)
Number of ED visits in past 6 months				t(4502) = 1.77, p = .0773
Mean (SD)	1.97 (4.1)	1.91 (4.0)	2.17 (4.5)
Number of medical-related hospital visits				t(1369.9) = 6.82, p < .001
Mean (SD)	0.30 (1.1)	0.23 (0.9)	0.54 (1.4)
Number of psychological-related hospital visits				t(2896.5) = −6.97, p < .001
Mean (SD)	0.31 (0.9)	0.35 (1.0)	0.18 (0.6)
Number of substance use-related hospital visits				t(3571.6) = −3.48, p = .005
Mean (SD)	0.03 (0.3)	0.04 (0.3)	0.01 (0.1)
ED Management Characteristics
Patient evaluated by mental health professional during visit				X2(1) = 1058.37, p < .001
Patient Not Evaluated	858 (21.9%)	367 (11.6%)	491 (66.6%)
Patient Evaluated	3,052 (78.1%)	2,806 (88.4%)	246 (33.4%)
ED disposition				X2(2) = 570.21 p < .001
Discharged	2,145 (48.4%)	1,441 (42.4%)	704 (67.9%)
Admitted (Medical)	498 (11.2%)	266 (7.8%)	232 (22.4%)
Admitted (Psych/Substance Use-Related Admission)	1,789 (40.4%)	1,688 (49.7%)	101 (9.7%)
Referral resources for mental health services provided to patient (e.g., provider lists, hotline numbers) *				X2(1) = 277.52, p < .001
No Resources Provided	1,214 (56.6%)	636 (44.1%)	578 (82.1%)
Resources Provided	931 (43.4%)	805 (55.9%)	126 (17.9%)
Appointment made with outpatient mental health treatment provider before patient discharge *				X2(1) = 52.62, p < .001
No Appointment Made	1,995 (93.0%)	1,300 (90.2%)	695 (98.7%)
Appointment Made	150 (7.0%)	141 (9.8%)	9 (1.3%)
Safety plan created for patient *				X2(1) = 40.43, p < .001
No Safety Plan	1,874 (87.4%)	1,213 (84.2%)	661 (93.9%)
Safety Plan Present	271 (12.6%)	228 (15.8%)	43 (6.1%)

Abbreviations: ED, emergency department; NA, not applicable; LA = Lifetime attempt; PSS-3 = Patient Safety Screener-3

^*Among several ED Management Characteristics (safety planning, appointment made with outpatient mental health provider, and referral resources), the sample was restricted to individuals who were discharged as these factors are most pertinent to this group.

For the longitudinal sample post-hoc test for PSS-3 Cat, a comparison was made between “No Ideation/LA but Not Current” and a collapsed category comprising all other PSS-3 Cat categories. Differences in proportions between congruent and incidental risk were significantly different [X² (1, N=4,504)=478.11, p<.001].

^1:Percentages are based on column totals.

^2:Percentages are based on row totals for study site and demographics and column totals for all others.

Table 3.

Comparison of Cumulative Suicide Composite Outcomes in ED-SAFE 2 Longitudinal Sample

Traditionala (Most Conservative) Suicide Composite Outcomes
		Suicide death or suicide-related ED/inpatient visit in 4 weeks		Suicide death or suicide-related ED/inpatient visit in 12 weeks		Suicide death or suicide-related ED/inpatient visit in 24 weeks
	Total1	No	Yes	No	Yes	No	Yes
Overall 2	n=4,504 (100%)	3,933 (87.3%)	571 (12.7%)	3,636 (80.7%)	868 (19.3%)	3,378 (75.0%)	1,126 (25.0%)
Congruent Risk Positive	3,451 (76.6%)	2,958 (85.7%)	493 (14.3%)	2,719 (78.8%)	732 (21.2%)	2,510 (72.7%)	941 (27.3%)
Incidental Risk Positive	1,053 (23.4%)	975 (92.6%)	78 (7.4%)	917 (87.1%)	136 (12.9%)	868 (82.4%)	185 (17.6%)
Chi-Square Test, p-value	--	X2(1) = 34.48, p < .001		X2(1) = 35.69, p < .001		X2(1) = 40.48, p < .001
Expandedb (Less Conservative) Suicide Composite Outcomes
		Suicide death or suicide-related ED/inpatient visit in 4 weeks (n)		Suicide death or suicide-related ED/inpatient visit in 12 weeks (n)		Suicide death or suicide-related ED/inpatient visit in 24 weeks (n)
	Total1	No	Yes	No	Yes	No	Yes
Overall 2	n=4,504 (100%)	3,931 (87.3%)	573 (12.7%)	3,631 (80.6%)	873 (19.4%)	3,147 (69.9%)	1,357 (30.1%)
Congruent Risk Positive	3,451 (76.6%)	2,957 (85.7%)	494 (14.3%)	2,715 (78.7%)	736 (21.3%)	2,378 (68.9%)	1,073 (31.1%)
Incidental Risk Positive	1,053 (23.4%)	974 (92.5%)	79 (7.5%)	916 (87.0%)	137 (13.0%)	769 (73.0%)	284 (27.0%)
Chi-Square Test, p-value	--	X2(1) = 33.72, p < .001		X2(1) = 35.71, p < .001		X2(1) = 6.51, p = .0107

