Alcohol Craving and Consumption in Borderline Personality Disorder: When, Where, and with Whom

Abstract

Substance use is highly prevalent in our society, and substance use disorders are comorbid with most psychiatric disorders, including borderline personality disorder (BPD; Grant et al., 2006, 2008). Craving is a fundamental feature of addiction and disorder, yet the contexts in which craving occurs and is associated with substance use is still under-researched. We examined alcohol craving and consumption in a sample of 56 BPD individuals and a comparison group of community drinkers (COM; n = 60) who carried electronic diaries for approximately 21 days. BPD individuals reported more craving than COM individuals in most contexts. Compared to COM individuals, elevated craving in BPD individuals was paralleled by more drinking when at work, at home, and with romantic partners, coworkers, and children. These findings identify contexts of particular relevance to those with BPD and other mood/anxiety disorders in which craving may lead to risky and maladaptive alcohol use.

Substance use disorders (SUDs) are among the most prevalent of all psychological problems within the general population (Grant et al., 2006; Kessler et al., 2005; Regier et al., 1990; Substance Abuse and Mental Health Services Administration [SAMHSA], 2013). SUDs are routinely associated with physiological, psychological, and social impairment (Schuckit, 2009; American Psychiatric Association [APA], 2013), including psychopathology more broadly (e.g. Grant et al., 2006). Individuals with borderline personality disorder (BPD) represent a particularly high-risk group for substance use given their propensity for emotion dysregulation and impulsive behaviors (Carpenter & Trull, 2013; Crowell, Beauchaine, & Linehan, 2009; Linehan, 1993), and BPD has been found to be highly associated with SUDs (Carpenter, Wood, & Trull, 2015; Skodol, Oldham, & Gallaher, 1999; Trull et al., in press).

Even taking into account shared variance across SUDs, alcohol use disorder (AUD) is associated with BPD (Carpenter, Wood, & Trull, 2015), and nearly half of BPD individuals meet criteria for an AUD diagnosis (e.g., Tomko et al., 2014; Trull et al., in press; Trull, Sher, Minks-Brown, Durbin, & Burr, 2000). The reasons for this high level of comorbidity are not fully understood, but multiple factors exist that may, at least in part, explain why those with BPD appear particularly susceptible to AUD. We briefly review three major overlapping models of etiology for SUDs (including AUD), namely pharmacological vulnerability, affect regulation, and deviance proneness (Sher, 1991; Sher & Trull, 1994; Sher at al., 1999), that highlight the interplay between several factors that appear to be associated with and give rise to both BPD and AUD. Importantly, these factors are not unique to BPD, but are also likely relevant to the comorbidity observed between AUD and other psychiatric disorders (e.g., depression, anxiety disorders, antisocial personality disorder; Grant et al., 2006).

The pharmacological vulnerability model evolved from the observation that there are individual differences in the effects of alcohol, especially the relationship between personality and alcohol sensitivity (Sher et al., 1999). Individual differences in reactions to alcohol have been shown to prospectively predict alcohol problems (Schuckit & Smith, 1996; Volavka et al., 1996), and it has consistently been found that the personality trait of impulsivity/disinhibition is associated with the stress-reducing properties of alcohol (e.g., Levenson, Oyama, & Meek, 1987; Sher & Levenson, 1982). Furthermore, impulsivity, a central feature of BPD (APA, 2013), may influence the decision to use alcohol and lead to its continued use (or overuse) in those situations in which many would discontinue alcohol use.

The negative affect regulation model posits that alcohol use and abuse may represent attempts to regulate or alleviate negative affect and, as a result, alcohol use may become negatively reinforcing (Baker et al., 2004). Both the psychopharmacological properties of alcohol and the belief that alcohol will alleviate negative affective states may contribute to this effect (e.g. Kober, 2014). Given their high levels of negative affect and proneness to negative affect instability, individuals with BPD may be especially vulnerable to engaging in alcohol use for this reason (Trull et al., 2008).

Finally, the deviance proneness model suggests that temperamental traits, especially those related to impulsivity/disinhibition, may interact in a transactional way with deficits in parental control, leading to problems in socialization (Sher et al., 1999). In turn, deficits in socialization are associated with a host of problems, including poor academic performance, delinquent behavior, and substance abuse (Sher, 1991). Consistent with this perspective, family and parenting factors are hypothesized to play a key role in the development of BPD (Crowell, Beauchaine, & Linehan, 2009; Linehan, 1993). Additionally, BPD individuals tend to engage in a variety of behaviors that violate social norms, in particular, substance abuse (APA, 2013), which in turn further strain interpersonal relationships (Gunderson, 2007).

Thus, multiple factors, in particular emotion dysregulation and impulsivity, may contribute to the comorbidity of BPD and AUD. Although there are several studies that have explored the use of alcohol in high-risk individuals such as BPD individuals, few have focused on the experience of alcohol craving or the relationship between alcohol craving and alcohol use in this population. The experience of craving, or the desire, frequently intense, to use a substance (often while attempting to abstain) is a core feature of SUDs (Drummond, 2001; Kozlowski & Wilkinson, 1987; Serre, Fatseas, Swendsen, & Auriacombe, 2015; Tiffany & Wray, 2012). As such, craving may be an important precursor to alcohol use and abuse. For some time, craving has been recognized as a fundamental feature of addiction. Craving is a diagnostic criterion of substance dependence in the International Statistical Classification of Diseases and Related Health Problems - Version 10 (ICD-10; World Health Organization, 1992) and of SUD in the newly revised Diagnostic and Statistical Manual of Mental Disorders, 5^th edition (DSM-5; APA, 2013). Furthermore, recent reviews report a robust link between craving and both concurrent and prospective substance use across a variety of substances (Buckner et al., 2014; Serre et al., 2015).

Craving can be conceptualized as a motivational mechanism that may be triggered or enhanced by stress or negative affect, often leading to substance use (Robinson & Berridge, 2003; Skinner & Aubin, 2010).. Therefore, craving may serve as a potential middle step between these two stages of response (i.e., emotion dysregulation/impulsivity and substance use), and it may play a particularly important role in the development and maintenance of substance use in BPD individuals. This may be especially true with regard to circumstances where substance use is typically not acceptable or normative (e.g., at work, with children) and where those with BPD are especially susceptible to emotion dysregulation or impulsivity (e.g. during interpersonal contact; Gunderson et al., 2011; Russell, Moskowitz, Zuroff, Sookman, & Paris, 2007).

Although there is substantial literature studying alcohol craving in experimental settings (Plebani at al., 2012), including studies using neuroscience methods (e.g., Goldstein & Volkow, 2011), relatively few studies have examined alcohol craving as it manifests in daily life (Serre et al., 2015). Instead, much of the research on substance craving has been conducted in controlled laboratory settings using experimental paradigms, most notably the cue-reactivity procedure (Carter & Tiffany, 1999; Drummond, Tiffany, Glautier, & Remington, 1995). This work has been instrumental for understanding the mechanisms underlying craving and its associations with substance use and relapse. However, it has also largely assessed craving at the person level, with limited assessment of potential variation across a range of naturally occurring, uncontrolled exposures. Such approaches limit the ability to parse between-subject and within-subject sources of variation that are likely to interact in everyday life.

This is not to say that previous research suggests that craving is context independent. Rather, many hypothesize and find variability in individuals’ responses to different cues (e.g., Carter & Tiffany, 1999). Furthermore, some have attempted to acquire snapshots of individuals’ context-dependent craving experiences in daily life using targeted retrospective reports (e.g. Annis, 1982; Lipperman-Kreda et al., in press). However, there are methodological pitfalls to each of these approaches, ranging from recall bias to limited ecological validity (c.f. Grimes & Schulz, 2002; Serre et al., 2015). Alternatively, time-intensive sampling methodologies, including ecological momentary assessment (EMA; Stone & Shiffman, 1994), minimize many of these limitations by assessing individuals using digital technologies as events happen in daily life.

