Association between patterns of alcohol use and short-term risk of suicide attempt among patients with and without reported suicidal ideation

Abstract

Objective:

To evaluate the association between patterns of alcohol use and short-term risk of suicide attempt among patients with and without reported suicidal ideation.

Methods:

Kaiser Permanente Washington electronic health record data were used to identify mental health visits (1/1/2010–6/30/2015) with documented assessments for unhealthy alcohol use (AUDIT-C) and suicidal ideation (PHQ-9 ninth question). Logistic regression fit using generalized estimating equations were used to conduct visit-level analyses, accounting for correlation between individuals’ assessments. Separate models evaluated the association between (1) level of alcohol consumption and (2) frequency of heavy episodic drinking (HED), in combination with suicidal ideation (any versus none), with suicide attempt within 90 days following each visit. Primary models adjusted for age, gender, race/ethnicity and visit year.

Results:

Of 59,705 visits (43,706 unique patients), 372 (0.62%) were followed by a suicide attempt within 90 days. The risk of suicide attempt was significantly higher for patients reporting suicidal ideation across all levels of alcohol consumption compared to patients reporting low-level alcohol use and no suicidal ideation, particularly high-level use (OR 9.77, 95% CI, 6.23–15.34). Similarly, risk of suicide attempt was higher for patients reporting suicidal ideation across all levels of HED relative to those reporting no HED or suicidal ideation, particularly HED monthly or more (OR 6.80, 95% CI 4.77–9.72). Among patients reporting no suicidal ideation, no associations were observed.

Conclusions:

Findings underscore the potential value of offering alcohol-related care to patient reporting suicidal ideation. Additional strategies are needed to identify suicide risk among those reporting no suicidal ideation.

INTRODUCTION

Alcohol use and suicidal ideation are independently associated with increased suicide risk. Both intoxication resulting from heavy drinking and having an alcohol use disorder (AUD) diagnosis are associated with increased risk for fatal (Wilcox et al. 2004, Flensborg-Madsen et al. 2009, Caetano et al. 2013, Kaplan et al. 2013, Darvishi et al. 2015) and non-fatal suicide attempts (Cherpitel et al. 2004, Lejoyeux et al. 2008, Boenisch et al. 2010, Sung et al. 2016, Borges et al. 2017, Richards et al. 2019b). Suicidal ideation is also a known risk factor for suicide attempt (Kessler et al. 1999, Borges et al. 2006, McGirr et al. 2007, Nock et al. 2008a, Han et al. 2015). Researchers have demonstrated that depression assessment tools that include questions about suicidal ideation, like the 9-item Patient Health Questionnaire [PHQ-9] (Kroenke et al. 2001), can identify patients at risk of suicide attempt in the clinical setting (Simon et al. 2013, Louzon et al. 2016, Simon et al. 2016), where many seek care in the months prior to attempt (Luoma et al. 2002, Ahmedani et al. 2014).

Alcohol use also potentially impacts the relationship between suicidal thoughts and suicide attempts. In a recent qualitative study conducted among individuals who survived a recent suicide attempt after reporting no suicidal ideation at a recent healthcare visit on the ninth PHQ-9 question, which asks about frequency of “thoughts that you would be better off dead, or of hurting yourself” in the prior two weeks (Kroenke et al. 2001), alcohol use was often described in the context of making an unplanned suicide attempt (Richards et al. 2019c). Thus, identifying alcohol use patterns that increase risk for suicide attempt may help improve suicide risk identification among patients who report no suicidal ideation. However, alcohol use may also strengthen the relationship between suicidal thoughts and attempts by disinhibiting patients from making a suicide attempt (Lamis and Malone 2012).

The utility of combining information about patterns of alcohol use (rather than AUD diagnoses) and suicidal ideation for purposes of suicide prevention in clinical settings has not been studied. However, health systems are increasingly integrating assessments for unhealthy alcohol use and depression in response to national recommendations (Moyer and Preventive Services Task 2013, Crowley et al. 2015, Siu et al. 2016, National Committee for Quality Assurance 2018, National Council for Behavioral Health 2018), generating opportunities for investigating this issue. For example, many health systems routinely assesses patients for depression (including suicidality) and/or unhealthy alcohol use, using tools like the PHQ-9 and the Alcohol Use Identification Disorders Test Consumption [AUDIT-C] (Bush et al. 1998, Bradley et al. 2007). Because the AUDIT-C measures high-level alcohol use and heavy episodic drinking, which causes intoxication, it may be particularly useful for identifying patients at high risk of unplanned suicide attempts while intoxicated.

