Introduction

Suicide is a widespread public health concern (Centers for Disease Control and Prevention, 2020), and there is a need to better understand key pathways conferring near-term suicide risk. Notably, alcohol use may be an important proximal factor. While problematic alcohol use has been established as a distal risk factor for suicide ideation, attempts, and deaths in both youth and adults (Darvishi et al., 2015; Sellers et al., 2019; Wilcox et al., 2004), acute alcohol use may also demonstrate proximal risk for suicidal thoughts and behaviors. The likelihood of a suicide attempt increases on days when alcohol is consumed (Borges et al., 2017), with the 24-h period following alcohol intoxication being associated with a sevenfold increased risk for suicidal behavior (Esang & Ahmed, 2018). Alcohol use may also contribute to the facilitation of death by suicide. Alcohol stands out as one of the most frequently detected substances in suicide decedents, with over one-third of decedents having alcohol present in their bloodstream (Centers for Disease Control and Prevention, 2014). Despite the potential salience of this association, limited research has aimed to better understand the near-term link between alcohol use and suicide risk. Given the potential implications for suicide prevention efforts, it is pertinent to understand which factors, or specific contexts (e.g., negative mood states), increase the likelihood of alcohol use among those at high risk for future suicidal thoughts and behavior (i.e., those with prior suicidal thoughts; Ribeiro et al., 2016).

There are several candidate risk factors that may be influential in predicting daily alcohol use experiences among those at elevated suicide risk – the most notable is heightened affect. Both alcohol use and suicidal thinking have been proposed to serve an emotion regulation function (i.e., Baumeister, 1990; Sayette, 2017), and it may be a link in their co-occurrence. For example, drinking during negative mood states has been associated with suicidal behavior, even among those without recent suicidal thinking (Schilling et al., 2009). Furthermore, empirical research has supported positive momentary associations between negative affect, drinking co** motivations, and alcohol use (Dvorak et al., 2014), as well as between negative affect and suicidal ideation (i.e., Gee et al., 2020). These empirical findings support theoretical models linking alcohol use and suicide risk (Conner et al., 2008; Hufford, 2001; Lamis & Malone, 2012), both suggesting that momentary negative affect may be uniquely related to alcohol use among those with elevated suicide risk.

Negative interpersonal experiences may also be contributing to alcohol use among those at elevated suicide risk. The experiences of thwarted belongingness (TB; i.e., an unmet need to belong) and perceived burdensomeness (PB; i.e., the perception that one’s death is worth more than one’s life to others) have been theoretically and empirically supported as risk factors for suicidal thinking (i.e., Chu et al., 2017; Joiner, 2005; Van Orden et al., 2010). There have also been suggestions that TB and PB may play a role in the association between alcohol use and suicide risk. For example, the association between alcohol use and suicide risk has been explained, in part, by the presence of interpersonal problems (i.e., Conner et al., 2008; Pompili et al., 2010). Research has supported the importance of TB in predicting suicide risk among those with a substance use disorder (Conner et al., 2007; You et al., 2011) and that TB and PB may faciliate the relationship between alcohol use and suicidal risk among college students (Lamis & Malone, 2011) and firefighters (Gallyer et al., 2018). While much of this research has utilized cross-sectional and longitudinal designs, recent studies demonstrate that momentary TB and PB have been concurrently and prospectively linked to momentary suicidal ideation (Hallensleben et al., 2019; Jacobucci et al., 2022; Kleiman et al., 2017). Despite the potential influence of TB and PB contributing to alcohol use among a high-risk population, these associations have yet to be examined in predicting alcohol use at the momentary level among high-risk individuals, which would provide great insight into the alcohol use-suicide risk proximal pathway.

Thus, this study’s primary aim was to investigate the relationship between identified risk factors (i.e., daily affect, TB, PB), response characteristics (i.e., day of the week, time of day), and momentary alcohol use experiences (i.e., urge to drink alcohol; alcohol use), and to determine how suicide risk history may influence these relationships. More specifically, we examined these associations among those with and without a history of suicidal ideation given that prior suicidal thinking increases risk for future suicidal thoughts and behavior (Ribeiro et al., 2016) and that the context of alcohol use may impact suicide risk, even among those with distal risk history (Schilling et al., 2009). Intensive time sampling methods, such as ecological momentary assessment (EMA), allow for the nuanced investigation of such daily dynamics. Indeed, EMA has seen a rise in the field of suicide (for reviews, see Ammerman & Law, 2022; Sedano-Capdevila et al., 2021) as this approach can reduce retrospective recall bias (Gratch et al., 2021) and capture fluctuation in cognitive-affective experiences over short timescales. Leveraging EMA data, we hypothesized that there would be an association between negative affect, TB, and PB with both alcohol use urges and alcohol use, and that these relationships would be unique to those with a suicidal ideation history.