Given that craving is thought to be elicited from cues in one’s environment (Drummond, 2001; Tiffany & Wray, 2012), it seems natural to measure the level of alcohol craving, as well as resulting alcohol use, elicited by different contexts commonly encountered in everyday life. Therefore, using EMA, the current study sought to identify the specific times when, places where, and people with whom individuals with BPD craved, and, subsequently, drank alcohol. Such work has the potential to serve as both a platform with which to corroborate various laboratory findings and a starting point for developing personalized and adaptive interventions that are administered in real-time as individuals report being in previously identified risky contexts (Morgenstern, Kuerbis, & Muench, 2015; Trull & Ebner-Priemer, 2013).

EMA approaches in clinical psychology, including addictions research, have been expanding in recent years, and the phenomenology of craving in this context has emerged as a particular construct of interest. Much of this research has assessed craving as a correlate of use, and the link between craving and substance use is well documented (Buckner et al., 2014; Serre et al., 2015; Wray, Merrill, & Monti, 2015). These associations range from weak to strong, but, importantly, contextual factors are almost always implicated in moderating the association between craving and use. Contextual factors are posited to be associated with a range of variables, including level of craving, level of substance use, and the strength of the association between craving and substance use (e.g., including both conducive contexts where drinking is socially appropriate and inhibiting contexts that provide limited opportunity for drinking). Contextual factors can also have simultaneously mixed effects, such that, for example, craving is enhanced due to a factor in the environment (e.g., stress) but drinking is inhibited because of a co-occurring factor (e.g. being at work or with children; Serre et al., 2015).

To date, much of the literature has explored the role of “internal” contextual factors on craving and substance use (e.g. mood, emotion; Serre et al. 2015) as opposed external environmental factors. The few studies that have explored environmental contextual factors have found that craving occurs more often in the evening (Piasecki et al., 2011; Ramirez & Miranda Jr., 2014), on weekends (Piasecki et al., 2011), and at bars/restaurants (Piasecki et al., 2011). In turn, craving occurs less frequently at work/school (Ramirez & Miranda Jr., 2014), and when around others (Piasecki et al., 2011; Serre et al., 2014), especially children (Linas et al., 2014). Some preliminary evidence suggests that actual substance use tends to occur more often at bars/restaurants and at home (Muraven et al., 2005; Simons et al., 2005), and with romantic partners and friends (Muraven et al., 2005; Linas et al., 2014; Simons et al., 2005).

Furthermore, to our knowledge, no studies have examined how person-level factors might moderate the effect of different contexts on craving and substance use, though some have assessed context effects in independent at-risk samples (e.g. Fatseas et al., 2015; Linas et al., 2014; Ramirex & Miranda Jr., 2014) and in those receiving treatment (see Serre et al., 2015). Recently, Freisthler, Lipperman-Kreda, Bersamin, and Gruenewald (2015) proposed a social-ecological model of substance use, specifically alcohol, that outlines links between person-level, psychosocial characteristics and situation-specific social, spatial, and temporal characteristics in influencing alcohol consumption and alcohol related problems. Similarly, we argue that within-person contextual factors may facilitate/inhibit craving and alcohol use depending, in part, on other between-person personality factors (i.e., the presence of BPD).

Current Study

We used EMA to examine when, where, and with whom two groups of regular drinkers were more likely to crave and consume alcohol. The first group consisted of BPD individuals currently in treatment. As mentioned above, BPD individuals are an especially high-risk group for AUD and, potentially, alcohol craving, due to the high levels of emotional dysregulation, impulsivity, and social deviance associated with BPD, as well as the high rates of comorbidity for BPD with SUDs in general, and AUD in particular (Carpenter & Trull, 2013; Crowell et al., 2009; Linehan, 1993; Trull et al., in press). The second group consisted of individuals recruited from the local community (COM). The COM group represented: (1) a potential baseline for characterizing normative times when, places where, and people with whom individuals reported craving and drinking; and (2) a comparison point for identifying contexts that may be especially risky for an already high-risk group (i.e. BPD individuals). Our study thus provides an opportunity to identify contexts that are particularly risky (or protective) for those whom the alcohol craving-consumption link might be expected to be especially strong, namely those with BPD.

We examined alcohol craving and consumption in terms of both binary and ordinal/continuous endorsement of craving and alcohol consumption at each report given recent work suggesting that craving intensity confers additional information beyond discrete endorsements (Ramirez & Miranda Jr., 2014; Wilson & Sayette, 2014; 2015). In addition, to address the theoretical interpretations presented above regarding the experience of specific symptoms related to BPD that may put such individuals at an increased risk for craving and subsequent alcohol use, we compared EMA self-reports of negative affect, impulsivity, and interpersonal conflict between groups. We predicted that the BPD group would report higher average levels of all three variables across the EMA period.

Based on findings from the pharmacological vulnerability model and its associations with impulsivity, we hypothesized that BPD individuals would have the sustained impulse to use alcohol (i.e. craving) across most contexts. In addition, while previous studies have found that craving occurs less often at work and when around others, specifically children, we hypothesized that BPD individuals, would conversely have a heightened tendency to crave in these socially non-normative or inappropriate contexts as a result of a propensity toward social deviance. Similarly, in contrast with past findings with non-BPD individuals, we also expected BPD individuals to crave more earlier in the day, on weekdays, and at home compared to controls. Finally, drawing from negative affect regulation models of substance use, we expected that craving would be higher in BPD individuals in contexts in which they have a higher likelihood, compared to non-BPD individuals, of experiencing stress and negative affect. Specifically, we predicted this in interpersonal contexts, which are known to be particularly stressful for those with BPD (Berenson, Downey, Rafaeli, Coifman, & Paquin, 2011; Coifman, Berenson, Rafaeli, & Downey, 2012; Gunderson, 2007; Linehan, 1993; Russell et al., 2007), and in contexts where interpersonal contact is more likely (e.g. at work, school, bars).

Furthermore, given the robust link between craving and subsequent drinking, we hypothesized that in situations where BPD individuals have heightened levels of craving (during the week, at home and at work, and when around romantic partners and children) compared to individuals without a BPD diagnosis, they will similarly have an increased likelihood to consume alcohol compared to non-BPD individuals. Importantly, consistent with past research, we hypothesized that comparison individuals without BPD would be most likely to crave and drink in socially normative contexts (in the evening, on weekends, at bars, and with friends). We also predicted that non-BPD individuals may crave and consume alcohol in these contexts to an even greater extent than BPD individuals, as a result of constraining themselves primarily to these more socially appropriate situations.

Method

Participants

The total sample included 116 participants, 56 of whom were in the BPD group and who endorsed the specific affective instability criterion for BPD (APA, 2000). We required this feature in order to ensure those in the BPD group were characterized by emotion dysregulation, a feature associated with substance use disorder (Kober, 2014), and AUD in particular. The remaining 60 in the COM group did not meet the syndromal requirements for BPD nor endorse affective instability. BPD participants were recruited from local psychiatric outpatient clinics and the general community. In clinics, potential participants were made aware of the study through flyers in the waiting rooms and from their assigned therapists or doctors. In addition, advertisements were placed in a local weekly advertisement circular, on bulletin boards of local businesses, and online via Craigslist and a weekly university-wide email announcement. Advertisements targeted individuals experiencing symptoms associated with BPD (i.e., intense mood swings, impulsive behavior, unstable relationships, and intense anger).

Community controls (COM) were recruited through advertisements which included no language regarding BPD symptoms. No restrictions were placed on controls regarding treatment or presence of psychiatric diagnoses, other than those regarding absence of BPD, affective instability, and general exclusion criteria noted below. Potential participants who were recruited from outpatient clinics or through BPD targeted advertisements could not be considered for the control group, even if they proved to be ineligible for the BPD group.