Therefore, among a population of adult patients receiving mental health specialty care, we evaluated the association between patterns of alcohol use, reported via AUDIT-C, and short-term risk of suicide attempt in combination with reported suicidal ideation. Specifically, our primary analysis tested whether greater levels of alcohol consumption and/or increased frequency of heavy episodic drinking identified individuals at higher risk of suicide attempt among patients reporting suicidal ideation (any versus none). Based on increased risk of suicide attempt associated with intoxication (Cherpitel et al. 2004, Lejoyeux et al. 2008, Caetano et al. 2013, Kaplan et al. 2013, Borges et al. 2017), higher-level alcohol use and more frequent heavy episodic drinking was hypothesized to be associated with greater risk of suicide attempt than low-level use and no heavy episodic drinking, especially among patients reporting no suicidal ideation.

METHODS

Data Source & Sample

Data sources included electronic health records [EHR], insurance claims, enrollment information, and cause-of-death data from state death certificates for patients at Kaiser Permanente Washington [KPWA], a large regional healthcare system. The dataset included adult (ages≥18) outpatient mental health specialty visits between 1/1/2010 and 6/30/2015. Data included sociodemographic information, inpatient and outpatient visit diagnosis codes and patient-reported measures—specifically, the AUDIT-C (Bush et al. 1998, Bradley et al. 2007) and PHQ-9 ninth question about suicidal ideation (Kroenke et al. 2001). Index visits included outpatient visits with documented AUDIT-C and PHQ-9 ninth question response on the same visit-day among patients enrolled for the 90-day period following the visit (i.e. did not disenroll or die from causes other than suicide attempt) (Supplement Figure S1). The sample was further limited to patient visits that occurred at least nine months apart, because: 1) the AUDIT-C asks individuals to report on alcohol consumption in the year prior, and 2) this approach approximates the scenario when individuals are assessed for unhealthy alcohol use annually, now common practice in some large health systems, including KPWA (Bradley et al. 2006, Glass et al. 2018). The KPWA IRB approved this research evaluation.

Measures

Predictor Measures: Patterns of Alcohol Use.

The three-item AUDIT-C questionnaire is a validated screen for unhealthy alcohol use that measures quantity and frequency of average consumption and frequency of heavy episodic drinking (HED) (formerly termed “binge drinking” (Hingson et al. 2017, Centers for Disease Control and Prevention 2018)), resulting in a total score of 0–12. Higher scores reflect greater alcohol consumption and consequences (Rubinsky et al. 2013). The AUDIT-C was used to assess two clinically meaningful patterns of alcohol use, level of alcohol consumption and frequency of HED, both of which are associated with serious health consequences (Bryson et al. 2008, Kinder et al. 2009, Williams et al. 2010, Bradley et al. 2011, Harris et al. 2011, Lembke et al. 2011, Williams et al. 2012a), unintentional injury and violence (World Health Organization. Management of Substance Abuse Unit 2014), and AUD (Saha et al. 2006, Saha et al. 2007). Level of alcohol consumption categories were defined using gender-specific AUDIT-C score thresholds: nondrinking, score 0; low-level, score 1–2 or 1–3 (women, men); moderate-level, score 3–7 or 4–7 (women, men), and high-level, score 8–12 (both women and men). Frequency of HED was measured using responses to the third AUDIT-C question, which asks patients to report how often they consume 6 or more drinks on one occasion. Response options include “never,” “less than monthly,” “monthly,” “weekly,” or “daily or almost daily,” and categories were collapsed into three groups to measure risk of suicide attempt associated with HED frequency less than monthly and monthly or more (compared to “never”).

Effect modifier:

Suicidal ideation was defined as a binary indicator variable using responses to the ninth PHQ-9 question, which asks about the frequency of thoughts of death or self-harm with response options: “not at all,” “several days,” “more than half,” or “nearly every day” (Kroenke et al. 2001, Uebelacker et al. 2011). The last three response options were combined to create a suicidal ideation indicator to evaluate the risk of suicide attempt across patterns of alcohol use in combination with the presence or absence of any suicidal ideation.