Method

Participants and procedures

To ensure adequate variability in alcohol use experiences at the daily level, we utilized data among a sample with elevated alcohol and substance use. Undergraduate students (n = 56) were recruited who met the following inclusion criteria: 18 +; enrolled as an undergraduate; and met screening criteria for probable alcohol use (i.e., > 7 on the Alcohol Use Disorder Identification Test; DeMartini & Carey, 2012; Saunders et al., 1993) or substance use (i.e., > 6 on the Drug Use Disorder Identification Test; Berman et al., 2007; Hildebrand, 2015) disorder. Participants completed 21 days of EMA (4 surveys per day within a 12-h window; randomized within 3-h blocks, occurring 30-min apart). Participants with less than 20 data points were excluded from the analysis, resulting in a final sample of 51 participants and 3,562 data points. This was done to ensure adequate power for within-person analyses. Although specific power analysis recommendations do not exist given the categorical nature of our outcome variables, recommendations for conducting individual-level analyses with continuous variables in group iterative multiple model estimation (GIMME; Gates & Molenaar, 2012) suggest at least 60 time points. Thus, our decision aimed to balance power for within-person analysis and the presence of all levels of the outcome variables.

Across all participants in the study, the average compliance over the EMA period was 79.4% (SD = 22.6%; range = 22.7–100%). Participants’ average age was 20.12 (SD = 1.29), 71.2% identified as a cisgender woman (21.6% a cisgender man; 4.9% transgender/other; 1.9% preferred not to answer); 73.1% identified as White (7.7% Black, 1.9% Asian, 15.3% other or more than 1 race; 1.9% preferred not to answer); and 19.6% identified as Hispanic / Latinx. Of the final sample, 52.9% of participants reported a lifetime history of active suicidal ideation; 13.7% reported suicide plan history; and 7.8% reported suicide attempt history.

Measures

Daily affect

Momentary positive and negative affect were assessed via the 20-item Positive and Negative Affect Scale (PANAS) (Watson et al., 1988). Items assess participants’ affective experiences in the present moment on a 5-point Likert scale (1 = very slightly or not at all; 2 = a little; 3 = moderately; 4 = quite a bit; 5 = extremely), wherein 10 items are summed to create a composite score for positive affect and negative affect. The PANAS has been widely used in suicide-related EMA research (Gee et al., 2020).

Thwarted belongingness

Momentary thwarted belongingness was assessed using a composite of two items (i.e., “Like I do not belong”; “Lonely”), that examined how a participant felt in the present moment on a 5-point Likert scale (1 = very slightly or not at all; 2 = a little; 3 = moderately; 4 = quite a bit; 5 = extremely). These items (and composite) have been previously utilized in high-risk samples in suicide-related EMA research (i.e., Forkmann et al., 2018).

Perceived burdensomeness

Momentary perceived burdensomeness was assessed using a composite of two items (i.e., “Like a burden”; “Useless”), that examined how a participant felt in the present moment on a 5-point Likert scale (1 = very slightly or not at all; 2 = a little; 3 = moderately; 4 = quite a bit; 5 = extremely). These items (and composite) have been previously utilized in high-risk samples in suicide-related EMA research (i.e., Forkmann et al., 2018).

Alcohol use experiences

At each survey prompt, participants were asked, “Since the last prompt have you… Drank alcohol?” Response options included, No; Had the urge but did not; Yes. For the purpose of the present study, the response “Had the urge but did not” was coded as an alcohol use urge, referred to herein as “Urge”, and the response “Yes” was coded as alcohol use behavior, referred to herein as “Use”.

Lifetime history of suicidal ideation

The Self-Injurious Thoughts and Behaviors Interview (Nock et al., 2007) was used to assess for elevated suicide risk. For the purpose of this study, elevated suicide risk was defined as having a positive lifetime history of suicidal ideation, as assessed by a single item, “Have you ever had thoughts of killing yourself?” (response options: No; Yes).

Data analysis

To assess the influence of covariates, we utilized within-person standardization for each variable, allowing us to compare the within-person effects of each variable on subsequent time alcohol Urge and Use, as opposed to between-person effects (Schuurman et al., 2016). As baseline models, we used the brms package (Bürkner, 2017) to run two separate multilevel models, one for alcohol Urge and one for alcohol Use. The outcome at time t + 1 was modeled with each covariate assessed at time t, Urget, and a random effect for the intercept.

Instead of attempting to assess lifetime suicidal ideation (SI) history as a level-2 predictor for each covariate effect, we used a method that assessed group differences at a more global level. The generalized linear mixed-model (GLMM) tree method uses permutation tests to assess whether individuals with and without a SI history are meaningfully different with respect to the multilevel model parameters. Whereas GLMM tree is a form of decision trees that automatically partitions the sample into groups based on covariates, the use of this method, with this data, is equivalent to the use of a multiple-group mixed effects model as there is only two possible groups. The GLMM tree algorithm is available in the glmertree package (Fokkema et al., 2017).