After either completing a form, which was returned to research staff, or contacting research staff directly, participants were briefly screened over the phone for presence of BPD features and absence of exclusionary criteria. If participants appeared eligible, they were next brought in for a face-to-face screening diagnostic interview. To be eligible, participants were required to be between the ages 18 and 45 and report alcohol consumption on average at least once a week (or four times over the previous month). Participants were excluded if they reported current psychosis, intellectual disability, severe neurological dysfunction, or history of head trauma that affected mood or concentration. They were also excluded if they were currently seeking treatment or interested in seeking treatment for alcohol use and/or problems, or if they reported significant unsuccessful efforts to cut down or stop using alcohol or physiological withdrawal symptoms when not using alcohol over the past year. These alcohol-related exclusion criteria were designed specifically to limit the sample to those without severe alcohol use problems, while the alcohol consumption minimum ensured variability in drinking reports but also necessarily oversampled above-average drinkers. If female, participants had to report not being pregnant or not planning to become pregnant. Eligible participants were scheduled for an orientation session. All potential participants who completed the screening interview were paid $20.

The overall sample was predominantly women (78.5%), of Caucasian ethnicity (84.5%), with an average age of 26.4 years (SD = 7.1). A majority of participants were single/never married (68.7%) or currently married (20.0%), did not have any children (74.1%), were employed (78.5%), and had an annual income less than $50,000 (76.7%).¹ All of the BPD participants were currently in treatment and most (73.2%) were taking psychotropic medication (e.g. antipsychotics, mood stabilizers, stimulants, anxiolytics, depressants, hypnotics, anticonvulsants). In contrast, only 2 COM participants were currently receiving treatment for conditions other than BPD and 5 (4.3%) were taking psychotropic medication. Community participants were less likely to qualify for current DSM-IV (APA, 2000) AUD (COM = 15.5%, BPD = 32.1%; χ²(1) = 4.23, p = .040) and lifetime AUD (COM = 51.7%, BPD = 81.1%; χ²(1) = 10.81, p = .001), to have a current mood disorder (COM = 1.7%, BPD = 39.3%; χ²(1) = 25.79, p < .001), or to have a current anxiety disorder (COM = 21.7%, BPD = 64.3%; χ²(1) = 19.81, p < .001).

Given that most BPD participants were receiving medication to address their symptoms, with almost all of the medications discouraging concurrent alcohol use, we considered this BPD group to be relatively low severity and at lower risk for alcohol problems compared to the overall BPD continuum (e.g. Tomko et al., 2014). While, we did not collect specific information regarding BPD individuals’ program of treatment, such as the inclusion of dialectical behavioral therapy (DBT; Linehan, 1993; Robins, Ivanoff, & Linehan, 2001) or Seeking Safety (Najavits, 2002), among others, many of these individuals may have been enrolled in such programs, which might further reduce their relative risk. In contrast, given the high lifetime prevalence of AUD in the COM group compared to the population (Hasin et al., 2007), we viewed this group to be at a relatively greater risk for alcohol problems compared to the general population. In the current study we view this as an advantage in that it results in the two groups being more comparable in terms of their alcohol behaviors and increases the likelihood that BPD features, and not any third variables, would account for differences between the two groups.

Procedure

Diagnostic information was obtained from two semi-structured interviews, Structured Clinical Interview for DSM-IV Axis I Disorders (SCID-I; First, Spitzer, Gibbons, & Williams, 1995) and Structured Interview for DSM-IV Personality (SIDP-IV; Pfohl, Blum, & Zimmerman, 1994), to assess Axis I and II disorders, respectively. Inter-rater reliability was computed for 20 participants in the larger sample by having a second trained interviewer rate videotapes of interviews; reliability was excellent for the diagnosis of BPD (κ = 1.00) and for presence of affective instability (κ= .89).

At the initial orientation session, for which they received $10, eligible participants first completed a battery of self-report questionnaires. Participants were then issued an electronic diary (ED; Palm Tungsten E2© handheld computer) programmed with customized software that they carried for approximately 21 days (M = 21.6; SD = 2.1, Range = 7–25). Two participants dropped out of the study early, but completed at least 7 days of diaries, and so are included in the present analyses. The other 114 participants each completed at least 18 days of diaries. Participants returned to the lab weekly for data downloads and to receive payment. They were paid $50 at each visit for adequate compliance (i.e. at least 80%) of the previous week’s random prompts; however, payment was reduced by $10 for every 10% below 80%. During the last visit, participants also completed a follow-up self-report battery for which they received $10. Thus, the maximum payment per participant was $190 (initial screening, orientation, three weeks of data collection, and follow-up).

Electronic Diary Protocol

The current protocol was a modified version of protocols from previous studies using different samples (see Piasecki et al., 2012; Trull et al., 2008). Participants completed seven different types of reports over the course of the study. First, morning reports were made each day upon waking and had to be completed by 12:00PM. Second, time-based random prompts occurred, on average, 6 times per day, starting after the morning report or 12:00PM, whichever came first. Random prompts were scheduled to be at least 60 minutes apart and random prompts could not occur within 30 minutes of any scheduled assessment. Thus, random prompts could not occur during a drinking episode sequence (see below). Third, user-initiated initial drink reports occurred when participants logged the completion of the first drink of an episode. Fourth, time-based drinking follow-ups were triggered at set intervals (30, 60, 120, and 180 minutes) following an initial drink report. This was extended by 60 minutes with an additional follow-up assessment each time a new drink was reported. Thus, a drinking episode was considered to end after 3 hours had passed with no additional drinks reported, when the participant completed a bedtime report, indicating they were going to sleep for the night, or when the participant failed to complete the last scheduled follow-up. During this time period, no random prompts were given and participants could not initiate new initial drink reports. Fifth, participants who were smokers completed user-initiated cigarette reports when they engaged in cigarette use. Sixth, participants completed user-initiated initial self-harm reports following a self-harm event. Seventh, similar to drinking follow-ups, time-based self-harm follow-ups were triggered at set intervals (30, 60, and 90 minutes) following an initial self-harm report.

Although these different report types varied in terms of specific items, all report types included questions about alcohol craving, alcohol consumption, and context, which were the focus of the present study. Thus, the current analyses utilized all available craving, consumption, and context data regardless of prompt type. The compliance in this sample was high, with an average completion rate of 90.26% of random prompts, 92.95% of follow-up prompts, and 93.28% of morning reports. On average participants completed 137.0 reports per person. A majority of the prompts were random (60.2%), followed by self-initiated or follow-ups (25.7%) and morning (14.1%). A majority of all self-initiated and follow-up reports were alcohol reports (78.7%), followed by cigarette reports (19.7%). Though assessed, the smoking and self-harm responses were not included in the present analyses. The total number of prompts included in the analyses was 15,889 over 2,508 person-days.

Measures

Alcohol Craving

At the beginning of each assessment individuals were asked to indicate the degree to which they were craving a drink in the “PAST 15 MINUTES” on a scale of ‘1’ (not at all) to ‘5’ (extremely). Of all reports, 71.1% indicated that individuals were not craving at all, with approximately 10% of reports each falling within the ‘2’ (9.3%), ‘3’ (10.0%), and combined ‘4’ and ‘5’ (9.6%) endorsement categories. Given the relatively low endorsement of craving, an additional, dichotomous craving variable was created corresponding to no craving (responses of ‘1’ were coded as ‘0’) and any craving (responses above ‘1’ were coded as ‘1’).² Both this binary version indicative of any craving and the ordinal version representing the intensity of craving were analyzed as separate outcomes.

Alcohol Consumption

At each random prompt, individuals were asked to indicate if they had consumed alcohol since the last recording, responding “yes” (‘1’) or “no” (‘0’). In addition, they were asked to report on the number of standard drinks they had consumed since the last prompt. Participants were also instructed to initiate a survey after finishing their first drink. A total of 85.3% of prompts indicated no drinking, 11.2% indicated a single drink since the last prompt, and the remaining 3.5% of prompts indicated between 2 and 6 drinks since the last prompt. Similar to craving, given the low endorsement of any drinking, a dichotomous drinking variable was created corresponding to any drinking (‘1’) and no drinking (‘0’). Again, binary and ordinal drinking were analyzed separately.

