Effects of sexual assault on alcohol use and consequences among young adult sexual minority women

Abstract

Objective

The purpose of this study was to examine effects of sexual assault victimization on later typical alcohol use and alcohol-related consequences among young sexual minority women (SMW).

Method

Data were collected over four annual assessments from a national sample of 1,057 women who identified as lesbian or bisexual and were 18 to 25 years-old at baseline. Marginal structural modeling, an analytic approach that accounts for time-varying confounding through the use of inverse probability weighting, was used to examine effects of sexual assault and its severity (none, moderate, severe) on typical weekly number of drinks consumed and number of alcohol-related consequences one-year later as well as two-year cumulative sexual assault severity on alcohol outcomes at 36-month follow-up.

Results

Findings showed that compared to not experiencing any sexual assault, severe sexual assault at the prior assessment was associated with a 71% higher number of typical weekly drinks (Count Ratio [CR] = 1.71; 95% confidence interval [CI]: 1.27, 2.31) and 63% higher number of alcohol-related consequences (CR = 1.63; 95% CI: 1.21, 2.20). Effects were attenuated when comparing moderate to no sexual assault; however, the linear trend across sexual assault categories was statistically significant for both outcomes. There were also effects of cumulative levels of sexual assault severity over two years on increased typical drinking and alcohol-related consequences at end of follow-up.

Conclusions

Sexual assault may be an important cause of alcohol misuse among SMW. These findings further highlight the need for strategies to reduce the risk of sexual assault among SMW.

Background

Although reasons for the disparity are unclear, numerous studies have shown that sexual minority women (SMW) are more likely than heterosexual women to experience sexual assault in childhood and in adulthood. In a recent systematic review of the literature, prevalence of lifetime sexual assault ranged from 16% to 85% among SMW, with a median estimate of 43% of lesbian and bisexual women reporting adult or childhood sexual victimization across studies (Rothman, Exner, & Baughman, 2011). As noted by Rothman and colleagues, most studies on sexual victimization among SMW have focused predominantly on childhood sexual abuse, indicating the need for additional research that examines adult sexual assault in this population.

Despite recent advances in societal acceptance of sexual minorities in the U.S., negative social attitudes and behaviors towards SMW are still widespread (Herek, 1997). The minority stress model posits that the cumulative stressors associated with sexual minority status can lead to negative physical and mental health outcomes and help to explain health disparities (Hatzenbuehler, Phelan, & Link, 2013; Meyer, 2003). There is a growing body of evidence indicating that experiences of interpersonal and institutional discrimination increase problematic drinking among sexual minorities (Hatzenbuehler, Keyes, & Hasin, 2009; McCabe, Bostwick, Hughes, West, & Boyd, 2010; McLaughlin, Hatzenbuehler, & Keyes, 2010). Indeed, young adult SMW are at significantly elevated risk of heavy episodic drinking (consuming 4 or more drinks in 2 hours; National Institutes on Alcohol Abuse and Alcoholism, 2004) and of experiencing problems related to alcohol use compared to heterosexual women (Drabble, Midanik, & Trocki, 2005; McCabe, Hughes, Bostwick, West, & Boyd, 2009; Wilsnack et al., 2008). For example, lesbian and bisexual women aged 20 to 34 reported higher weekly alcohol consumption and less abstinence compared to older lesbian and bisexual women and to heterosexual women (Gruskin, Hart, Gordon, & Ackerson, 2001). In a study of college females, lesbian/bisexual women were 10.7 times more likely to consume alcohol than heterosexual women (Ridner, Frost, & Lajoie, 2006). Elevated risk of sexual assault has been generally included as an example of structural stigma or minority stress disproportionately affecting sexual minorities (Balsam, Lehavot, & Beadnell, 2011; Hughes, Johnson, & Wilsnack, 2001; Hughes et al, 2010). As such, one contributor to this elevated risk for heavy episodic drinking may be SMW’s greater exposure to sexual assault (Drabble, Trocki, Hughes, Korcha, & Lown, 2013). In both heterosexual women and SMW, those who experience sexual assault may be at greater risk for a variety of health consequences, including alcohol misuse. For example, a history of childhood sexual abuse is associated with later alcohol abuse and dependence (Afifi, Henriksen, Asmundson, & Sareen, 2012; Danielson et al., 2009; Gilmore et al., 2014). Although not entirely consistent, cross-sectional research has found that SMW who report sexual assault in adulthood are also more likely to engage in greater alcohol consumption and heavy episodic drinking than those who report no lifetime sexual assault (Hughes et al., 2010; Ullman, 2003; Wilsnack, Wilsnack, Kristjanson, Vogeltanz-Holm, & Windle, 2004). This may be explained at least in part by negative reinforcement models that suggest that alcohol use may reduce distress, which negatively reinforces continued and increased use of alcohol (Baker et al, 2004).

