We recruited 464 first-year college students for one of two parallel studies. Study 1 (n = 148) consisted of a neuroimaging session as well as online self-report questionnaires. Study 2 (n = 316) was designed, largely in response to the COVID-19 pandemic, as an online alternative to the neuroimaging study in which participants completed the same set of self-report questionnaires but did not attend an in-person imaging session. Participants gave written (Study 1) or electronic (Study 2) informed consent in accordance with ethical standards of the UNC Office of Human Research Ethics.

Inclusion criteria for both studies was 18–19 years of age and in their first year in a 4-year undergraduate degree program. Given the strong effects of the college environment on alcohol consumption, and to reduce variability associated with this factor, college attendance at time of study enrollment was an inclusion criterion. The neuroimaging study, Study 1, had additional exclusion criteria: MRI contraindications (i.e., claustrophobia, non-removable metal in the body, pregnancy), left-handedness, psychoactive drug use (including medications), neurological disorders, and psychiatric disorders with the exception of lifetime—but not current—mood or anxiety disorders. Psychiatric disorders were assessed via Mini-International Neuropsychiatric Interview (M.I.N.I.) for DSM-IV (23). We excluded for current or lifetime AUD or substance use disorder using DSM-5 criteria. Study 1 subjects also screened negative for alcohol and other drug use (i.e., cocaine, cannabis, opioids, amphetamines, methamphetamine) on the day of the neuroimaging session via breathalyzer test (FC-10, Lifeloc Inc., Wheat Ridge, CO) and urine drug screen (Biotechnostix, Inc., Markham, ON), respectively.

Subjects completed a computerized battery of self-report questionnaires administered with Research Electronic DATA Capture (REDCap) tools hosted at UNC (24), which included instruments measuring personal and family alcohol use, attention, impulsiveness, resilience, and symptoms of depression.

We administered three items from the Alcohol Use Questionnaire (#10-#12) that were combined to create a “binge drinking score,” as described elsewhere (25).

Timeline followback calendars (26) were administered to collect self-report data on the number of alcoholic drinks consumed on each of the 30 previous days. We calculated the number of binge drinking episodes in 30 days, where a binge drinking was defined as any day in which 5 or more drinks (males) or 4 or more drinks (females) were reported.

AUD among family members was assessed with the Family History Assessment Module (FHAM) (27). We created a binary FH variable based on having a first-degree biological relative (i.e., parent or sibling) or two or more biological second-degree relatives (i.e., grandparents, aunts, and uncles) with AUD. This definition was based on the utility of considering both first-degree and second-degree relatives in AUD research (28, 29).

The CD-RISC is a measure of resilience that contains 10 questions relating to ability to cope with stressors and has been validated for use in college students with histories of stress and trauma (30). Scores range from 10 to 40, with higher scores corresponding with greater stress resilience and mean scores of a community sample of adults being ~32 (31).

Although the BIS was designed as a measure of impulsivity, the BIS factor structure contains a first-order factor that is comprised of items measuring attentional ability (32). Questions contained in the first-order “attention” factor relate to the ability to focus and pay attention, for example, “I don't pay attention,” “I concentrate easily,” and “I am restless at the theater or lectures.” Higher scores correspond with more impaired attentional ability. As our a priori hypothesis centered around attention as a mediator of resilience and binge drinking, we used this subscale in our primary analyses. Secondary analyses explored relationships using other first-order and second-order subscales.

Depressive symptoms were measured with the Beck Depression Inventory (33), and a total score was computed to represent current levels of depression.

For descriptive purposes, we calculated the means and standard errors of the means for the above self-report measures, separately by FH status and study. A chi-square test in SAS 9.4 PROC FREQ tested for differences in the distribution of the sexes by FH status and study. Two-way analyses of variance (ANOVAs) in SAS (Cary, NC) PROC ANOVA were performed on BDI scores, CD-RISC scores, and BIS scores to test for main effects of FH status and study. Although we administered the Customary Drinking and Drug Use Record [CDDR, Brown et al. (34)] to assess substance use in addition to alcohol, cannabis use and tobacco use were almost exclusively reported among users of alcohol, with the exception of only two subjects who reported using both cannabis and tobacco but not alcohol. Therefore, statistical analyses could not reliably separate effects of these other substances from alcohol effects, and thus these variables were not included in analyses.