Time of Day

When participants were either randomly prompted or self-initiated a diary report, the date and time were recorded by the electronic diary (ED). In the current analyses we used the time at which individuals initiated the survey and binned responses according to the hour of the day. Figure 1A depicts the proportion of reports that COM and BPD individuals reported craving and drinking across all reports. Based on Figure 1A we grouped reports into three categories according to time of the day. Reports from 6:00:00AM through 11:59:59AM were grouped into the morning category, reports from 12:00:00PM though 5:59:59PM were grouped into the afternoon category, and reports from 6:00:00PM of the current day though 5:59:59AM of the next day were grouped into the evening category. These thresholds were made based on normative social conventions for partitioning a day, creating approximately balanced categories in terms of the number of waking hours included, and identifying relative inflection points in craving and drinking. Of all reports 25.1% occurred in the morning, 34.4% in the afternoon, and 40.5% in the evening. Sensitivity analyses were undertaken where the thresholds were varied by an hour in either direction and did not change the pattern of reported results.

Figure 1.

Craving probabilities for COM and BPD individuals across (A) time and (B) location and person.

Day of Week

The day of the week that each report occurred on was extracted from the time stamp recorded by the ED. Based on the definition of an evening report (reports from 6:00:00PM of the current day though 5:59:59AM of the next day), reports between 12:00:00AM and 5:59:59AM on a given day were back-cast to the previous day. Figure 1B depicts the proportion of reports that COM and BPD individuals reported craving and drinking across all days. Reports were relatively evenly distributed across days with approximately 14% of reports occurring each on Sunday, Monday, Tuesday, Wednesday, and Thursday, and 15% on both Friday and Saturday.

Location

During each report, individuals were asked to indicate their current location (checking all that applied). In the current investigation we focused on individuals’ reports of whether they were: (1) at school, (2) at work, (3) at a bar or restaurant, (4) at their home or residence, or (5) somewhere else. Individuals reported being at home a majority of the time (58.3%), followed by someplace else (27.3%), at work (12.2%), at a bar or restaurant (5.9%), and at school (5.4%).

People

During each report individuals were asked to indicate “In the PAST 15 MINUTES, WHO have you been with,” (checking all that applied). Possible responses included: (1) a romantic partner, (2) a friend, (3) a coworker, (4) child(ren), (5) a parent, (6) other family, or (7) someone else. In addition, responses were combined to indicate if individuals reported being with anyone (i.e. an affirmative response to at least one of the seven items) in the past 15 minutes (yes = 1, no = 0). Individuals reported being with anyone 38.7% of all reports, with friends (24.7%) and romantic partners (22.8%) being most likely, followed by coworkers (11.4%), children (10.5%), anyone else (7.1%), other family (5.6%), and parents (4.2%).³

Negative Affect

At each momentary assessment, negative affect was assessed using 21 items from the Positive and Negative Affect Schedule-Extended version (PANAS-X; Watson & Clark, 1999). Items were presented to participants on the ED during each of the momentary assessments. For each affect item, respondents were asked to rate the extent to which they felt the particular affective state on a five point Likert scale (1 = very slightly or not at all, to 5 = extremely) in the past 15 minutes. Items were averaged to create an estimate of negative affect for each occasion. All of each individual’s occasion-level reports were then averaged to create an overall person-level estimate of negative affect across the EMA period

Impulsivity

At each momentary assessment, participants were asked to rate their impulsivity in the past 15 minutes.⁴ Participants responded to 4 items, one each based the four subscales of the UPPS Impulsivity Scale (Whiteside & Lynam, 2001), using a 5-point Likert scale (1 = very slightly or not at all, to 5 = extremely). The individual items were, “I felt and acted on a strong impulse (Negative Urgency),” “I did something without really thinking it through (Lack of Premeditation),” “I gave up easily (Lack of Perseverance),” and “I did something for the thrill of it (Sensation Seeking).” Responses were averaged to create an index of impulsivity for each occasion and all occasions were then averaged to create a person-level estimate across the EMA period.

Interpersonal Conflict

At each random prompt, participants answered whether they had had a “disagreement” with anybody since the last prompt (0 = “No”, 1 = “Yes”). Since this item was not asked at every prompt (only random) we aggregated this measure to the daily level to create an index of whether a person reported a disagreement with anybody at any point during a given day. These were then summed to create a measure of the number of EMA days individuals reported a disagreement.

Analytic Strategy

Given that the EMA sampling methodology results in multiple days, in which each participant provided responses, and multiple responses within any given day, we employed regression models with generalized estimating equations (GEE; Liang & Zeger, 1986; see Burton, Gurrin, & Sly, 1998; Carlin, Wolfe, Brown, & Gelman, 2001; Hubbard et al., 2010 for discussions comparing GEE and multilevel approaches) assuming an independent covariance structure with days nested within person (McNeish, 2014) for all analyses. Such models account for the likelihood that certain individuals may crave and/or drink more than other individuals across all reports, and that some days may be higher craving and/or drinking days than others. More importantly, it results in more appropriate calculations of standard errors for rates and degrees of endorsement when comparing between BPD and COM individuals due to the likely non-independence of repeated observations.

We conducted analyses for each context separately, whereby effects for the particular context, group membership, and their interaction, were entered into the model as predictors. This was done because we were particularly interested in the unconditional effects of being in a specific context, acknowledging that a person might also report simultaneously being in other contexts (i.e. being with a friend and romantic partner at a bar). Moreover, as can be observed in the Supplementary Material (Table S1), the correlations between different contexts (where and whom) did differ substantially between BPD and COM individuals in a number of instances, especially within a given domain, which could obscure the interpretation of multivariate results across groups (Rabe-Hesketh & Skrondal, 2012a, 2012b). Time of day and day of week were also analyzed in separate GEE models, but with their different levels included within the same model, respectively (i.e. morning, afternoon, evening; Monday through Sunday), given that the levels are explicitly defined to be mutually exclusive. In all models we included the person mean of the dependent variable across all occasions (i.e. number of craving and drinking days) as a predictor 1) to serve as a functional equivalent to a random person intercept in multilevel models, and 2) to isolate the idiosyncratic effect of context on craving and drinking independent of person-level, trait-like tendencies to crave/drink. For the binary versions of craving and drinking, logistic models were fit with a logit link function, whereas for the ordinal versions, normal models with an identity link function were fit.⁵ We used the GENMOD procedure in SAS 9.4 (SAS, 2014) for all analyses.

Results

First, we sought to establish whether or not the BPD group reported a greater degree of negative affect, impulsivity, and interpersonal conflict compared to the COM group across the entire EMA period. As expected, the BPD group indicated that they experienced greater overall negative affect (M_BPD = 1.48, M_COM = 1.09, t(114) = 6.34, p < .001), impulsivity (M_BPD = 1.31, M_COM = 1.12, t(51) = 2.58, p = .013), and interpersonal conflict (M_BPD = 4.73, M_COM = 2.32, t(114) = 4.48, p < .001) compared to the COM group.

Prior to examining the different contexts in which alcohol craving and drinking were more likely to occur, we first examined group level differences in craving and drinking between BPD and COM individuals across the entire diary period. At any given report (occasion), although all participants were generally unlikely to report craving (i.e., Odds < 1.0), BPD individuals (Odds_BPD = .47) were more likely to report craving (presence/absence) compared to COM individuals (Odds_COM = .35; OR_COM/BPD = .75, p < .001). In addition, BPD (M_BPD = 1.73) and COM (M_COM = 1.54) groups differed in the average intensity of craving at any given report (b_Group = −.18, p < .001). On the other hand, again, while unlikely in general, COM participants were more likely to report drinking (yes/no) at any given occasion (Odds_COM = .20) compared to BPD participants (Odds_BPD = .14; OR_{C OM/BPD} = 1.43, p < .001). This difference was accounted for by the fact that COM individuals reported drinking on 41.5% of days (9.0 days) compared to 30.7% of days (6.6 days) for BPD individuals (OR_COM/BPD = 1.60, p = .005). Importantly, although COM individuals drank more often (i.e., had more drinking days), on drinking days they drank a similar amount (M = 3.47 drinks) compared to BPD individuals (M = 3.51 drinks; b_Group = −.04, p = .931).

When do craving and drinking occur?

Figure S1A displays the raw proportions of observations in which individuals reported binary craving and drinking grouped by hour of the day, across all days (see Supplementary Material, Figure S2A for ordinal versions). Craving tended to increase throughout the day; however, BPD individuals appeared to crave more in the morning and afternoon while group differences were attenuated in the evening. Drinking, in contrast, was negligible in the morning for both groups, but COM individuals appeared to drink slightly more in the afternoon and evening compared to BPD individuals.