Primary Outcome:

Suicide attempts (fatal or non-fatal) were ascertained from state death certificates (fatal) and EHR and insurance claim data (non-fatal). Suicide deaths were identified, following common recommendations (Bakst et al. 2016, Cox et al. 2017), as any mortality codes (ICD-10) of self-inflicted injury (X60–X84) or injury/poisoning with undetermined intent (Y10–Y34). Non-fatal suicide attempts were identified using cause of injury codes (ICD-9-CM) indicating intentional self-harm (E950–E958) or undetermined intent (E980–E989), based on prior research, which also demonstrated high and consistent rates of E-code use at KPWA (Simon et al. 2018).

Other Measures included demographic characteristics known to be associated with alcohol use and suicide attempt, including age, sex, and race/ethnicity (Nock et al. 2008b), as well as common comorbidities and mental health conditions that increase suicide risk (Nock et al. 2010). Demographics were extracted from the medical record. Diagnostic codes (ICD-9-CM) within 365 days prior to the index visit were used to measure mental and physical health comorbidity, specifically anxiety, depressive, or serious mental illness disorders (bipolar, schizophrenia, other psychosis or personality disorders), suicide attempts, and Charlson comorbidity index score (Klabunde et al. 2000). For descriptive purposes, we also created an indicator for presence/absence of an AUD diagnosis (within 365 days prior to the index visit) and the PHQ-8 score recorded at the index visit.

Statistical Analysis

Patient-level descriptive statistics were calculated (using the first visit in study period) to describe the sample. We inspected all variables for missing data and compared demographic and clinical characteristics of the analytic sample (i.e. patients with a documented PHQ-9 and AUDIT-C) to the general population of patients who received mental health specialty care from a KPWA provider during the study period. Separate visit-level models were fit to evaluate the risk of suicide attempt within 90 days of the index visit associated with 1) level of alcohol consumption and 2) HED frequency, among patients with and without reported suicidal ideation (i.e., PHQ-9 ninth question response 0 versus 1–3). Visit-level logistic regression models were fit using generalized estimating equations (Zeger et al. 1988), to account for the correlation between multiple assessments for individual patients (Zeger et al. 1988, Zeger and Liang 1992). Because mental and physical health comorbidities could be caused or exacerbated by alcohol use, primary models adjusted only for demographics and visit year. All analyses were performed using Stata/MP 15.0 (StataCorp LLC 2017). Marginal suicide attempt prevalences were estimated from the primary model using the covariate distribution (i.e., Stata “margins” command), and presented graphically across categorical measures of alcohol use stratified by suicidal ideation. Secondary models additionally adjusted for indicators of past-year mental health diagnoses (depressive, anxiety, and serious mental illness disorders), past-year suicide attempt and Charlson score, to evaluate the utility of the AUDIT-C (in combination with suicidal ideation reported on the PHQ-9) for identifying increased risk of suicide attempt, among patients with similar mental health and medical comorbidity (in addition to demographics). The AUD diagnosis indicator and PHQ-8 score were not included in secondary analyses due to strong correlation with alcohol use and suicidal ideation (Dawson et al. 2005, Uebelacker et al. 2011). All models included the binary indicator for any suicidal ideation, reported on the PHQ-9 ninth question, and a multiplicative interaction term between this suicidal ideation indicator and the predictor of interest (level of alcohol consumption or HED frequency). Consistent with prior studies, (Au et al. 2007, Harris et al. 2009, Kinder et al. 2009, Williams et al. 2012b) and study hypotheses, models assessing level of alcohol consumption, were fit with low-level drinking (rather than non-drinking) as the referent group. This approach is aligned with findings of multiple prior studies that have demonstrated a non-linear association (i.e. U or J shaped curve) between levels of alcohol use and risk of adverse health consequences (Rodgers et al. 2000, Connor 2006), a pattern hypothesized to relate to residual confounding (Jackson et al. 2005) and the fact that people who do not drink include persons who abstain from use over the course of a lifetime and those who abstain for health-related reasons (a so called “sick quitter” effect) (Shaper et al. 1988, Klatsky and Udaltsova 2013). The latter issue is particularly relevant because the past-year timeframe of the AUDIT-C does distinguish these groups from each other (Bush et al. 1998, Bradley et al. 2007). For models assessing HED frequency, reporting “never” HED was used as the referent group. For all models, standard errors were calculated using the robust sandwich estimator (Zeger and Liang 1992), and odds ratios were used to approximate relative risk because suicide attempt was rare (Schechtman 2002). Confidence intervals (alpha 0.05) and p-values associated with two-sided Wald tests (Wald 1943) comparing predictor categories are presented and Wald heterogeneity tests, which tested the interaction between the indicator for suicidal ideation with all alcohol consumption levels.