To examine evidence for attention as a resilience factor among FHP subjects, we tested for statistical differences in attention between FHP binge drinkers and non-binge drinkers as well as between FHN binge drinkers and non-binge drinkers. Because there is not a specific binge drinking score that clearly delineates binge drinkers from non-binge drinkers due to the nature of the calculation (35), we designated non-binge drinkers as those with a score of 0, representing the first quartile of our sample, who also reported no binge episodes in the previous 30 days based on timeline followback data. Twenty-two subjects with missing timeline followback data were excluded from the non-binge-drinking group due to an inability to confirm their status as non-binge drinkers. We designated binge drinkers as those in the upper quartile of binge scores, which included scores ≥20, regardless of the number of binge episodes in the previous 30 days. Binge drinkers reported an average of 3.44 binge episodes in the previous 30 days (range 0–19).

We conducted a two-way analysis of covariance (ANCOVA, SAS PROC ANOVA) testing factors of FH, binge drinking, and their interaction, covarying for sex and study. Exploratory analyses tested other BIS subscales as dependent variables in similar ANCOVA models without correcting for the multiple tests.

We assessed correlational relationships between CD-RISC and BDI scores, binge drinking scores, and BIS scores. Spearman rank-order partial correlations were conducted in SAS 9.4 separately by FH status, covarying for study. Statistical significance was determined by calculating 95% confidence intervals from 10,000 bootstrap iterations. We included biological sex as a covariate due to sex differences in binge drinking (Center for Behavioral Health Statistics and Quality 2015) as well as in stress-related psychopathology within attention systems (36). Due to known associations of depressive symptoms with both binge drinking (25) and resilience (13), correlations also covaried for BDI scores to ensure relationships were not driven by depressive symptoms (with the exception of the correlation between CD-RISC and BDI scores).

Next, we sought to examine causal relationships between attention, resilience, and binge alcohol drinking. Specifically, we used mediation analyses to test whether BIS attention scores mediated the relationship between CD-RISC scores and binge drinking scores, and whether the indirect effect was moderated by FH.

Missing questionnaire data was imputed with the mean value from other subjects from the same study, FH status, and sex. For all variables, missing data represented <4% of values. Mediation analyses were performed using the Multilevel Mediation and Moderation (M3) MATLAB Toolbox (37) in MATLAB 2019b. Due to differences across our study samples (Table 1), we utilized multilevel analyses to merge the samples from Study 1 and Study 2, preserving power to detect variable relationships, while simultaneously allowing separate estimates of means and slopes for the two studies. Significant paths were evaluated by performing 100,000 bootstrap iterations and calculating 95% confidence intervals. Mediation analyses were conducted separately for FHP and FHN groups.

Prior to each mediation analysis, study variables were converted to rank-ordered values due to non-normal distributions. Next, we used linear regression models to estimate effects of sex and depression on the mediator (BIS attention) and the outcome variable (binge drinking score). These effects were subsequently subtracted from the mediator and outcome variables to adjust our mediation model for the effects of these nuisance variables. This adjustment was conducted separately for each study and FH group.

After testing mediation models for the FHP and FHN groups, we additionally tested the null hypothesis of equal indirect paths for the two models, thereby testing whether the mediation effect was moderated by FH. To test this, we took the difference in the indirect path estimates between the FH groups and the variance estimates for those paths to compute a z-statistic of the difference between models for FHP and FHN (38). A significant effect would indicate that FH moderated the mediating effect of attention on the relationship between resilience and binge drinking.