These patterns are corroborated by the analyses in Tables 1 and 2 and Figures 1 and 2 such that, for binary endorsement, BPD participants reported a higher likelihood of craving in the morning (OR_BPD/COM = 2.31, 95% CI = [1.68, 3.17], p < .001) and afternoon (OR_BPD/COM = 1.43, 95% CI = [1.17, 1.74], p < .001), but not in the evening (OR_BPD/COM = 1.00, 95% CI = [.84, 1.20], p = .989). The ordinal scoring of craving in comparison suggested significantly higher craving among BPD participants in the afternoon (M_BPD-COM = .16, SE = .04, p < .001) and evening (M_BPD-COM = .18, SE = .05, p < .001), but not in the morning (M_BPD-COM = .04, SE = .03, p = .119). Contrasts testing between-group differences by time of day indicated that the difference in binary craving was greater in the morning than the afternoon (OR = 1.61, 95% CI = [1.16, 2.25], p = .005), but the observed group differences in craving intensity did not differ in the afternoon compared to the evening (p > .401).

Table 1.

Raw prevalence, binary odds, and continuous parameter estimates for craving endorsement by context for BPD and COM individuals.

		COM					BPD
		Binary			Ordinal		Binary			Ordinal
Context		Prev.	Odds	95% CI	Est.	SE	Prev.	Odds	95% CI	Est.	SE
When
Morning		4.8%	.03a	[.03, .04]	1.16	.02	12.8%	.08a	[.06, .09]	1.20	.02
Afternoon		21.2%	.19a	[.16, .22]	1.46b	.03	29.7%	.27a	[.24, .32]	1.62b	.03
Evening		43.5%	.67	[.59, .75]	1.90b	.03	45.6%	.67	[.58, .77]	2.08b	.04
When
Monday		19.6%	.20	[.15, .25]	1.46	.05	29.4%	.28	[.21, .37]	1.60	.06
Tuesday		23.0%	.24	[.19, .29]	1.53	.05	29.1%	.27	[.21, .35]	1.59	.06
Wednesday		21.5%	.21a	[.17, .26]	1.47b	.04	31.4%	.31a	[.24, .40]	1.68b	.06
Thursday		28.6%	.32	[.26, .41]	1.57	.05	31.2%	.32	[.25, .42]	1.72	.07
Friday		32.7%	.40	[.32, .49]	1.69	.05	34.2%	.39	[.31, .50]	1.74	.06
Saturday		31.7%	.39	[.31, .49]	1.72	.06	36.2%	.44	[.36, .55]	1.86	.07
Sunday		23.9%	.24	[.19, .31]	1.48b	.05	31.5%	.32	[.25, .42]	1.68b	.07
Where
School		20.5%	.21	[.15, .29]	1.37b	.05	26.1%	.29	[.21, .39]	1.53b	.06
Work		13.4%	.11a	[.09, .14]	1.33	.03	20.3%	.19a	[.15, .24]	1.44	.05
Bar/Restaurant		57.6%	1.28a	[.96, 1.70]	2.28	.08	46.8%	.79a	[.58, 1.09]	2.12	.09
Residence		23.0%	.24a	[.21, .27]	1.48b	.02	31.0%	.30a	[.27, .34]	1.68b	.03
Other		36.1%	.46	[.40, .53]	1.76	.04	39.1%	.47	[.41, .54]	1.86	.04
With Whom
Nobody		22.3%	.20a	[.17, .23]	1.50	.03	27.8%	.26a	[.23, .29]	1.55	.03
Anybody		28.1%	.33a	[.30, .37]	1.60b	.02	35.3%	.40a	[.35, .45]	1.82b	.06
Partner		27.7%	.34a	[.29, .40]	1.60b	.03	41.1%	.50a	[.40, .62]	1.98b	.06
Friend		42.8%	.63	[.54, .74]	1.89b	.05	41.3%	.56	[.47, .66]	2.04b	.05
Co-worker		18.5%	.15a	[.12, .19]	1.37b	.03	26.9%	.28a	[.22, .37]	1.62b	.07
Child		19.9%	.21	[.17, .27]	1.38b	.03	38.3%	.28	[.22, .36]	1.75b	.08
Parent		27.2%	.35	[.25, .50]	1.55	.06	34.5%	.40	[.26, .61]	1.74	.11
Other Family		31.6%	.35	[.26, .47]	1.60b	.06	42.1%	.45	[.32, .64]	1.84b	.09
Other		29.6%	.38	[.29, .51]	1.63	.06	35.9%	.37	[.29, .46]	1.72	.06

Note. BPD = Borderline Personality Disorder; COM = Community; Prev. = Prevalence; CI = Confidence Interval; Est. = Estimate; SE = Standard Error.

^aIndicates that Odds in the same row are significantly different at p < .05.

^bIndicates that Estimates in the same row are significantly different at p < .05.

Table 2.

Raw prevalence, binary odds, and continuous parameter estimates for drinking endorsement by context for BPD and COM individuals.

		COM					BPD
		Binary			Ordinal		Binary			Ordinal
Context		Prev.	Odds	95% CI	Est.	SE	Prev.	Odds	95% CI	Est.	SE
When
Morning		.5%	.00	[.00, .01]	−.02b	.00	.4%	.00	[.00, .01]	.05b	.01
Afternoon		8.8%	.07	[.06, .08]	.08	.01	5.1%	.06	[.04, .07]	.09	.01
Evening		33.4%	.38	[.34, .43]	.44	.02	26.2%	.37	[.33, .42]	.39	.02
When
Monday		10.2%	.09	[.06, .12]	.10	.02	9.3%	.11	[.08, .16]	.16	.02
Tuesday		14.6%	.13	[.10, .17]	.16	.02	8.2%	.09	[.06, .13]	.14	.02
Wednesday		12.5%	.10a	[.08, .13]	.11b	.02	13.2%	.16a	[.12, .21]	.22b	.03
Thursday		18.9%	.17	[.14, .22]	.23	.03	12.3%	.15	[.12, .20]	.20	.02
Friday		22.3%	.22	[.17, .27]	.32b	.04	14.1%	.17	[.13, .23]	.22b	.02
Saturday		23.1%	.23	[.18, .29]	.29	.03	17.0%	.23	[.17, .29]	.27	.03
Sunday		13.9%	.12	[.09, .16]	.15	.03	11.7%	.14	[.11, .19]	.20	.02
Where
School		2.9%	.03	[.02, .06]	.08	.03	2.7%	.04	[.02, .08]	.10	.02
Work		1.3%	.01a	[.01, .01]	−.02b	.01	2.2%	.02a	[.01, .04]	.05b	.01
Bar/Restaurant		57.1%	.95	[.74, 1.22]	.78b	.07	40.3%	.68	[.50, .93]	.55b	.05
Residence		14.2%	.13	[.11, .14]	.15	.01	11.2%	.13	[.11, .15]	.18	.01
Other		25.2%	.25	[.21, .29]	.33	.03	17.5%	.21	[.18, .25]	.27	.02
With Whom
Nobody		8.9%	.08	[.06, .09]	.10	.01	5.8%	.07	[.05, .08]	.11	.01
Anybody		20.7%	.20	[.17, .22]	.25	.02	17.6%	.23	[.20, .26]	.28	.02
Partner		21.8%	.20a	[.18, .24]	.25b	.02	25.7%	.31a	[.25, .38]	.37b	.03
Friend		35.7%	.44	[.38, .51]	.50b	.03	26.1%	.35	[.30, .42]	.39b	.03
Co-worker		7.3%	.05a	[.04, .07]	.08b	.02	11.3%	.12a	[.08, .18]	.15b	.03
Child		13.1%	.10	[.08, .12]	.08b	.01	11.6%	.13	[.10, .17]	.17b	.02
Parent		22.5%	.25a	[.17, .35]	.27b	.05	7.0%	.11a	[.06, .20]	.16b	.02
Other Family		31.2%	.33	[.25, .42]	.35	.04	14.4%	.25	[.15, .41]	.28	.05
Other		19.5%	.18	[.13, .25]	.28	.05	21.4%	.27	[.20, .35]	.32	.04

Note. BPD = Borderline Personality Disorder; COM = Community; Prev. = Prevalence; CI = Confidence Interval; Est. = Estimate; SE = Standard Error.