To further assess degree of heterogeneity in the relationships between alcohol Urges and Use, risk factors, and lifetime SI history, we detail subgroup and individual-level analyses across multiple datasets. To accomplish this, we used GIMME (Gates & Molenaar, 2012), implemented as the gimme package (Lane & Gates, 2017) in R. Pertinent to our aims, subgroup-level models can be estimated (Gates et al., 2017), which is based on finding models specific to subgroups of individuals, which allows us to specify confirmatory subgroup-level models, using lifetime history of SI (data is subset to only those individuals with non-zero variance on each variable included in analysis; 11 without SI history, 16 with SI history). We modeled the relationship between momentary alcohol Urge and Use along with the same set of covariates (i.e., positive affect, negative affect, TB, and PB), with response characteristics of hour of day and day of response (i.e., weekend vs. weekday), specified as exogenous variables.

Results

Alcohol Urge was reported on 260 survey responses across 32 individuals; alcohol Use was reported on 367 survey responses across 47 individuals.

Assuming no group differences based on lifetime SI history, mixed-effects model estimates for each outcome are displayed in Tables 1 and 2. For the GLMM tree, splits did not occur for alcohol Use; however, for both alcohol Urget and alcohol Urget+1, individuals were partitioned into groups based on SI history. Firstly, across both sets of results, individuals with an SI history reported higher rates of alcohol Urges than those without an SI history. Note that this relationship did not hold for alcohol Use, where a logistic regression modeling alcohol Use with SI history had a coefficient near zero and p > 0.05. For modeling alcohol Urget+1, the main distinction (Table 3) is that individuals with an SI history had a stronger alcohol Urge autoregressive effect, with almost every other covariate effect near zero. In contrast, the autoregressive effect for those without an SI history was near zero, with stronger effects for positive affect, PB, TB, and whether it was the weekend. For modeling alcohol Urget, the main distinction was the effect of negative affect, which had a larger positive relationship with alcohol Urget among individuals without a SI history, whereas, for those with an SI history, PB had a stronger positive relationship with alcohol Urget.

Table 1 Baseline multilevel model addressing group-level analyses, with alcohol Urget+1 as the outcome
Full size table
Table 2 Baseline multilevel model addressing group-level analyses, with alcohol Uset+1 as the outcome
Full size table
Table 3 Individual-level analyses: GLMM tree results for modeling alcohol Urge
Full size table

For the subgroup analysis, the paths at the group, subgroup, and individual levels are displayed in Fig. 1. At the group level (solid black line), the only relationship found was a contemporaneous path from PB to negative affect. At the subgroup levels, individuals without an SI history had contemporaneous paths from negative affect and PB to TB. In contrast, individuals with an SI history had a contemporaneous relationship from TB to PB. There was significantly more heterogeneity at the individual level, with the individual path counts for each group displayed in Table 4. Among those with an SI history, there were a greater number of individuals with significant paths during weekends (i.e., significant relationships more likely to occur on weekends) and hours (i.e., significant associations more likely to occur later in the day) to both alcohol Urge and Use.

Fig. 1
figure 1

Subgroup analyses for alcohol Urge. Note: Those without a lifetime suicidal ideation history are presented on the left; those with a lifetime suicidal ideation history are presented on the right. Gray lines denote individual effects, while black lines denote effects at the group level, and green denotes subgroup level paths. Dashed lines denote lagged effects, while solid are contemporaneous effects. SI = suicidal ideation; TB = thwarted belongingness; PB = perceived burdensomeness; PA = positive affect; NA = negative affect; Urge = alcohol use urge; Use = alcohol use; Hour = hour of the day; Weekend = weekend versus weekday

Full size image
Table 4 Individual-level analyses: GIMME confirmatory subgroup analysis
Full size table

Discussion

The current study was the first study, to our knowledge, aimed at improving our understanding of the alcohol use—suicide risk proximal pathway by investigating the momentary relationships between key risk factors (i.e., negative affect, thwarted belongingness, perceived burdensomeness) and alcohol use experiences (i.e., Urges, Use) among those with a history of suicidal ideation. Our hypotheses were not supported; we did not find differences in associations between risk factors and alcohol use experiences among those with and without a lifetime SI history. However, findings at the subgroup and individual levels highlight the likely complexity of the proximal pathways between alcohol use experiences and suicide risk..