Abbreviations: ED, emergency department.

The relative frequencies of each of the individual events comprising the suicide composite outcome (i.e., ED/inpatient visit related to suicidal ideation only, ED/inpatient visit related to suicide attempt, and suicide death) are presented in Supplemental Table S-3.

^aThe traditional suicide composite outcome was defined as suicide death or any ED or inpatient visit with an International Classification of Diseases −10 code reflecting suicidal ideation or intentional self-injury, or research assistant abstraction noting that suicidal ideation or suicidal behavior was present during the visit.

^bThe expanded suicide composite outcome was defined as suicide death or any ED or inpatient visit with an International Classification of Diseases −10 code reflecting suicidal ideation or intentional self-injury, indeterminate self-injury, or unknown intent, or research assistant abstraction noting that suicidal ideation or suicidal behavior was present during the visit.

^1:Percentages are based on column totals.

^2:Percentages are based on row totals.

3.2.1. Study site.

Sites had markedly different proportions of positives that were incidental, with proportions ranging from 4.9% to 39.2%, [X²(6, N =4,504)=293.98, p<.001].

3.2.2. Patient Safety Screener-3.

225 (21.4%) of those with incidental risk, versus 75 (2.2%) of those with congruent risk had no recent active ideation and only a lifetime attempt that caused them to be positive, [X² (1, N=4,504)=478.11, p<.001]. Nevertheless, among those with incidental risk, the majority had recent risk factors, with 78.6% (n=828) having active suicidal ideation in the past 2 weeks, and 3.5% (n=37) reporting an attempt on the day of the visit.

3.2.2. Demographics.

Similar to the cross-sectional analysis, there were age differences in incidental vs congruent risk [t(1560.2)=8.56, p<.001], with incidental risk patients being older (M=41.88, SD=16.2) than congruent risk patients (M=37.14, SD=14.0). Among risk positive patients, there was a greater proportion of women with incidental risk (54.3%; n=572) compared to men (45.7%; n=481) [X²(1, N=4,504)=17.86, p<.001].

3.2.3. Clinical characteristics.

At the index visit, incidental risk patients were less likely to report substance use problems [alcohol problems: 12.6% vs 29.9%, X² (1, N=4,504)=125.55, p<.001; drug problems: 25.9% vs 43.4%, X² (1, N=4,504) = 102.70, p<.001] and other clinical risk factors, namely, intoxication at index visit, depressed mood, history of non-suicidal self-injury, and experience of domestic violence than congruent risk patients [all p<.001, see Table 2]. For hospitalizations, incidental risk patients had more medical [t(1369.9)=6.82, p<.001] and fewer psychiatric [t(2896.5)=−6.97, p<.001] and substance use hospitalizations [t(3571.6)=−3.48, p<.001].

3.2.4. Clinical management.

Incidental risk patients were much less likely to be evaluated by a mental health professional [X² (1, N=3,910)=1,058.37, p<.001], less likely to be admitted for a psychiatric reason, and more likely to be admitted for a medical reason [X² (2, N=4,432)=570.21, p<.001]. Of patients who were discharged, incidental risk patients were less likely to get mental health referral resources [X² (1, N=2,145)=277.52, p<.001], an outpatient mental health appointment made before discharge [X² (1, N=2,145)=56.62, p<.001], or a safety plan [X² (1, N=2,145) = 40.43, p<.001]. Congruent risk patients had significantly longer lengths of stay in the ED [t(1873) = −15.09, p < .001] (Supplemental Table S-1).