^aIndicates that Odds in the same row are significantly different at p < .05.

^bIndicates that Estimates in the same row are significantly different at p < .05.

Figure 2.

Drinking probabilities for COM and BPD individuals across (A) time and (B) location and person.

In terms of drinking behavior, BPD and COM participants did not differ in their likelihood of drinking in the morning (OR_BPD/COM = 1.22, 95% CI = [.36, 4.10], p = .744), afternoon (OR_BPD/COM = .80, 95% CI = [.58, 1.10], p = .161), or evening (OR_BPD/COM = .97, 95% CI = [.87, 1.15], p = .689). The ordinal scoring of number of drinks, in contrast, suggested significantly more drinking among BPD participants in the morning (M_BPD-COM = .07, SE = .00, p < .001), but not in the afternoon (M_BPD-COM = .01, SE = .01, p = .352) or evening (M_BPD-COM = −.05, SE = .03, p = .114).

Figure S1B displays the proportion of reports in which COM and BPD individuals indicated that they were drinking, split by day of the week. In general, a weekly trend was observed for both craving and drinking such that craving and drinking gradually increased in both groups from Monday to Saturday and then returned to near Monday levels on Sunday. As indicated by Table 1 and Figure 1, BPD individuals craved significantly more than COM individuals on Wednesdays (OR_BPD/COM = 1.51, 95% CI = [1.09, 2.08], p = .011; M_BPD-COM = .21, SE = .08, p = .005), but also showed trends on Mondays (OR_BPD/COM = 1.44, 95% CI = [.99, 2.09], p = .054; M_BPD-COM = .13, SE = .08, p = .099) and Sundays (OR_BPD/COM = 1.31, 95% CI = [.94, 1.84], p = .109; M_BPD-COM = .20, SE = .08, p = .013). There were no differences in craving on Fridays and Saturdays (all ps > .113).

Table 2 and Figure 2 show the same comparisons for drinking. Similar to craving, BPD individuals drank significantly more than COM individuals on Wednesdays (OR_BPD/COM = 1.52, 95% CI = [1.05, 2.19], p = .022; M_BPD-COM = .10, SE = .03, p = .002), and showed a trend for the continuous measure on Mondays (OR_BPD/COM = 1.27, 95% CI = [.79, 2.03], p = .311; M_BPD-COM = .06, SE = .03, p = .053). In contrast COM individuals drank larger quantities on Fridays (M_BPD-COM = −.10, SE = .05, p = .032), though were not necessarily more likely to have at least one drink (OR_BPD/COM = .80, 95% CI = [.57, 1.13], p = .197).

Where do craving and drinking occur?

Craving was most likely to occur when individuals were at a bar or restaurant (Table 1, Figure 1), with COM individuals craving more than BPD individuals in terms of binary endorsement of craving (OR_BPD/COM = .62, 95% CI = [.41, .95], p = .025), but not in terms of craving intensity (M_BPD-COM = −.17, SE = .12, p = .166). In contrast, individuals experienced the least amount of craving at work, with BPD individuals craving more than COM individuals while there (OR_BPD/COM = 1.74, 95% CI = [1.23, 2.44], p = .001; M_BPD-COM = .11, SE = .06, p = .051). A similar pattern, with BPD individuals craving more than COM individuals, was observed when individuals reported being at home (OR_BPD/COM = 1.26, 95% CI = [1.08, 1.47], p = .002; M_BPD-COM = .20, SE = .04, p < .001) and to a lesser extent at school (OR_BPD/COM = 1.38, 95% CI = [.87, 2.21], p = .164; M_BPD-COM = .16, SE = .08, p = .038).

Similar to craving, drinking was most likely to occur and occurred in the greatest quantities when at bars and restaurants (Table 2, Figure 2), and COM individuals tended to drink more often (OR_BPD/COM = .72, 95% CI = [.48, 1.07], p = .095) and in greater quantities (M_BPD-COM = −.23, SE = .09, p = .007) when there. COM individuals also drank marginally more than BPD individuals when somewhere else (OR_BPD/COM = .85, 95% CI = [.67, 1.08], p = .169; M_BPD-COM = −.06, SE = .03, p = .098). In contrast, BPD individuals were more likely to drink at work (OR-_BPD/COM = 2.65, 95% CI = [1.22, 5.77], p = .012), and drink greater quantities at work (M_BPD-COM = .07, SE = .01, p < .001) and home (M_BPD-COM = .03, SE = .02, p = .055). The two groups did not differ in their drinking behavior when at school, where there tended to be the least amount of drinking in general.

With whom does craving and drinking occur?

Both groups craved more when with others than when alone, but a group difference was also observed (OR_BPD/COM = 1.21, 95% CI = [1.03, 1.41], p = .017; M_BPD-COM = .23, SE = .04, p < .001). In examining particular relationships, BPD individuals reported more frequent and intense craving when around their romantic partners (OR_BPD/COM = 1.47, 95% CI = [1.13, 1.91], p = .003; M_BPD-COM = .38, SE = .07, p < .001), coworkers (OR_BPD/COM = 1.90, 95% CI = [1.35, 2.67], p < .001; M_BPD-COM = .25, SE = .08, p < .001), and children (OR_BPD/COM = 1.32, 95% CI = [.95, 1.85], p = .094; M_BPD-COM = .37, SE = .09, p < .001). They reported more intense craving only, when around friends (OR_BPD/COM = .88, 95% CI = [.70, 1.11], p = .270; M_BPD-COM = .15, SE = .07, p = .028) and other family members (OR_BPD/COM = 1.30, 95% CI = [.83, 2.03], p = .246; M_BPD-COM = .24, SE = .11, p = .027).

Concerning alcohol consumption, all individuals were more likely to drink when with someone compared to when alone, but there were no consistent group differences in drinking frequency or quantity when around anybody (Table 2, Figure 2; OR_BPD/COM = 1.16, 95% CI = [.98, 1.37], p = .080; M_BPD-COM = .03, SE = .02, p = .236), or when by themselves (OR_BPD/COM = .87, 95% CI = [.66, 1.15], p = .319; M_BPD-COM = .01, SE = .02, p = .549). By comparison, there were different patterns when focusing on specific individuals. BPD individuals drank with greater frequency and in larger quantities when around their romantic partners (OR_BPD/COM = 1.51, 95% CI = [1.18, 1.93], p < .001; M_BPD-COM = .13, SE = .04, p = .001), coworkers (OR-_BPD/COM = 2.29, 95% CI = [1.37, 3.84], p = .001; M_BPD-COM = .07, SE = .03, p = .023), and children (OR_BPD/COM = 1.34, 95% CI = [.94, 1.91], p = .095; M_BPD-COM = .09, SE = .02, p < .001), while COM individuals reported the greater drinking than BPD individuals when around friends (OR_BPD/COM = .81, 95% CI = [.65, 1.01], p = .051; M_BPD-COM = −.10, SE = .04, p = .013) and parents (OR_BPD/COM = .46, 95% CI = [.24, .90], p = .020; M_BPD-COM = −.11, SE = .05, p = .039).⁶