Secondary exploratory analyses tested the remaining BIS factors as mediators using identical methods as above. The goal of these analyses was to explore whether other aspects of impulsivity, outside of our a priori hypothesis regarding attention, mediated resilience effects on binge drinking.

Demographic characteristics of subjects separated by FH status and study are presented in Table 1. FHP subjects demonstrated greater BDI scores and higher binge drinking scores. Study 1 and Study 2 demonstrated differences in CD-RISC scores, BDI scores, binge drinking scores, and BIS total and subscale scores except for those related to motor impulsivity.

Scores on the BIS first-order attention subscale by FH and binge drinking group are displayed in Figure 1. Results of a two-way ANCOVA controlling for sex and study indicated a significant interaction of FH and binge drinking [F_(1,203) = 7.17, p = 0.008]. Specifically, non-binge-drinking FHP subjects reported significantly lower attention problems than not only the FHP binge-drinkers, who reported the highest attention problems, but also FHN non-binge drinking subjects, consistent with a marker of resilience to binge drinking.

Exploratory analyses testing other BIS first-order factors in similar ANCOVAs found no other significant interaction effects. Among second-order factors, there was a significant interaction of FH and binge drinking on the attentional subscale [F_(1,203) = 5.85, p = 0.017], mirroring the primary finding (Supplementary Figure 1).

Correlations between CD-RISC, binge drinking, and BIS scores are presented in Table 2. There was a significant correlation between resilience and attention, similar to previous findings (12, 14), in both FHP and FHN subjects. Additionally, we identified significant correlations between BIS attention scores and binge drinking scores, but only among FHP subjects. Similarly, significant correlations between CD-RISC scores and BIS total and subscale scores were more robust among FHP. On the other hand, the correlation between resilience and depression, replicating previous work (13, 14), was strongest in FHN subjects.

Models of the mediating effects of BIS first-order attention scores on the relationship between CD-RISC and binge drinking are presented in Figure 2. The direct path between CD-RISC scores and binge drinking score was significant for FHP, but not FHN, subjects. There were significant a and b paths in both groups, indicating an indirect effect of resilience on attention and an effect of attention on binge drinking. However, the total indirect effect (a×b) of BIS Attention scores was stronger for FHP than for FHN (z = −3.05, p = 0.002).

Mediation effects for all BIS subscales are presented in Supplementary Table 1. Of note, there was a sizable and significant mediation by the non-planning second-order subscale of the BIS, suggesting reductions in this form of impulsiveness may also promote stress coping resilience to reduce binge drinking.

Several limitations of the study should be considered. This study was conducted only in first-year undergraduate students attending 4-year colleges and thus the results may not necessarily translate to adolescents, older populations, or 18–19-year-olds in other academic or non-academic environments. Because analyses did not control for other substance use, it is possible that reported relationships are not specific to alcohol use. Additionally, this study relied on self-report measures of substance use, attention, and resilience. The cross-sectional nature of this study is another limitation, and future studies are needed to determine whether the association of attentional ability with resilience and binge drinking represents a causal mechanism. It is also possible that many of the non-binge drinking subjects will develop AUD later in life. The utility of attentional ability as a predictor of future AUD remains to be determined. Additionally, a large proportion of our sample was female, and the study was not powered for testing sex differences, so it is possible that effects differ between males and females. Increased severity of clinical symptoms in the online sample (Study 2) was likely driven by the relaxed exclusion criteria, which was intended to provide improved real-world translation. However, differences in recruitment methods, geographical regions, and the pandemic may also have contributed to sample differences. Future studies are needed to identify the neural circuits that mediate the relationships between attentional ability, resilience, and binge drinking.

This study revealed a novel link between attention and resilience to binge alcohol use among individuals at risk for AUD. Reducing hazardous alcohol use on college campuses remains an important goal of intervention research (52). The current findings suggest approaches targeting attention in college students with familial risk for AUD have the potential to reduce the immediate and future burden of binge drinking in this high-risk group.