RESULTS

Patient Characteristics

Among all patients in the analytic sample (N=43,706), the majority were women (63.7%), white (77.9%) and aged 18 to 90 years (mean=43.2) (Table 1). At the first visit in the study period, 21.9% reported non-drinking, and 38.7%, 34.6%, and 4.8% reported low-, moderate-, and high-level drinking, respectively. Further, 65.0% reported no HED in the past year, 22.0% reported HED less than monthly, and 13.0% reported HED monthly or more. Over half the sample (55.2%) had moderate to severe depressive symptoms (PHQ-8 score ≥ 10) and past-year prevalence of mental health diagnoses was 58.5%, 71.1% and 14.1% for anxiety, depressive and serious mental illnesses, respectively. Among all patients at their first eligible visit, 1.2% had a past-year suicide attempt diagnosis and 3.9% had a past-year AUD diagnosis.

Table 1:

Patient characteristics and alcohol use measures stratified by indicator of any patient reported suicidal ideation, at the time of the first outpatient mental health visit within study period.

	Total Sample N=43706		Patient Reported Suicidal Ideation Frequency (‡PHQ9 Q9)
			None N=32430		Any N=11276
Age (M, SD)	43.2	16.6	43.9	16.6	41.3	16.5
Male (N, %)	15878	36.3%	11641	35.9%	4237	37.6%
Race (N, %)
White, non-Hispanic	34050	77.9%	25646	79.0%	8404	74.5%
Black, non-Hispanic	1964	4.5%	1412	4.4%	552	4.9%
Asian	1953	4.5%	1353	4.2%	600	5.3%
Hispanic	1492	3.4%	1048	3.2%	444	3.9%
Other	2666	6.1%	1863	5.7%	803	7.1%
Unknown	1581	3.6%	1108	3.4%	473	4.2%
Alcohol Consumption Level† (N, %)
Nondrinking	9570	21.9%	6959	21.5%	2611	23.2%
Low-Level	16947	38.7%	12917	39.8%	4030	35.7%
Moderate-Level	15113	34.6%	11327	34.9%	3786	33.6%
High-Level	2076	4.8%	1227	3.8%	849	7.5%
Heavy Episodic Drinking Frequency†† (N, %)
Never	28419	65.0%	21663	66.8%	6756	59.9%
Less than monthly	9604	22.0%	7001	21.6%	2603	23.1%
Monthly or More	5683	13.0%	3766	11.6%	1917	17.0%
Level of Depressive Symptoms‡‡ (N, %)
Minimal or None	8424	19.3%	8188	25.3%	236	2.1%
Mild	11124	25.5%	9808	30.3%	1316	11.7%
Moderate	10363	23.7%	7715	23.8%	2648	23.5%
Moderately Severe	8429	19.3%	4702	14.5%	3727	33.1%
Severe	5310	12.2%	1973	6.1%	3337	29.6%
Mental Health Diagnoses, Prior Year (N, %)
Anxiety Disorder	25602	58.5%	18465	56.9%	7134	63.2%
Depressive Disorder	31068	71.1%	21340	65.8%	9728	86.3%
Serious Mental Illness Diagnosis*	6141	14.1%	4142	12.8%	1999	17.7%
Suicide Attempt	526	1.2%	204	0.6%	322	2.9%
Charlson Score**	0.5	1.2	0.5	1.1	0.5	1.2
Alcohol Use Disorder Diagnosis, prior year (N, %)	1700	3.9%	1091	3.4%	609	5.4%

^†AUDIT-C: Nondrinker=score 0; Low-Level= Score 1–2 Women, 1–3 Men; Moderate-Level=Score 3–7 Women, 4–7 Men; High -Level=Score 8–12 Women & Men