Context of craving associated with drinking

Given the established link between craving and concurrent/subsequent substance use (Linas et al., 2014; Serre et al., 2015), and our findings that craving and drinking appear to be contextually influenced, the question of whether craving in a specific context leads to drinking in that context remains. We conducted follow-up analyses, with craving as an additional predictor of concurrent drinking, including its two- and three-way interactions with group and context. As suggested by Figures 1 and 2, the higher degree of craving reported by BPD individuals during weekdays was associated with a relative increase in their risk for concurrent drinking on Wednesdays, such that while the association between craving and drinking was typically lower for BPD compared to COM individuals (see Table S1), this difference was reduced on Wednesdays (OR_BPD/COM = 1.97, 95% CI = [.91, 4.25], p = .080; M_BPD-COM = .10, SE = .04, p = .013; Figure S5). When at work, BPD individuals reported greater craving and drinking than COM individuals, with the association between craving and drinking also being stronger for BPD individuals (OR_BPD/COM = 2.56, 95% CI = [.97, 6.76], p = .054; M_BPD-COM = .11, SE = .02, p < .001; Figure S5). Similarly, when around most people, BPD individuals experienced greater craving compared to COM individuals, but when craving around romantic partners (OR_BPD/COM = 1.41, 95% CI = [.94, 2.11], p = .089; M_BPD-COM = .13, SE = .03, p < .001) specifically, their normatively comparable drinking levels compared to the COM group also significantly eclipsed those of the COM group. Interestingly, this finding was not replicated for coworkers (OR_BPD/COM = 1.03, 95% CI = [.57, 1.87], p = .923; M_BPD-COM = .05, SE = .05, p = .293) or children (OR-_BPD/COM = .88, 95% CI = [.55, 1.42], p = .845; M_BPD-COM = −.05, SE = .06, p = .436). These two results appear to be accounted for by the finding that BPD individuals are especially likely to drink around coworkers (OR_BPD/COM = 5.26, 95% CI = [2.46, 10.96], p < .001; M_BPD-COM = .11, SE = .02, p < .001) and children (OR_BPD/COM = 3.13, 95% CI = [1.48, 6.47], p = .002; M_BPD-COM = .17, SE = .02, p < .001) when they are not craving. Just as there were certain situations where craving was more strongly associated with drinking in the BPD group, there were separate contexts in which craving was more strongly associated with drinking for the COM group. In particular, this link appeared to be stronger for COM participants in the afternoons (OR_BPD/COM = .62, 95% CI = [.42, .92], p = .015; M_BPD-COM = −.09, SE = .03, p = .001), on Tuesdays (OR_BPD/COM = .51, 95% CI = [.31, .84], p = .007; M_BPD-COM = −.22, SE = .08, p = .007) and Fridays (OR-_BPD/COM = .61, 95% CI = [.41, .91], p = .013; M_BPD-COM = −.38, SE = .10, p < .001), and when around their parents (OR_BPD/COM = .30, 95% CI = [.13, .70], p = .005; M_BPD-COM = −.49, SE = .15, p = .001; see Figure S5).

Discussion

The current EMA study provided a unique opportunity to examine alcohol craving and consumption as it occurred in daily life in a sample at risk for AUD (i.e., the BPD group) and among generally healthy (COM group) participants. In this way, we were able to measure the effect on craving and drinking of both within-person contextual factors previously found to be important in substance use research (Linas et al., 2014; Ramirez & Miranda Jr., 2014; Serre et al., 2015), a relevant between-person personality factor (BPD), and their interaction. We observed various distinct contexts in which the risk for craving and drinking was elevated for both groups of individuals, as well as contexts that were particularly risky for the BPD (and COM) group. Our analyses, unsurprisingly, indicate that craving and drinking alcohol was most likely to take place in socially normative contexts (in the evening, on weekends, at a bar/restaurant, with friends) for both groups. This was particularly true for the COM group, who craved and drank primarily in socially sanctioned contexts, where BPD individuals’ craving and drinking were comparatively muted. However, BPD individuals were more likely to crave and consume alcohol, compared to COM individuals, across a variety of other contexts, which may be due to a number of countervailing factors (Serre et al., 2015). This elevated craving profile across most contexts in those with BPD is consistent with Verheul and colleagues’ (Verheul, Van Den Brink, & Geerlings, 1999) obsessive craving psychobiological pathway, characterized by impulse and mood dysregulation (i.e., two core features of BPD, which are elevated for the BPD group in this sample), serotonin deficiency, and vulnerability to obsessive craving and possible relapse. It is also consistent with motivations and behaviors that might be expected from psychopharmacological vulnerability, negative affect regulation, and deviance proneness mechanisms (Levenson, Oyama, & Meek, 1987; Schuckit & Smith, 1996; Sher, 1991; Sher & Levenson, 1982; Trull et al., 2008; Volavka et al., 1996).

In some of these contexts, craving also appeared to be associated with increased drinking. In particular, BPD individuals were more likely to crave on weekdays compared to COM individuals, and this coincided with a relative increase in their drinking behavior on those days. Similarly, social interaction appeared to elicit greater craving for BPD individuals compared to COM individuals, especially when around romantic partners, coworkers, and their own children. Notably, BPD individuals, who typically consumed alcohol on a fewer number of days than COM individuals across most contexts, consumed alcohol in greater amounts than COM individuals when with their romantic partners, coworkers, and children.

Although relatively little research has examined alcohol use and craving across different contexts in daily life, our results are consistent with contextual findings concerning craving and use with other populations and substances. Linas and colleagues (2014) observed that drug users were more likely to report drug craving when with a child or at work, and reported more drug use relative to craving when at home or with a spouse. Similarly, we found that, when at work, BPD individuals were more likely to report craving and consuming alcohol compared to COM individuals. They were also much more likely to crave and drink when around their children. Partially consistent with Linas and colleagues’ results, we found that BPD individuals both crave and drink more when around romantic partners, and, though they do not drink more than COM individuals while at home, they do crave more overall and drink more than average, compared to themselves, when at their residence.

Consistent with previous interpretations (Linas et al., 2014), generally “unstructured” social and physical environments where prescriptions as to how one should behave are more relaxed (e.g. when alone, at a bar, or with friends or a romantic partner compared to when at work, with a child, or in the morning) were associated with more drinking in both groups of individuals, and these environments were also associated with greater craving. The finding that drinking and craving were both associated with a number of different contexts is not surprising, considering the robust positive association between craving and drinking (and substance use more generally; Serre et al., 2015). Previous work has not examined the co-occurrence of craving and substance use, as Linas and colleagues (2014) employed a mutually exclusive scoring system for craving/use (i.e., to the extent craving led to use, an event would only be identified as a substance use event and the antecedent craving would be lost). In our study, all individuals generally reported both craving and drinking more in the evenings and during the weekend, presumably when they were not at work or school and freer to engage in self-directed activities. They also reported more craving and drinking when at bars/restaurants and in other undefined locations. Whereas the relative increase in craving and drinking in other locations may be associated with fewer cues and influences on drinking behavior, being at a bar or restaurant may additionally elicit craving and drinking because they are normative places in which drinking occurs, serving as a prime. Similarly, all individuals tended to crave and drink the most when they were around friends. Friends, compared to other individuals, may serve as contextual cues indicating that drinking is acceptable and even encouraged.

As hypothesized, the current findings are largely consistent with what would be expected from etiological theories of SUDs. For example, the negative affect regulation model suggests that individuals with BPD would crave more under stressful conditions, which would be likely to lead to increased levels of negative affect. This may, in part, explain our findings that BPD individuals craved considerably more at work and school, both typically stressful environments (Baer, 2005; Karasek & Theorell, 1990; Mirsa & McKean, 2000; Sax, 1997). Additionally, it is well-documented that BPD individuals often have histories of problematic relationships (APA, 2013; Gunderson, 2010), suggesting that interpersonal interaction may serve as a particularly poignant stressor for them. Given this, it is perhaps not surprising that we found that BPD individuals craved considerably more when around most individuals (except friends). Furthermore, consistent with our hypotheses derived from both the pharmacological vulnerability and the deviance proneness models, which highlight the relevance of impulsivity/disinhibition for the development of SUDs, BPD individuals engaged in greater craving/drinking in places (e.g., work or school), at times (e.g., morning or early in the week), and with people (e.g., romantic partners, coworkers, children) that are generally non-normative.

