Interactive effects of genetic liability and combat exposure on risk of alcohol use disorder among US service members

Abstract

Background:

An improved understanding of pathways to alcohol use disorder (AUD) among service members may inform efforts to reduce the substantial burden of AUD on this population. This study examined whether the relationship between a service-related risk factor (combat exposure) and later AUD varied based on individual differences in genetic liability to AUD.

Methods:

The sample consisted of 1,203 US Army soldiers of genetically determined European ancestry who provided survey and genomic data in the Army STARRS Pre/Post Deployment Study (PPDS; 2012–2014) and follow-up survey data in wave 1 of the STARRS Longitudinal Study (2016–2018). Logistic regression was used to estimate the conditional effect of combat exposure level (self-reported in PPDS) on odds of probable AUD diagnosis at follow-up, as a function of a soldier’s polygenic risk score (PRS) for AUD.

Results:

The direct effect of combat exposure on AUD risk was non-significant (AOR=1.12, 95% CI=1.00–1.26, p=.051); however, a significant combat exposure x PRS interaction was observed (AOR=1.60, 95% CI=1.03–2.46, p=.033). Higher combat exposure was more strongly associated with elevated AUD risk among soldiers with heightened genetic liability to AUD.

Conclusions:

The effect of combat exposure on AUD risk appeared to vary based on a service member’s level of genetic risk for AUD. Continued investigation is warranted to determine whether PRS can help stratify AUD risk within stress-exposed groups such as combat-deployed soldiers. Such efforts might reveal opportunities to focus prevention efforts on smaller subgroups at the intersection of having both environmental exposures and genetic vulnerability to AUD.

INTRODUCTION

The impact of alcohol misuse on members of the US Armed Forces is widely recognized [1, 2]. Alcohol use disorder (AUD)—defined as drinking despite adverse social, occupational, or health consequences—is of particular concern within active-duty [3] and veteran populations [4] and may be especially prevalent among veterans of recent versus earlier conflicts [5]. This raises serious concerns given the extensive morbidity, mortality, and disability attributed to AUD [6, 7] which includes increased risks of non-fatal suicidal behaviors and suicide death [8, 9].

Given the harms associated with AUD, prevention of this disorder would benefit both individual service members as well as military and veterans organizations. Identification of risk factors is integral to this effort and requires examination of a wide range of characteristics (e.g., biological, environmental, and social/systemic). Within the military population, combat exposure is a salient environmental factor to consider, with several large-scale investigations identifying links between combat exposure and alcohol misuse [10–13, cf. 14]. Of course, most combat-exposed service members do not ultimately develop AUD—implying the relationship between combat exposure and AUD risk depends on other characteristics.

Among the many factors that could influence responses to stress exposure, individual differences in genetic vulnerabilities might affect the likelihood of—though not determine—certain behavioral health outcomes. Thus, one plausible moderator of the effect of combat exposure on AUD risk is genetic liability to AUD. Knowledge of the genetics of alcohol misuse has grown considerably in recent years, with genome-wide association studies (GWAS) identifying specific risk variants associated with AUD and other alcohol-related phenotypes [for a review, see 15]. Investigators have further leveraged GWAS findings to create polygenic risk scores (PRS), which aggregate the small effects of many variants across the genome into an overall index of individual genetic liability for a phenotype of interest (e.g., AUD diagnosis). One informative and potentially powerful way to leverage these PRS is through gene-environment interaction (G x E) models that test how environmental influences might vary based on genetic factors [e.g., 16]. Such an approach recognizes that polygenic vulnerabilities may be more likely to be expressed in the context of environmental stressors, and vice versa.

The current study tests the hypothesis that the effect of combat exposure on AUD risk varies in part based on a service member’s genetic vulnerability to AUD. Consistent with the diathesis-stress framework [16, 17], we predict that greater combat exposure will be more strongly associated with subsequent AUD among service members with higher PRS for alcohol use disorder (AUD-PRS). In addition, we explore whether the hypothesized effect is specific to polygenic risk for AUD—or if the G x E effect generalizes to polygenic risk for a neighboring phenotype (problematic drinking).