^††AUDIT-C Question 3: Never=Score 0, Less than monthly=Score 1, Monthly or More=Score 2–4

^‡PHQ-9 Question 9 regarding frequency of self-harm thoughts in prior 2 weeks: None=Score 0, Any=Score 1–3 (some of the days to nearly every day)

^‡‡PHQ-8: Minimal or None=Score 0, Mild=Score 5–9, Moderately Severe= Score 15–19, Severe=Score 20–24; missing for individuals (N=56) with more than 2 missing values

^*Diagnosis of bipolar, schizophrenia, other psychosis or personality disorders

^**Missing for individuals (N=2428) with no in-patient or out-patient utilization in prior year

About a quarter of the sample (25.8%, N=11,276) reported suicidal ideation. Patient demographic characteristics were similar across report of suicidal ideation, but patient-reported measures of alcohol use, depressive symptoms and past-year diagnoses differed. Patients reporting some level of suicidal ideation reported more severe depressive symptoms than those with no ideation reported. A higher proportion of patients reporting suicidal ideation also reported high-level alcohol consumption and HED frequency monthly or more than those reporting no suicidal ideation. Prevalence of all past-year diagnoses, including AUD and suicide attempts, were higher among patients reporting suicidal ideation.

About two-thirds of the overall population of patients who received outpatient mental health care during the study period completed an AUDIT-C and PHQ-9 during a visit and very few (N=79) were missing outcome data in the subsequent 90 days (Supplement Figure S1). The demographic characteristics and past-year mental health diagnoses of the analytic sample used for this study were comparable to the overall study population (Supplement Table S2).

Association Between Patterns of Alcohol Use and Risk of Suicide Attempt by Reported Suicidal Ideation

Of 59,705 patient visits with a documented AUDIT-C and PHQ-9 ninth question response, 372 (62.3 per 10,000) were followed by a suicide attempt (353 nonfatal, 19 deaths) within 90 days, including 241 (163 per 10,000) among patients reporting suicidal ideation and 131 (29 per 10,000) among patients reporting no suicidal ideation. Counts across categorical measures of alcohol use and suicidal ideation are presented in Table 2.

Table 2:

Counts of suicide attempts observed within 90-days following the index visits with documented AUDIT-C and PHQ-9 assessments, stratified by binary indicator of patient-reported suicidal ideation

	Total Visit Sample			Patient-Reported Suicidal Ideation Frequency
				None			Any
	Visits N=59705	Suicide Attempts N=372		Visits N=44941	Suicide Attempts N=131		Visits N=14764	Suicide Attempts N=241
Alcohol Consumption Level†	Count	N	%	Count	N	%	Count	N	%
Nondrinking	11688	78	0.67%	8494	27	0.32%	3194	51	1.60%
Low-Level	25595	142	0.55%	19870	58	0.29%	5725	84	1.47%
Moderate-Level	19976	114	0.57%	15121	38	0.25%	4855	76	1.57%
High-Level	2446	38	1.55%	1456	8	0.55%	990	30	3.03%
Heavy Episodic Drinking Frequency††	Count	N	%	Count	N	%	Count	N	%
Never	40572	223	0.55%	31339	91	0.29%	9233	132	1.43%
Less than monthly	12343	80	0.65%	9083	23	0.25%	3260	57	1.75%
Monthly or More	6790	69	1.02%	4519	17	0.38%	2271	52	2.29%

^†AUDIT-C: Nondrinker=score 0; Low-Level= Score 1–2 Women, 1–3 Men; Moderate-Level=Score 3–7 Women, 4–7 Men; High -Level=Score 8–12 Women & Men

^††AUDIT-C Question 3: Never=Score 0, Less than monthly=Score 1, Monthly or More=Score 2–4

^‡PHQ-9 Question 9 regarding frequency of self-harm thoughts in prior 2 weeks: None=Score 0, Any=Score 1–3 (some of the days to nearly every day)

Level of Alcohol Consumption.