However, effects for drinking were not universal across contexts, and, overall, COM individuals unexpectedly reported drinking on more days than BPD individuals. In some respects, this is surprising, given that BPD individuals consistently craved more than COM individuals and that AUD was more prevalent in the BPD group. Similarly, if craving constitutes a prepotent impulsive tendency, for which BPD individuals are disproportionately at risk and as we find in the current sample, then it suggests they should drink more. However, the craving and consumption of alcohol is only one of many outlets for impulsivity, and conversely, not all drinking behavior is necessarily impulsive. Also, while BPD individuals are more prone to experiencing AUD related pathology (e.g., craving), perhaps as a result of associated emotion dysregulation and impulsivity/disinhibition, this does not necessarily translate straightforwardly into drinking more often (e.g. Maclean & French, 2014; Saha, Stinson, & Grant. 2007). Additionally, as stated above, the sampling methods we used may have led to recruiting BPD and community individuals at the lower and higher ends, respectively, of the spectrum of risk for alcohol problems. We also note that all of the BPD individuals were currently in treatment and a large majority were taking medications that explicitly warn against alcohol consumption. Thus, a simpler explanation of the difference in number of drinking days between groups may be that their drinking behavior was otherwise being partially successfully altered through therapy and/or psychopharmacological risks. Besides medicinal contraindications, individuals in the BPD group, while explicitly not seeking/receiving treatment for alcohol abuse, may have acquired behavioral skills from treatment platforms such as DBT (Linehan, 1993; Robins, Ivanoff, & Linehan, 2001) or Seeking Safety (Najavits, 2002) that allowed them to control and not act on such impulses. If this is the case, then the contexts we identified with elevated drinking for BPD individuals compared to COM individuals may be particularly compelling because it suggests that those contexts may be resistant to such skill mobilization.

Beyond exploring the different contexts in which craving and drinking occur in those with BPD, an additional goal of the current study was to determine the convergence/divergence of findings when treating craving (and drinking) as a discrete versus continuous experience. Recent research has argued for the importance of considering craving in terms of its intensity, because it can begin to separate craving that is experienced as a function of wanting or desire from craving that represents more of a psychological or physiological need (Wilson & Sayette, 2014, 2015). Our results indicate that binary and continuous treatment of craving (and drinking) produced very similar patterns of results, though there were some notable departures. For example, while there were no differences in the frequency of endorsing any degree of craving (binary) in the evening between BPD and COM individuals, BPD individuals endorsed higher intensity craving in the evening. This same pattern was observed when individuals reported being around their friends, parents, and other unspecified individuals. When at a bar or restaurant there was evidence that COM individuals craved more often than BPD individuals, though they did not necessarily crave to a greater degree. These effects suggest that craving intensity may be more revealing in some cases, and for BPD and other at-risk groups more generally, by identifying different levels of craving. In support of this, BPD individuals did endorse more intense craving (i.e. either a ‘4’ or ‘5’) more often (37.8% of craving reports) than did COM individuals (27.9%).

The EMA methodology, rich intensive longitudinal data, inclusion of an at-risk group, and a community comparison group are identifiable strengths of the current study; however, there are also limitations to be considered. One core limitation that is shared with much other craving research is that our craving measure was individually defined. That is, what was meant by craving was left up to the participant to decide. Given the lack of agreement in the literature as to what defines craving, namely the wanting versus liking of a substance (Robinson & Berridge, 1993; Sinha, 2013), responses may confound the different reasons for wanting a drink (e.g. physiological/psychological need or intense desire) and liking a drink (e.g. celebratory or affective coping). Our approach of quantifying craving as discrete versus continuous allows for an indirect assessment of this potential issue. Wilson and Sayette (2014; 2015) suggest that most researchers are interested in craving as the intense wanting of a substance. On a continuum this would correspond to the more extreme values, while the liking of a drink may be relegated to less extreme values. In general, our discrete and continuous results were consistent, but there were subtle, and perhaps substantively important differences in some patterns, suggesting a continued need to further disentangle these aspects of craving.

Although the multiple types of reports (i.e. time fixed, random, event-contingent, follow-up) might also be considered a strength, with it comes the limitation that they are disproportionately distributed over the course of the day. For example, a vast majority of prompts completed later in the evening (9:00PM or later) consisted of self-initiated and follow-up reports. Those are largely defined as drinking reports, so responses of craving and drinking during those times could be susceptible to bias to the extent to which there are group differences in the likelihood for BPD and COM individuals to be awake, in specific places (i.e. bars), and with specific people (i.e. friends, partners) during those times. We do note, however, that there were similar percentages of reports for BPD (20.5% of BPD total) and COM (20.7% of COM total) between the hours of 9:00PM and 6:00AM.

Another concern is the difference in the number of drinking days between the BPD and COM groups. The difference in observed AUD prevalences between the groups might lead to the expectation that BPD individuals would drink more than COM individuals, which we did not find. However, it is also plausible that the negative emotionality, negative affective instability, and impulsivity associated with BPD, in combination with drinking, contribute to problems that increase the likelihood of an AUD diagnosis in those with BPD without also leading to an increase in drinking quantity. For example, high levels of negative emotionality and affective instability are associated with occupational, interpersonal, and psychological problems, which, if in the context of drinking, also meet criteria for an AUD (e.g., conflicts with one’s partner, conflicts at work, impulsive decision-making). Alternatively, drinking may be less likely to be problematic for COM individuals. The situations in which drinking typically occurs may more often be socially normative (e.g., at bars and restaurants) as well as linked with the association with motives to enhance positive affect, as opposed to, for example, motives to regulate negative affect or to satisfy impulsive urges (e.g., Piasecki et al., 2014).

A related concern is whether the differences in craving and drinking between the two groups is explained by the higher proportion of AUD observed in the BPD group. To assess this, we included current AUD status as a factor in all of our models and examined if the BPD/COM group differences in craving and drinking persisted. Overall, all reported effects remained when AUD status and its interaction with context was added into each of the craving and drinking models, though the magnitude and significance of some effects was reduced due to the shared variance between the two diagnoses. In particular, the higher levels of craving for BPD individuals when around their children was reduced and the contrast with the COM group was reduced by approximately half (and became marginally significant). The analogous drinking difference was also reduced, but still significant. Overall, we would conclude that the observed effects are specific to BPD and not simply a product of shared symptomatology between BPD and AUD.

Lastly, we note that a majority of our sample was female, and as a result, the generalizability of our findings to both non-clinical and BPD males is unclear. We reanalyzed the data including sex as a factor and did not observe any consistent differences between males and females as a function of BPD status; however, we are likely underpowered to detect such differences and would warn against interpreting this as an absence of gender differences. We do note that, within clinical samples, BPD diagnosis is much more prevalent in females (e.g. APA, 2000), so while we may not be able to generalize our results to males, they likely do generalize to treatment seeking individuals.

Though not considered explicitly in the current study, future work should explore the temporal link between craving and substance use as contextually influenced. Our follow-up analyses provided some evidence for concurrent associations unique to different contexts, but examining lagged and more dynamic relationships are logical next steps. To the extent that such approaches can integrate contextual limiting factors they could be useful in mapping the time course of different craving processes. For example, some places afford immediate satiation of craving (e.g. bars) while others do not (e.g. work, school). If craving shares parallel processes with more general concepts like impulsivity (Joos et al., 2013), then craving that builds over time due to constraints of the environment might be dissociable from craving that is instantaneously experienced and immediately relieved.

To our knowledge this is one of the first studies to collect real-time, independent craving and alcohol use data on both clinically at-risk and non-clinical samples, and to compare the environmental situations in which craving and drinking naturally occur. We found that craving and drinking are contextually influenced, and the contexts appear to be different for non-clinical individuals compared to those who are predisposed toward emotion dysregulation, impulsivity, social deviance, and problematic alcohol use behaviors (i.e., BPD individuals). As noted earlier, emotion dysregulation and impulsivity are factors that are not unique to BPD, but important in many other psychiatric disorders that are also comorbid with SUD. Future work should examine the context of substance craving and use in individuals with other disorders of interest. In addition, defining craving continuously in terms of intensity as opposed to discretely appears to confer additional information as to when craving is especially risky. This work represents a first step towards understanding the conditional links between craving and alcohol use (as well as how alcohol use may spurn additional craving), including how contexts may influence their temporal dynamics. Furthermore, it identifies an opportunity to develop targeted, adaptive interventions that can catch individuals in the moment when they are more likely to engage in potentially problematic behavior (e.g., when entering a bar), and turn them onto a less risky path.