The potential value of findings from this investigation are twofold. First, evidence of a G x E interaction between combat exposure and polygenic risk could inform etiological models of AUD in service members by providing better frameworks for understanding the combined genetic and environmental origins of AUD. Second, findings could enhance prevention efforts by enabling more precise targeting. For instance, if a large proportion of a population has a risk factor (e.g., combat exposure), it may be cost-prohibitive to implement preventive efforts; however, if only a subset of the population is genetically susceptible to the outcome, it may be feasible to implement interventions in that smaller high-risk group. Relating this to the current study, if combat exposure interacts with AUD-PRS in the expected manner, there may be opportunities to target interventions to the subset of service members at the intersection of having both genetic vulnerability to AUD and exposure to an environmental risk factor for AUD.

METHODS

Overview

Pre/Post Deployment Study (PPDS).

The PPDS component of the Army Study to Assess Risk and Resilience in Servicemembers (Army STARRS) is a multi-wave panel survey of three Brigade Combat Teams (BCTs) that deployed to Afghanistan in 2012 [18, 19]. Baseline surveys were administered 1–2 months prior to deployment (T0) and follow-ups were conducted approximately 1 month (T1), 3 months (T2), and 9 months (T3) after return from deployment. PPDS respondents were also invited to participate in the biomarkers portion of Army STARRS, which entailed providing a blood sample for genotyping.

STARRS Longitudinal Study (STARRS-LS).

STARRS-LS follows a probability sample of Army STARRS (including PPDS) participants who agreed to linkage of their survey responses and Army/Department of Defense (DoD) administrative data. STARRS-LS oversampled Army STARRS baseline survey respondents with a history of suicidality or mental disorder, women, Reserve/Guard members, and Special Operations soldiers. Prior publications give detailed descriptions of the sampling and other STARRS-LS procedures [20, 21]. This study uses data from wave 1 of STARRS-LS (LS1) to examine outcomes of PPDS participants several years after combat exposure. LS1 data were collected September 2016-April 2018 resulting in a mean interval of 46 months (range = 36–58) between the last wave of PPDS and LS1.

Written informed consent was obtained for PPDS and LS1 survey participation, linkage of survey responses to Army/DoD records, and collection of genomic data. Procedures were approved by the Institutional Review Boards of the collaborating institutions.

Participants

Of 9488 soldiers present for duty in the three BCTs at T0, 8773 consented to participate in the PPDS baseline survey. The current analysis was further restricted to participants who: deployed to Afghanistan with their BCT (n = 7738); completed the PPDS T2 survey (n = 6375) provided genomic data in PPDS (n= 3455); were of genetically determined European ancestry (the ancestry group for which our PRS was based; n = 2796); were recruited for and completed the LS1 survey (n = 1278); and provided necessary information for all measures used in the current analyses (n = 1203).

Measures

Combat exposure.

The PPDS T2 survey assessed combat exposure during the index deployment. Participants indicated the frequency with which they were exposed to 7 types of events (e.g., combat patrols/other dangerous duty; unit casualties). Response options were 0, 1, 2–4, 5–9, or 10+ times. Level of combat exposure was operationalized as the number of different combat-related events a participant reported being exposed to during the index deployment. To derive this score, raw responses were first recoded to reflect whether participants were exposed to a particular experience at least once during deployment (coded 1) or were not exposed to that experience during deployment (coded 0). The dichotomous variables were then summed, yielding a combat exposure score for each participant ranging from 0 (no exposure to the combat-related events assessed in the survey) to 7 (exposure to all combat-related events assessed in the survey).

Polygenic risk scores (PRS).

Further information on the DNA collection process, genotyping, quality control, and ancestry assignment, can be found in a prior report [22]. PRS for AUD were generated using summary statistics from a discovery GWAS of DSM-IV or DSM-III-R-based alcohol dependence, which were predecessors to the DSM-5 definition of AUD [23]. Effect weights were derived for each SNP using PRS-CS-auto [24] and a European ancestry reference panel. The weighted SNP effects were summed with PLINK 2.0 [25] to create a PRS for AUD (AUD-PRS), which was standardized within the overall PPDS genomic sample for subsequent analyses. We adjusted for residual within-ancestry variation by including the top 10 principal components as covariates in all regression models [26]. As a sensitivity analysis to examine whether findings were specific to AUD liability or could be captured by other alcohol-related genetic factors, we also generated PRS for problematic drinking (AUDIT-P-PRS) based on a GWAS of scores on the “problems” subscale of the Alcohol Use Disorders Identification Test [27].