The marginal prevalences of suicide attempt, estimated from the primary model adjusted for demographics and visit year, were substantially higher for patients reporting suicidal ideation across all levels of alcohol consumption, particularly for those reporting high-level use (Figure 1). Primary model results (Table 3, Model A) indicated that among patients reporting suicidal ideation, the risk of suicide attempt was significantly increased for those reporting high-level alcohol use (OR 9.77, 95% CI, 6.23–15.34), moderate-level use (OR: 4.94, 95% CI 3.49–6.98), low-level use (OR: 4.97, 95% CI 3.55–6.95), and non-drinking (OR 5.86, 95% CI 4.00–8.58) relative to those with low-level alcohol use and no suicidal ideation. However, among patients reporting no suicidal ideation, the suicide attempt risk was not significantly increased for any level of alcohol use, including high-level (OR 1.80, 95% CI 0.86–3.80), moderate-level (OR 0.82, 95% CI 0.54–1.23), and non-drinking (OR 1.20, 95% CI 0.75–1.89) relative to low-level alcohol use. After further adjustment for past-year mental health and medical comorbidity (Table 3, Model B), results were similar to the primary analysis. The Wald test for heterogeneity, which tested the interaction between suicidal ideation and all alcohol consumption levels, was statistically significant in both primary (A) and secondary (B) models (p<0.0001), confirming the relationship between alcohol consumption and suicide attempt risk was not the same for those with and without suicidal ideation.

Figure 1:

Predicted prevalence of suicide attempt by alcohol consumption level for patients with and without suicidal ideation, adjusted for age, sex, race/ethnicity and visit year

Table 3:

Estimated risk of suicide attempt by alcohol consumption level and frequency of heavy episodic drinking by indicator for any patient reported suicidal ideation, within 90 days of assessment during 59,705 patient visits (N= 43,706 individuals)

			A: Primary Model*			B: Secondary Model**
			OR	95% CI	p-value	OR	95% CI	p-value
	Alcohol Consumption Level†
Suicidal Ideation‡	None	Nondrinking	1.20	(0.75–1.89)	0.447	0.99	(0.62–1.61)	0.995
		Low-Level	Ref			Ref
		Moderate-Level	0.82	(0.54–1.23)	0.332	0.90	(0.59–1.37)	0.633
		High-Level	1.80	(0.86–3.80)	0.120	1.86	(0.88–3.94)	0.106
	Any	Nondrinking	5.86	(4.00–8.58)	<.001	3.89	(2.60–5.81)	<.001
		Low-Level	4.97	(3.55–6.95)	<.001	3.53	(2.47–5.04)	<.001
		Moderate-Level	4.94	(3.49–6.98)	<.001	3.57	(2.47–5.16)	<.001
		High-Level	9.77	(6.23–15.34)	<.001	6.07	(3.72–9.89)	<.001
Suicidal Ideation ‡	Heavy Episodic Drinking Frequency††
	None	Never	Ref			Ref
		Less than monthly	0.75	(0.47–1.19)	0.221	0.87	(0.54–1.38)	0.546
		Monthly or More	1.12	(0.66–1.90)	0.676	1.19	(0.69–2.05)	0.533
	Any	Never	4.86	(3.71–6.36)	<.001	3.54	(2.66–4.72)	<.001
		Less than monthly	5.16	(3.67–7.26)	<.001	3.74	(2.59–5.40)	<.001
		Monthly or More	6.80	(4.77–9.72)	<.001	4.72	(3.21–6.94)	<.001

^*Adjusted for visit year, age, sex, race/ethnicity (including unknown as separate category)

^**Additionally adjusted for indicators for past-year diagnosis indicators for depression, anxiety, serious mental illness disorders) and a suicide attempt, and Charlson score, excluding patients (N=2428) with no in-patient or out-patient utilization in prior year

^†AUDIT-C: Nondrinker=score 0; Low-Level= Score 1–2 Women, 1–3 Men; Moderate-Level=Score 3–7 Women, 4–7 Men; High -Level=Score 8–12 Women & Men

^††AUDIT-C Question 3: Never=Score 0, Less than monthly=Score 1, Monthly or More=Score 2–4

^‡PHQ-9 Question 9 regarding frequency of self-harm thoughts in prior 2 weeks: None=Score 0, Any=Score 1–3 (some of the days to nearly every day)

Wald heterogeneity tests for each model, testing interaction between the suicidal ideation indicator and all categories of each measure of alcohol use, were significant p<0.0001

Heavy Episodic Drinking.