Longitudinal research is needed to establish temporal ordering and potential causation. Multiple studies in general female samples that did not explicitly consider sexual orientation have observed associations between sexual assault victimization and subsequent alcohol outcomes (e.g., Bryan et al., 2015; Parks, Hsieh, Taggart & Bradizza, 2014; Testa, Hoffman, & Livingston, 2010; Ullman, 2016). To our knowledge, however, no prospective studies have examined these associations among SMW specifically. As longitudinal research moves forward on this topic, there are important methodologic issues that should be considered. First, the definition of sexual assault varies across studies and ranges from unwanted sexual contact to attempted or completed rape. Given that more severe forms of sexual assault tend to be more strongly associated with negative outcomes (Turchik, 2012), differing levels of assault severity may be associated with differing levels of drinking and alcohol-related problems. The tactics used by the perpetrator to obtain unwanted sex or sexual contact (i.e., verbal coercion, intoxication, and force or threats of force) may play a role in subsequent drinking by the victim and are not always considered (Littleton, Grills-Taquechel, & Axsom, 2009; Zinzow et al., 2012). Further, researchers often ignore the frequency of sexual assault experiences which is problematic given both high occurrence of repeated victimization and associations between the frequency of victimization and negative health consequences (Jozkowski & Sanders, 2012). Thus, a measure of sexual assault severity that incorporates both the severity of the assault experienced as well as the frequency of prior assaults may provide a more sensitive measure of “dose” of exposure and, thus, be a stronger predictor of health consequences.

Second, longitudinal studies with repeated measures of sexual assault exposure and alcohol-related outcomes provide opportunities to examine different aspects of how sexual assault could lead to alcohol misuse. For example, longitudinal studies can examine lagged longitudinal effects of sexual assault on alcohol outcomes at the following year across multiple study assessments. Further, the effects of cumulative exposure to sexual assault across multiple study waves on alcohol use and problems can be examined. SMW appear to be at elevated risk for not only any lifetime sexual assault victimization, but also revictimization (Hughes et al., 2010). Cross-sectional research suggests that individuals experiencing multiple victimization experiences may be at particularly elevated risk for substance use disorders (Hughes, McCabe, Wilsnack, West, & Boyd, 2010).

Finally, obtaining unbiased estimates of effects in longitudinal studies with repeated measures of the exposure can be challenging due to the additional possibility of time-varying confounding. For example, when examining effects of sexual assault on alcohol-related outcomes, it is possible that associations may be confounded by other time-varying factors such as prior levels of mental health symptoms (e.g., depression, anxiety, post-traumatic stress disorder), alcohol use, and prior trauma exposure. Approaches such as marginal structural modeling can be used to account for time-varying confounding and reduce potential bias through the use of inverse probability weights (IPWs) of exposure (Robins, Hernan, & Brumback, 2000). Marginal structural models use a counterfactual framework that can estimate an average causal effect comparing the potential outcome had a given individual been set to one covariate history vs. a different covariate history, possibly contrary to the fact (VanderWeele, Hawkley, Thisted, & Cacioppo, 2011). Rather than include time-varying and other confounders in the statistical model, this approach weights subjects by the inverse probability of their own exposure (e.g., sexual assault) status according to covariates. Thus, when applying the IPWs, individuals who are underrepresented for their sexual assault status according to covariate history are given greater weight, whereas those who are overrepresented for a certain exposure level are given lower weight. This results in a “pseudo-population” with balanced distribution of time-varying and time-fixed covariates across levels of the exposure history, and application of IPWs should yield unconfounded estimates of the effects of sexual assault. Further introduction to marginal structural modeling and its applications are found elsewhere (e.g., Bodnar, Davidian, Seiga-Riz & Tsiatis, 2004; Thoemmes & Ong, 2016). To our knowledge, no longitudinal studies in general samples or in SMW-specific samples have utilized marginal structural modeling to investigate the potential causal relation between sexual assault and subsequent alcohol use and consequences.