Probable alcohol use disorder (AUD).

The LS1 survey evaluated probable DSM-5 AUD with 6 items adapted from the Composite International Diagnostic Interview Screening Scales [28] that assessed DSM-5 criteria for AUD [29]. Sample items include “How often did your drinking interfere with your responsibilities at home or work?” and “How often did you continue to drink even though it caused ongoing problems with your family, friends, or coworkers?” Participants reported how often they experienced each alcohol-related problem in the past 30 days (never, 1–3 days/month, 1–2 days/week, 3–4 days/week, or every/nearly every day). Probable DSM-5 AUD was defined as responding to at least two of the items with a response of 1–2 days/week or more (coded 1; otherwise coded 0).

Controls.

Socio-demographic information was acquired via self-report in the PPDS T0 survey; this included information about the participant’s sex, age, education level, and marital status. These variables were used to describe the sample and were included as controls in the regression models. Sex was assessed with the item “Are you male or female?” yielding a dichotomous variable (1 = Male; 0 = Female). Education was categorized as possessing a high school diploma or GED/equivalent, an undergraduate college degree, or a graduate degree. Marital status was categorized as married, never married, or other (divorced, separated, or widowed). Current military status was assessed at LS1 and categorized as active, deactivated (applies to National Guard and Reserve members), retired, or separated.

Data analysis

Analysis of the effect of combat exposure on the predicted probability of AUD was based on logistic regression models using the ‘glm’ function in the statistical software R version 4.3.1 [30]. Logistic regression models estimate the logarithm of the odds ratio (log-odds) for a dichotomous outcome based on one or more independent variables [31]. We followed a procedure that involved two logistic regression models. The first model estimated the direct effects of combat exposure on the log-odds of AUD diagnosis. The second model estimated the interactive effect of combat exposure and AUD-PRS on the log-odds of AUD diagnosis. Once estimated, log-odds were converted to odds ratios to aid interpretability [32]. To test whether findings were specific to the AUD-PRS, we performed a sensitivity analysis that entailed fitting analogous models using an alternative alcohol-related PRS (AUDIT-P-PRS) [27].

RESULTS

The sample of 1203 participants was 95.8% male (n = 1153) and 4.2% female (n = 50). At PPDS baseline, the mean age of the sample was 26.6 (SD = 5.9) years. In terms of education level, 68.5% (n = 824) of participants had a high school diploma or GED/equivalent, 28.5% (n = 343) had an undergraduate college degree, and 3.0% (n = 36) had a graduate degree. With respect to marital status, 60.5% (n = 723) of the sample was married, 32.5% (n = 389) was never married, and 7.0% (n = 84) was either divorced, separated, or widowed. The military status of the sample at the LS1 follow-up was: 48.8% (n = 587) active, 9.2% (n = 111) deactivated, 9.8% (n = 118) retired, and 32.2% (n = 387) separated.

The average combat exposure score was 3.6 (SD=1.8). With respect to specific combat-related experiences during the index deployment, 77.5% of participants were involved in combat patrols or other dangerous duty, 65.0% fired rounds at the enemy or received incoming fire, 8.0% were directly wounded, 57.9% had someone in their unit die, 31.8% had someone close to them (e.g., friend, relative, mentor) die, 59.5% were exposed to sights/sounds of wounded or dying people, and 58.4% experienced other highly stressful events. At the LS1 follow-up, 9.6% (n=116) reported symptoms consistent with AUD. Correlations among combat exposure score in PPDS, AUD-PRS, and probable AUD at LS1 may be found in Table 1. The lack of a significant correlation between the combat exposure score and the AUD-PRS (r = .03, p = .33) suggests negligible gene-environment correlation, at least at the polygenic level, that could complicate interpretation of the results.

Table 1.