Foremost, there were no group differences in predicting alcohol Use. In examining alcohol Urges, there was a stronger concurrent association between negative affect and alcohol Urge among those without a lifetime SI history. This finding extends prior cross-sectional research finding that drinking while down was associated with suicide attempts among those not reporting prior-year suicidal thinking (Schilling et al., 2009) to a momentary level. Together, it may suggest that drinking in the context of negative affect may be indicative of negative outcomes, including suicidality, regardless of recent suicide risk markers. Alternatively, there was a stronger concurrent association between PB and alcohol Urge among individuals with a SI history. While prior research has supported the role of PB in the link between substance use and suicidal ideation in cross-sectional research (Baer et al., 2022; Gallyer et al., 2018), to our knowledge, this is the first study to consider the association with alcohol use urges and this link at the momentary level. While findings could potentially underscore the salience of the theoretically implicated suicide risk factors of TB and PB (Joiner, 2005; Van Orden et al., 2010), it also supports theoretical claims that interpersonal difficulties– potentially as a result of alcohol use– (Conner et al., 2008; Pompili et al., 2010), may facilitate alcohol use among those with prior suicidal thinking.

It is notable, however, that our pattern of prospective findings are in contrast to these conclusions: there was a stronger effect of both PB and TB on next timepoint alcohol Urge among those without a lifetime SI history. It is possible that there is a strong influence of PB and TB on alcohol use experiences, regardless of suicide risk history (e.g., Le et al., 2021), thus preventing the detection of such effects among those with an SI history given the small sample size. However, these findings were unexpected and require replication given they are not in line with prior research supporting the role of TB and PB in the alcohol use– suicide risk relationship (Gallyer et al., 2018; Lamis & Malone, 2011).

All group associations are important to consider in the context of several other potential explanations. It is possible that analyses were underpowered to detect some group-level risk relationships, particularly for alcohol Use. While some work suggests otherwise (Schilling et al., 2009), it may also be that lifetime SI history does not directly impact current (i.e., momentary) alcohol use experiences; it may be that imminent suicide risk elevations (i.e., increases in momentary suicidal thinking) may have an influence on current alcohol use experiences. Similarly, results from prior work (i.e., Gallyer et al., 2018; Lamis & Malone, 2011) may not have been extended to the current study given the timescale of data collection (i.e., daily versus cross-sectional). Finally, it is possible that the significant heterogeneity found in the risk associations (as demonstrated in subgroup and individual-level analyses) prevented a clear pattern of findings at the group level (e.g., Hoekstra et al., 2023).

Individual-level analyses demonstrated notable heterogeneity in associations between risk factors, alcohol Urges, alcohol Use, and SI history. However, a few patterns of results emerged. We see those with a lifetime SI history demonstrated a greater number of significant associations between risk factors and alcohol Urges and Use during the weekends (vs. weekdays) and later in a given day (vs. in the morning). Although less pronounced, there were also a small number of individuals with an SI history that displayed associations between PB and alcohol Urge (not present among anyone without a history). It will be necessary to consider these associations among a larger sample to determine if a subgroup of individuals demonstrating similar patterns of results emerges.

Interpretations of findings should be in the context of study limitations. While compliance rates were relatively high (~ 80%), it is possible that data collection did not capture all instances of SI and alcohol use experiences, particularly those that happened late into the evening (i.e., after the last prompt of the day) and were not reported on the next morning’s survey. It is also possible that relationships between risk factors and alcohol use experiences may have differed based on compliance; for example, lower compliance rates have been found among more clinically severe samples (Ammerman & Law, 2022), who may exhibit different patterns of risk relationships. Relatedly, our single-item assessment of alcohol use experiences may have impacted study findings, as it assessed experiences “since the last prompt”, whereas other variables assessed momentary experiences. Thus, future research examining these associations at a more fine-grained timescale would be beneficial. Our sample was fairly homogenous and not representative of all individuals with elevated alcohol use and suicide risk. Notably, the majority of our sample identify as women, preventing an examination of gender differences. While the relationship between alcohol use disorders and death by suicide may not differ by gender (Lange et al., 2024), pathways focused on in the current study (i.e., TB, PB) may (i.e., Gallyer et al., 2018). Finally, as previously noted, our sample sizes were relatively small and may have been underpowered to detect group-level associations (Hoekstra et al., 2023).

Despite these limitations, the current findings highlight the importance of further investigations of the relationship between risk factors for momentary alcohol use urges and alcohol use among those with suicidal ideation history. Despite the contradictory nature of group-level analyses, and individual-level analyses highlighting significant heterogeneity within these associations, this study highlights the importance of further investigating the role of negative affect, thwarted belongingness, and perceived burdensomeness in the proximal risk pathways between alcohol use and suicide risk. Although the number of individuals in the analysis was small, findings also raise questions as to whether there is a “one-size fits all” theoretical model that exemplifies the relationship between alcohol use and suicide risk, potentially highlighting the need for distinctions in associations that occur across a longer timescale (i.e., chronic alcohol use; distal risk pathways) as well as intensive associations (i.e., momentary alcohol use urges/behaviors; proximal risk pathways).