In the current study we used marginal structural modeling to examine the effects of sexual assault severity assessed at one study wave on typical alcohol use and drinking-related consequences at the next annual study wave in a national sample of young adult SMW. In addition, we also examined whether cumulative exposure to sexual assault over two years was associated with alcohol use and consequences at the final study wave.

Method

Participants and Procedures

Participants in this study were part of a longitudinal study of young adult (ages 18 to 25) SMW’s health and health behaviors. Women were recruited to participate via advertisements placed on the social networking website Facebook in such a way that only women who reported lesbian or bisexual identity on their profile and who were between the ages of 18–25 were shown the ad. Upon logging into Facebook, potential participants were shown the study advertisement (displayed in the side bar) with a link to the screening assessment. We used two types of advertisements: those that included sexual-minority-specific content (e.g., “LGB women needed for an online study on health behaviors”) and those that were non-LGB-specific (e.g., “we need you for an online study on partying”). Online advertisements were also placed on Craigslist in 12 metropolitan areas in the U.S.: Atlanta, Austin, Boston, Chicago, Houston, Los Angeles, New York, Philadelphia, San Francisco, Seattle, South Florida, and Washington, D. C. Craigslist postings included a brief summary of the project and a URL link to the screening assessment.

Women who responded to the advertisements and accessed the study screening site were first shown information about the study. After agreeing to participate, potential participants were then routed to a 5-minute screening assessment. A total of 4,119 women completed the online screening survey. Study eligibility criteria included: 1) residing in the U.S., 2) having a valid e-mail address, 3) being between the ages of 18 and 25 years, and 4) reporting lesbian or bisexual identity at the time of the assessment. Eligible women (n = 1,877) were then invited to participate in the study. Of those eligible, 1,083 (57.7%) provided consent for participation in the larger study. Inconsistencies in the data that suggested a small number of participants were falsifying information (e.g., inconsistent birth dates over time) led us to omit 2.4% of the participants in the baseline sample; 1,057 were retained in the study. Compared to those who were retained in the study, those women who were eligible and did not consent were less likely to be White race (67.9% vs. 78.8%; χ²(1) = 30.5; p<.001) and more likely to be of Hispanic ethnicity (13.4% vs. 10.2%; χ²(1) = 4.6; p = .03). However, there were no statistically significant differences in age or sexual orientation.

Data were collected online at four annual assessments. Participants were paid $25 for completion of the baseline survey and $30 for completion of each of the three annual follow-up assessments. A Federal Certificate of Confidentiality was obtained for the study. The University’s Institutional Review Board reviewed and approved all study procedures.

Measures

Sexual Assault

The revised Sexual Experiences Survey (SES) was used to assess sexual assault severity (Koss et al., 2007). This measure asks about experiences of different types of unwanted sexual behaviors (e.g., fondling, attempted or completed oral, vaginal or anal penetration) and tactics used to obtain each outcome (e.g., coercion, intoxication, and threat or use of physical force). Participants indicated how often (0 = never to 3 = 3 or more times) they experienced each unwanted sexual behavior by each tactic (e.g., how often they experienced attempted vaginal sex by coercion). At the baseline assessment, questions were asked in reference to “since age 18” and in follow-up assessments the reference period was “in the past year” (i.e., the time since the last assessment). An overall severity score was calculated as described by Davis and colleagues (2014). First, each sexual experience and tactic combination was assigned a severity rank from 0 (no history of sexual assault) to 6 (attempted or completed rape by threat or use of physical force). For each sexual experience and tactic combination, the severity rank was multiplied by the frequency of its occurrence and then summed for a total combined severity-frequency score with a possible range of 0–63. This severity score has shown strong convergent validity, especially among populations that report higher rates of assault (Davis et al., 2014).