The marginal prevalences of suicide attempt, estimated from the primary models adjusted for demographics and visit year, were substantially higher for patients reporting suicidal ideation across all levels of HED frequency and elevated for those reporting HED monthly or more (Figure 2). Primary model results (Table 3, Model A2) also indicated that, among patients reporting suicidal ideation, the risk of suicide attempt was significantly increased for those reporting HED monthly or more (OR 6.80, 95% CI 4.77–9.72), less than monthly (OR 5.16, 95% CI, 3.67–7.26) and no HED (OR 4.86, 95% CI 3.71–6.36), relative to those reporting no HED and no suicidal ideation. However, among patients reporting no suicidal ideation, suicide attempt risk was not significantly increased for those reporting any level of HED, including monthly or more (OR 1.12, 95% CI 0.66 – 1.90) and less than monthly (OR 0.75, 95% CI 0.47 – 1.19), relative to those reporting no HED. Results of secondary HED analysis (Table 3, Model B) were similar. The Wald test of the interaction between suicidal ideation and HED frequency was statistically significant in both primary (A) and secondary (B) models (p<0.0001).

Figure 2:

Predicted prevalence of suicide attempt by frequency of heavy episodic drinking [HED] for patients with and without suicidal ideation, adjusted for age, sex, race/ethnicity and visit year

DISCUSSION

In this large study of adult patients who completed an AUDIT-C and PHQ-9 during outpatient visits with a mental health provider, information about alcohol use patterns combined with information about suicidal ideation identified increased short-term suicide attempt risk, but only among patients reporting suicidal ideation. Patients reporting high-level alcohol use and suicidal ideation were nearly ten times more likely to attempt suicide than those with low-level alcohol use and no suicidal ideation. Similarly, patients reporting HED monthly or more in combination with suicidal ideation were nearly seven times more likely to attempt suicide than those reporting “never” and no suicidal ideation. Notably, patients reporting suicidal ideation in combination with nondrinking, low-level, and moderate-level consumption had a similar increased short-term likelihood of suicide attempt, relative to patients reporting low-level drinking and no suicidal ideation. Similarly, patients reporting suicidal ideation in combination with never or less than monthly HED had similar increased likelihood of suicide attempt. These findings were robust to adjustment for other factors known to predict suicide attempt, including mental health and medical comorbidities and prior year suicide attempt. However, patterns of alcohol use reported on the AUDIT-C did not appear to be useful for identifying increased short-term suicide attempt risk among patients not reporting suicidal ideation.

Our findings build on prior research that has demonstrated associations between alcohol use and suicide attempt and death (Powell et al. 2001, Cherpitel et al. 2004, Lejoyeux et al. 2008, Bagge et al. 2013, Caetano et al. 2013, Richards et al. 2019b). These findings support prospective identification of patients with suicidal ideation and risky patterns of alcohol use, particularly high-level consumption and HED more than monthly. Combining information from brief standard assessments used for screening and monitoring purposes, like the AUDIT-C and PHQ-9, can facilitate such identification (Richards et al. 2019b). Findings from the present study further underscore the potential value of addressing suicidal ideation and risky patterns of alcohol use together for purposes of suicide prevention (Kalk et al. 2019). For example, providers may counsel patients that high-level consumption and HED more than monthly may substantially increase risk of suicide attempt, by intensifying suicidal ideation (Bagge et al. 2014), increasing impulsivity and constricting coping/problem solving (Hufford 2001, Conner et al. 2008). Providers may also recommend treatments for alcohol use as part of their care plan for patients identified as being at risk of suicide attempt, including psychosocial interventions (Campbell et al. 2018) and/or pharmacotherapy (Holt and Tobin 2018). Relatedly, routinely assessing patients for suicidal thoughts in alcohol treatment settings may be a useful suicide prevention strategy (Conner et al. 2014).

Findings for patients reporting no suicidal ideation were contrary to our hypothesis generated from qualitative work (Richards et al. 2019c). Specifically, there were no significant differences in suicide attempt risk associated with higher-level alcohol use and more frequent HED among patients reporting no suicidal ideation. It is unclear what accounts for the lack of association. However, because both alcohol use and suicide are substantially stigmatized conditions (Room 2005, Keyes et al. 2010, Clement et al. 2015), it is possible that these findings relate to social desirability and measurement bias—specifically, some patients may be unwilling to disclose both suicidal ideation and alcohol consumption, and, thus measurement of these factors may be limited in these patients.