Means, standard deviations, and correlations of study variables

	Variable	Mean	SD	1	2	3	4	5	6	7	8	9	10	11	12	13	14	15	16
1.	Combat exposure	3.58	1.83	1
2.	Probable AUD	.10	.30	.07	1
3.	AUD-PRS	.00	.24	.03	.02	1
4.	Age	26.6	5.93	−.13	−.03	−.03	1
5.	Male	.96	.20	.13	.01	−.02	.02	1
6.	Female	.04	.20	−.13	−.01	.02	−.02	−1	1
7.	High School	.68	.46	−.01	.02	.06	−.31	.00	−.00	1
8.	College	.29	.45	.01	−.01	−.05	.22	−.02	.02	−.93	1
9.	Graduate School	.03	.17	.01	−.01	−.02	.26	.04	−.04	−.26	−.11	1
10.	Married	.60	.49	−.06	−.07	−.01	.33	.08	−.08	−.03	.01	.07	1
11.	Never Married	.33	.47	.05	.04	−.01	−.38	−.04	.04	.03	−.00	−.06	−.86	1
12.	Other Rel. Status	.07	.25	.03	.05	.04	.06	−.07	.07	.02	−.01	−.03	−.34	−.20	1
13.	Active Duty	.49	.50	−.06	−.10	−.02	.13	.03	−.03	−.11	.08	.09	.21	−.21	−.01	1
14.	Inactive	.10	.29	−.06	.00	−.05	−.10	.02	−.02	−.00	.01	−.02	−.10	.11	−.03	−.31	1
15.	Retired	.10	.30	.07	−.00	.05	.23	−.00	.00	.02	−.01	−.03	.11	−.12	.02	−.32	−.10	1
16.	Separated	.32	.47	.06	.12	.03	−.23	−.04	.04	.10	−.08	−.07	−.23	.23	.02	−.67	−.22	−.23	1

Note. N = 1203. Correlations above the absolute value of .06 are statically significant at p <.05. AUD = alcohol use disorder. AUD-PRS = polygenic risk score for alcohol use disorder.

The first hypothesis test examined the effect of combat exposure on probable AUD. Model 1 in Table 2 shows that participants with higher combat exposure were more likely to screen positive for AUD at LS1 (AOR = 1.12), but this effect was not statistically significant (p = .051). The second hypothesis test examined the conditional effect of combat exposure on probable AUD as a function of AUD-PRS. Model 2 in Table 2 shows a statistically significant combat exposure x AUD-PRS interaction (AOR = 1.60, p = .033), suggesting that the effect of combat exposure on odds of AUD varies based on the level of genetic liability to AUD. Figure 1 illustrates that, among individuals with low genetic risk (1 SD below mean AUD-PRS), combat exposure does not appear associated with risk of AUD. In those with average genetic risk (mean AUD-PRS), a pattern is observed whereby the risk of probable AUD increases as the combat exposure score increases. Moreover, among individuals with high genetic risk (1 SD above mean AUD-PRS), the positive association between combat exposure and risk of probable AUD is even stronger. A comparable interaction effect was not observed in the sensitivity model that substituted the AUDIT-P-PRS for the AUD-PRS (AOR = 1.26, p = .53 for combat exposure x AUDIT-P-PRS; see Supplementary Table 1 for full results).

Table 2.

Logistic regression models of the effects of combat exposure and AUD-PRS on odds of probable AUD diagnosis

Independent Variable	Control Model		Model 1		Model 2
	Adjusted OR	95% CI	Adjusted OR	95% CI	Adjusted OR	95% CI
Age	.99	(.96 – 1.04)	1.01	(.96 – 1.05)	1.01	(.97 – 1.05)
Male vs. Female	1.47	(.57 – 4.99)	1.19	(.45 – 4.10)	1.15	(.44 – 4.00)
Education Status
College vs. High School	.98	(.61 – 1.55)	.98	(.61 – 1.55)	.98	(.61 – 1.55)
Graduate School vs. HS	1.10	(.25 – 3.45)	1.05	(.23 – 3.36)	1.07	(.24 – 3.42)
Marital Status
Never vs. Married	1.22	(.77 – 1.93)	1.22	(.77 – 1.95)	1.23	(.77 – 1.96)
Other vs. Married	1.97	(.98 – 3.72)	1.72	(.85 – 3.29)	1.74	(.86 – 3.34)
Military Status
Inactive vs. Active	1.53	(.72 – 3.05)	1.71	(.79 – 3.44)	1.63	(.75 – 3.31)
Retired vs. Active	1.50	(.70 – 2.99)	1.45	(.67 – 2.93)	1.38	(.64 – 2.81)
Separated vs. Active	2.32**	(1.49 – 3.69)	2.32**	(1.47 – 3.71)	2.35**	(1.48 – 3.75)
Combat Exposure			1.12ψ	(1.00 – 1.26)	1.12ψ	(1.00 – 1.26)
AUD-PRS			1.20	(.52 – 2.74)	.19	(.03 – 1.27)
Combat Exposure x AUD-PRS					1.60*	(1.03 – 2.46)