3.2.5. Suicide-related outcomes.

Compared to congruent risk patients, our traditional and expanded suicide composites were less common for incidental risk patients within 4 weeks [X² (1, N=4,504)=34.48, p<.001], 12 weeks [X² (1, N=4,504)=35.69, p<.001], and 24 weeks [X² (1, N=4,504)=40.48, p<.001] after the visit. While congruent risk patients had more incidents of suicide-related behavior, the absolute rates for incidental risk patients were still high, with 185 (17.6%) of the incidental risk patients having died by suicide or having a suicide-related acute care visit in the 24 weeks after the index visit.

Among incidental risk patients, being admitted medically did not appear to modify suicide composite outcome rates (Supplemental Table S-2). Most traditional suicide composite outcome events were attributable to ED or inpatient visits related to suicidal ideation alone (Supplemental Table S-3), and among incidental risk patients, a substantial proportion of composite outcomes occurred among those whose index ED visit had a first-listed complaint related to any aches or pains (Supplemental Table S-4).

4. DISCUSSION

Universal screening in EDs offers the advantage of detecting suicide risk among patients with medical complaints (incidental risk) beyond just those presenting with psychiatric complaints (congruent risk).^1,9,20–31 This approach identifies opportunities for intervention to reduce future suicidal behavior.^9,14,32–36 However, little is known about how incidental risk patients compare to congruent and negligible risk patients, particularly after ED discharge.

Our findings replicated many trends reported with the original ED-SAFE 1 data,¹⁵ which is notable given the studies’ different methods. Both studies found incidental risk patients tended to be older and more likely female and non-white, suggesting universal screening may better identify hidden risk in these population segments.^15,37–40 Similarly, both studies found that incidental risk patients had more medical illness than those with congruent risk. This study extended results from ED-SAFE 1 by showing that in the prior six months, incidental risk patients had more medical admissions but fewer psychiatric or substance use hospitalizations compared to congruent risk patients. They were also more likely to be admitted to a medical versus psychiatric unit during the index visit than those with congruent risk or negligible risk. These healthcare utilization patterns depict incidental risk patients as medically complex, with acute medical concerns and often undetected suicide risk. Interventions must therefore address both medical and behavioral health needs of these patients. However, the degree to which medical factors are directly driving suicide risk in incidental risk patients is unknown, as we have only observed a correlation.

Clinical management of the two risk groups differed markedly. Universal screening is intended to prompt behavioral health evaluation and intervention for those who might otherwise be missed. Yet, most (66%) incidental risk patients received no behavioral health evaluation during their ED visit, despite that all EDs were participating in a grant-funded suicide prevention study. As receiving a behavioral health assessment during the ED visit is protective and associated with lower return rates for suicide attempts,^16,41–43 this intervention gap is particularly concerning. Many EDs have limited behavioral health staffing, and requesting an evaluation can require long patient waits. Standard safety measures may be implemented regardless of risk level, including removal of hazards and continuous observation, but these may be stigmatizing and resource-intensive for patients whose risk is driven by only a distant lifetime suicide attempt. Our current findings suggest the possible need for further evaluation and lower intensity interventions, such as provision of discharge resources.

Examining those discharged home, only ~18% of incidental risk patients received mental health or crisis service resources, the least intensive intervention conceivable. This represents the group that the ED has the most influence over, and responsibility for, since those admitted to inpatient units will presumably get interventions there. Fewer than 2% had a behavioral health appointment arranged before they were discharged, a practice shown to improve outpatient care engagement.^44–47 No patients with incidental risk received an evidence-based personalized safety plan^48,49 despite safety planning being an ED focus area for continuous process improvement. Incidental risk patients in EDs without special suicide prevention initiatives may receive even less intervention. Educating and training providers to boost their familiarity and confidence with safety planning may be an opportunity to overcome barriers to low rates of delivery. Our current findings are particularly compelling, as many medical conditions are exclusionary for accessing behavioral health services, including inpatient and residential levels of care.

These low rates of intervention delivery are likely a function of multiple contributing causes. The compartmentalization of care in the ED, from a process perspective, may make it more challenging to deliver intersecting treatments, particularly since an ED’s responsibility is to address a patient’s most acute concerns. If these are chiefly medical, then patients with incidental risk are less likely to receive psychiatric interventions, as we have shown. Potential opportunities for improvement include implementing digital health interventions in moments of downtime, such as while patients wait for care. Such interventions are increasingly being evaluated. Improving delivery of evidence-based interventions would also address the needs of congruent risk patients with co-occurring problems, such as alcohol use disorder and intimate partner violence, which represent modifiable risk factors for suicidal behavior. At a broader level, medical schools should develop curriculums that emphasize caring for patients with intersecting medical and behavioral health needs. Ultimately, ED-initiated interventions can only reduce suicidal behavior if they are actually delivered.