Further work is needed to understand the lack of association observed here and improve our ability to identify suicide risk among patients reporting no suicidal ideation. Though the overall suicide attempt rate was much lower among patients reporting no suicidal ideation (29 versus 163 per 10,000 patient visits), suicide attempts in this group still accounted for about a third of all suicide attempts observed within 90 days of a visit. One promising new approach includes development of predictive algorithms that identify patients at risk of suicide by harnessing large combinations of discrete EHR data elements, such as indicators for different types of healthcare utilization (e.g. inpatient/outpatient, primary care, specialty care), diagnosis codes, and pharmacy records, as well as sociodemographic information (Kessler et al. 2017, Simon et al. 2018). These predictive algorithms may be particularly powerful when health systems have access to comprehensive longitudinal medical records for their patients. However, in situations when health systems provide short-term care to transient populations, other strategies will be needed. One promising strategy could be to focus on improving screening and risk assessment. Patients’ fears of disclosing stigmatizing information about suicidality may be alleviated by normalizing disclosures about suicidal thoughts, in combination with providers’ expressions of caring and active listening, without panic or raising unnecessary alarm (Ganzini et al. 2013, Richards et al. 2019a).

Limitations of this study include generalizability; KPWA cares for a population of patients primarily insured by employer-sponsored health insurance and Medicare. Future research is needed to assess the relationship between alcohol use, suicidal ideation, and suicide risk in other organizations and settings (e.g. primary care). Next, two-thirds of the overall population of patients receiving mental healthcare during the study period completed the AUDIT-C and PHQ-9 ninth question. We used complete case analyses; while no strong differences were observed between those included and excluded from our analyses, unobserved differences may remain. Relatedly, non-fatal suicide attempts were assessed from EHR data and some events may have been missed or misclassified by providers. Furthermore, the AUDIT-C asks about alcohol use in the year prior to screening, which does not distinguish people with lifetime abstinence from those who may have stopped drinking due to problems with alcohol or other illnesses, thus limiting interpretation of study results for the non-drinking population. Finally, due to the small number of suicide deaths, we were not able to examine this outcome separately from suicide attempts or to evaluate findings in specific demographic sub-groups (Nock et al. 2008b). Future research is needed to confirm and extend these findings.

Despite these limitations, this is the first study investigating the utility of combining information about patterns of alcohol use and suicidal ideation for purposes of identifying patients at high-risk of suicide in the clinical setting. These study findings suggest one important future direction for suicide prevention research is integration and evaluation of evidence-based alcohol-related care into standard care for patients reporting suicidal ideation. This may include integration of repeated brief interventions for unhealthy alcohol use (Bischof et al. 2008, Guth et al. 2008) and/or alcohol-related treatment (e.g. psychotherapy, pharmacotherapy) into collaborative safety planning interventions with patients at high-risk of suicide (Stanley and Brown 2012, Bryan et al. 2018), including those in transition to outpatient settings following inpatient psychiatric hospitalizations (Conner et al. 2014). Study findings also suggest additional strategies will be needed to identify patients at high-risk of suicide who do not report suicidal ideation. This may include identification via suicide risk prediction algorithms deployed in large health systems (Kessler et al. 2017, Simon et al. 2018), or other strategies designed to reduce stigma and encourage disclosures about both alcohol use and suicidal ideation (Ganzini et al. 2013, Richards et al. 2019a).

Conclusions

In this large population of adults receiving outpatient mental health specialty care, results from the AUDIT-C were useful for identifying patterns of alcohol use associated with increased suicide-risk among patients reporting suicidal ideation on the PHQ-9 ninth question. These results suggest future research should evaluate patient outcomes associated with integrating evidence-based alcohol-related care into standard suicide-related care. Identification of patients at risk of suicide who do not report suicidal ideation will require testing other strategies to improve the suicide risk assessment process in healthcare settings.

Supplementary Material

Supplement Table S2

Table S2: Characteristics of Adult Population of Patients Receiving Mental Health Care Compared to Study Sample Screened, at the time of the first outpatient mental health visit within study period.

Supplemental Figure S1

Figure S1: Flow diagram analytic sampling: adult patients visits to a mental health provider that included an AUDIT-C and PHQ-9 with and outcome data available 90-days following the AUDIT-C 1/1/2010-6/30/2015