Note. N = 1203. Both models include the 10 ancestral principal components as covariates; these estimates were omitted here to reduce table clutter. AUD-PRS = polygenic risk score for alcohol use disorder, AUD = alcohol use disorder.

^ψp < .10 (two-tailed),

^*p < .05 (two-tailed),

^**p < .01 (two-tailed).

Figure 1.

Visualization of the interactive effect of combat exposure level and polygenic risk for alcohol use disorder (AUD-PRS) on odds of probable alcohol use disorder at follow-up. Shading around each line illustrates the confidence interval for each interaction effect.

DISCUSSION

An improved understanding of the biopsychosocial pathways to AUD within military populations may help reduce incidence of this costly and disabling disorder. In this study, we found that greater combat exposure was linked to higher odds of probable AUD approximately four years post-deployment, but this relationship varied as a function of AUD-related genetic risk. Consistent with the diathesis-stress framework, we observed that greater combat exposure was more strongly related to later AUD among soldiers with higher genetic liability to AUD diagnosis. This effect appeared specific to AUD-related genetic risk—a similar interaction was not observed in a sensitivity model that replaced the AUD-PRS with an index of genetic risk for problematic drinking (AUDIT-P-PRS).

Our findings are notable for several reasons. First, they suggest that soldiers exposed to combat may show differential risk for later AUD in part based on their individual differences in genetic risk profiles for AUD. In this study, combat exposure levels alone were not significantly associated with subsequent AUD, suggesting that, overall, this environmental factor did not have a robust effect on risk of later AUD. Nonetheless, combat exposure emerged as a salient risk factor for AUD among soldiers with higher genetic risk. It is possible that more “potent” environmental risk factors for AUD would have interacted with AUD-PRS in a similar (or more pronounced) manner. Alternatively, it may be the case that PRS will prove particularly informative in scenarios involving exposures with smaller main effects, which allow the spectrum of individual genetic differences to exert an influence [33]. For instance, examination of G x E effects may reveal that the effect of an environmental factor is negligible for individuals with low genetic susceptibility, but substantial for individuals with high genetic susceptibility to the outcome of interest (along the lines of the pattern observed here). Replication of the current results is needed before firm conclusions can be drawn, especially given that—like many studies examining interactions between genetic risk and environmental exposure [16]—the interaction effect size was modest. Our interaction model suggests that the combination of high combat exposure and high AUD-PRS was associated with an approximately 60% increased risk of later AUD, relative to the risk associated with average levels of combat exposure and AUD-PRS (holding other factors in the model constant, and with “high” defined as 1 SD above the mean).

Second, the current results may have implications for understanding the role of genetic vulnerability when predicting longer-term outcomes following exposure to environmental risk factors. In this study, the interval between the environmental exposure and the AUD outcome was several years. In the period between when combat exposure and AUD were assessed, many soldiers would have undergone a variety of significant stressors to include relocating to other Army units and installations and potentially separating from the military. Nonetheless, level of combat exposure was linked to probable AUD diagnosis several years later when considered in conjunction with a soldier’s polygenic risk for that outcome. One possibility is that, in some cases, PRS may have greater influence on the course vs the emergence of behavioral outcomes. Whereas acute symptoms may be triggered by certain stressors in a large proportion of exposed individuals, persistence of those symptoms well beyond the initial risk period may only be likely among individuals with high genetic susceptibility to the focal outcome (e.g., AUD).