In our study, the trajectory of incidental risk in the 24 weeks post-discharge tells an important story. Nearly 18% of incidental risk patients either died by suicide or had ideation or an attempt severe enough that it led to an ED visit or hospitalization. This result highlights that patients presenting with medical complaints and heterogeneous suicide risk have elevated risk for serious suicidal behavior in a relatively short time after the ED visit. We did not follow those with negligible risk over time, so we do not have good estimates of the suicide composite within that sample. Nonetheless, the 18% composite rate in this group is an alarming statistic.

Interestingly, this rate was significantly lower than the composite rate of 27% observed among those with congruent risk, contrary to our hypothesis of similar rates and in contrast with ED-SAFE 1, which found both groups had comparable probability of suicidal behavior. This may reflect methodological differences between the two studies. ED-SAFE 1 used telephone interviews, which have been shown to capture many more suicide attempts than medical records alone.¹² It is possible that some suicide attempts among incidental risk patients did not lead to ED visit at a higher rate than among congruent risk patients and were therefore under-detected in the medical record for ED-SAFE 2. Future studies using both ascertainment methods are needed to disentangle this.

4.1. Strengths and Limitations

This study sampled from seven geographically and sociodemographically diverse EDs. It used robust random selection to help ensure representative samples and leveraged two distinct samples with complementary strengths that improved our understanding of the impact of universal screening and how incidental risk patients compare to negligible and congruent risk patients. It provided a rare glimpse into suicide-related outcomes in congruent and incidental risk patients after the ED visit that helped address a major weakness in the ED-SAFE 1 study. The larger longitudinal sample in this study (n=4,504, 23% incidental risk) compared to ED-SAFE 1 (n=1,362, 8% incidental risk) allowed for better assessment of suicide-related outcomes.

Despite the larger sample, we may have still under-represented persons with incidental suicide risk by excluding those with incomplete safety screenings—potentially more common among patients, with severe or otherwise limiting medical conditions, unable to participate in screening (e.g., unconscious, direct to surgery). Separately, congruent risk patients’ post-visit suicidal behavior may have been differentially driven by their presenting complaint, which reasonably influenced the course of care at the ED visit. While the current paper did not distinguish subcategories of psychiatric complaints in the analysis, the risk of suicidal behavior after the ED visit may have been higher among patients presenting with more severe psychiatric problems (e.g., decompensated psychosis or severe major depressive disorder).

Next, our dependence on manual chart abstraction may have introduced limitations, such as interrater variability, and for many characteristics it limited the assessment of patient encounters to available clinical documentation only. Additionally, we were unable to further contextualize the composite rate in the incidental risk group as negligible risk patients were not followed up. This data would help inform how we think about patients with incidental risk. Regarding sample construction, since both were generated by randomly sampling unique patient encounters, a small number of patients may have been included more than once if they had multiple ED presentations. While we used the PSS-3, future research may find value in comparing screening outcomes between different screeners, where such data are available. Lastly, our exclusive reliance on the electronic health records of the participating health systems culminated in an under-count of suicide attempts since attempts not resulting in an ED visit, or those presenting to EDs outside participating systems, were not captured. While a limitation, this makes the findings even more alarming, meaning the 24-week 18% composite rate is likely an underestimate of the suicidal behavior that actually occurred.

5. CONCLUSION

While universal screening clearly identifies hidden risk in medical patients, this risk is not being further evaluated by a trained behavioral health professional, and interventions, even low intensity ones like post-discharge resources for behavioral health and crisis support, are not being delivered. This is true even in EDs with a focus on improving suicide-related care. Incidental risk patients are at elevated risk for serious suicidal behavior after the ED visit, though our findings that this risk is less than those with congruent risk contradicts findings from ED-SAFE 1, which showed they were comparable.

Figure 1. Cross-Sectional and Longitudinal Analytic Samples Created from ED-SAFE 2.

Study sample flowchart. ED-SAFE 2, Emergency Department Safety Assessment and Follow-up Evaluation 2; ED, emergency department; PSS-3, Patient Safety Screener-3.

Data Availability Statement:

Data for ED-SAFE 2 are available by request from the corresponding author.