As noted earlier, our sensitivity analysis revealed no significant effects involving an index of genetic liability to problematic drinking captured by the AUDIT (AUDIT-P-PRS). The AUD-PRS is derived from a GWAS of the binary outcome of alcohol dependence, established via clinician diagnosis or structured interview [23], whereas the AUDIT-P-PRS is derived from a GWAS of self-assessed alcohol-related problems on a continuum based on the “problems” subscale of the AUDIT measure [27]. The phenotype examined in the former GWAS is more precisely matched to our outcome (i.e., probable AUD diagnosis); however, the AUDIT-P score was of secondary interest because it also captures problems related to drinking. The null results obtained with the AUDIT-P-PRS may imply that the combat exposure x AUD-PRS interaction is driven by genetic features unique to individuals who surpass thresholds for clinical diagnosis (i.e., those who continue to drink despite serious social, occupational, or health problems).

An important caveat to the above interpretations is that behavioral health outcomes are influenced by a wide range of factors and, in many cases, increased genetic risk does not lead to manifestation of the focal outcome. Any implementation strategy must educate those who receive information pertaining to genetic vulnerabilities about the limits of genetic prediction, to avoid interpretations of model-estimated risk as evidence of genetic determinism [34]. Beyond this general point, it is important to highlight limitations specific to the current study. First, the clinical significance of the small G x E effect is uncertain. Our findings only suggest that PRS have promise for the applications we have discussed (e.g., facilitating targeting of preventive interventions for combat-exposed service members). A more robust evidence base is needed before it can be concluded that PRS have utility for these purposes. Second, the discovery GWAS of alcohol dependence diagnosis [23] was relatively small and there are emerging, larger GWAS of alcohol-related outcomes. Despite this, we calculated PRS based on that GWAS because its phenotype aligned most closely with our study outcome. Third, our outcome measure (probable AUD diagnosis) was based on respondents’ self-assessments of their alcohol use and its consequences. Self-report measures are generally susceptible to response biases and substance use measures are specifically vulnerable to under-reporting. Moreover, in the case of AUD evaluation, respondents may not always have good insight into the negative consequences of their alcohol use. Fourth, some cases of AUD would not have been detected in the LS1 survey because the timeframe for AUD assessment was the past 30 days, implying that fairly recent AUD cases that were temporarily remitted or below the diagnostic threshold at LS1 would have been missed. Fifth, the small number of females in the sample precluded an analysis of sex differences. Additionally, we were only able to test our hypothesis among individuals of European ancestry due to lack of availability of PRS for other groups in this sample. The current results may not generalize to females or individuals of other (non-European) ancestries. Investigating whether a similar G x E effect is observed among females and individuals of other ancestries should be prioritized in future research on this topic. Recent GWAS data [35] and novel data analytic methods [36] will enable the calculation of PRS for alcohol-related outcomes among individuals of other ancestries.

Conclusion

In a sample of US Army soldiers who deployed to Afghanistan in 2012, the relationship between level of combat exposure and risk of later AUD varied in part based on individual differences in genetic liability to AUD. Specifically, greater combat exposure was more strongly related to AUD risk among soldiers with heightened genetic vulnerability to AUD. Continued investigation is warranted to determine whether PRS have a role to play in stratifying AUD risk levels within stress-exposed groups such as combat-deployed service members. Advances in quantification of genetic risk for AUD and in understanding of complex effects of genetic risk factors may reveal opportunities to focus preventive efforts on smaller groups of stress-exposed individuals with high genetic susceptibility. The utility of PRS in this endeavor will ultimately depend on whether prediction of AUD is substantially enhanced by focusing on subgroups with high polygenic risk for AUD and/or related traits and disorders.

Supplementary Material

Supplementary Table 1

Supplementary Table 1. Logistic regression model of the effects of combat exposure level and AUDIT-P-PRS on odds of probable AUD diagnosis at follow-up.

Abbreviations used are as follows:

AUD

alcohol use disorder

GWAS

genome-wide association study

PRS

polygenic risk score

G x E

gene-environment interaction

AUD-PRS

polygenic risk score for alcohol use disorder

PPDS

Pre/Post Deployment Study

Army STARRS

Army Study to Assess Risk and Resilience in Servicemembers

BCT

Brigade Combat Team

STARRS-LS

Study to Assess Risk and Resilience in Servicemembers-Longitudinal Study

LS1

wave 1 of STARRS-LS

SNP

single nucleotide polymorphism

AUDIT-P-PRS

polygenic risk score for problem drinking per the Alcohol Use Disorders Identification Test