Alcohol Consumption

Typical weekly alcohol consumption during the previous three months was assessed using the Daily Drinking Questionnaire (DDQ; Collins, Parks, & Marlatt, 1985). Participants were asked, “Consider a typical week during the last three months. How much alcohol (measured in number of standard drinks), on average, do you drink each day of a typical week?” Standard drinks were defined as 1.5 oz. of liquor, 5 oz. of wine, or 12 oz. of beer. Typical weekly drinking was the sum of the number of standard drinks for each day of the typical week.

Alcohol Consequences

Alcohol consequences were measured using the Young Adult Alcohol Consequences Questionnaire (YAACQ; Read, Kahler, Strong, & Colder, 2006). The YAACQ obtains self-ratings of 48 possible drinking consequences. For each of the 48 consequences, participants indicate whether or not they experienced that consequence in the previous 30 days. The sum of the responses was calculated for a count of the number of past month alcohol consequences.

Covariates

A number of demographic and other measures were used for estimation of the predicted probability of past year sexual assault and level of severity. Demographic characteristics included age at baseline, race/ethnicity, sexual identity (lesbian, bisexual), and parent’s highest level of education. Additional psychosocial constructs were included because they tend to co-occur with sexual assault. Mental health problems were assessed using validated and commonly used measures including the PTSD Checklist (PCL) to assess post-traumatic stress disorder symptoms (Ruggiero, Del Ben, Scotti, & Rabalais, 2003), the Center for Epidemiologic Studies Depression (CES-D) scale to assess depression symptoms (Radloff, 1977), and the GAD-7 to assess generalized anxiety symptoms (Spitzer, Kroenke, Williams, & Lowe, 2006). The Daily Heterosexist Experiences Questionnaire (DHQ) was used to assess participants’ perceived minority stress due to their LGBT identity (Balsam, Beadnell, & Molina, 2013). Items ask about 38 stressors that LGBT individuals might experience such as difficulty finding a partner, pretending to be heterosexual, and being verbally harassed due to being LGBT. For each of the above psychosocial scales, the internal consistency was >.90 in this sample. Other traumatic events were assessed using the Traumatic Life Events Questionnaire (TLEQ; Kubany et al., 2000). At baseline, this measure asked about events in reference to one’s lifetime, but at annual follow-up visits the reference period was the past year. Further, the baseline TLEQ asked about sexual assaults that were experienced before age 13 and between age 13 and 18. Perceived drinking norms for sexual minority women were also assessed using a modified version of the Drinking Norms Rating Form (Baer, Stacy, & Larimer, 1991; Litt, Lewis, Rhew, Hodge, & Kaysen, 2015) that asked participants about the typical number of drinks consumed per week by a typical lesbian or bisexual woman.

Analytic Plan

We used marginal structural modeling to examine effects of sexual assault severity on alcohol use and drinking-related consequences. As the first step for these analyses, the inverse probability weights were calculated. To derive the IPWs for past year sexual assault, we first calculated the probability for being at one’s own level of sexual assault severity at each of the follow-up time points according to time-varying covariates measured at prior assessments as well as baseline covariates. Time-varying covariates included earlier levels of alcohol use and drinking-related consequences, past year sexual assault severity, other traumatic events, perceived heterosexism, perceived descriptive norms, PTSD symptoms, depression symptoms, and generalized anxiety symptoms. Baseline time-fixed covariates included age, race (White, Black, other race), Hispanic ethnicity, sexual identity (bisexual, lesbian), highest level of parents’ education, and sexual assault prior to age 18 (none, childhood [<13 years], adolescent [13–18 years]).

The distribution of the SES score at each of the follow-up visits was severely positively skewed with the vast majority of scores being 0 (>70%). Because of this, calculation of weights for the continuous SES score based on a probability density using a linear or log-linear model could lead to biased estimates (Naimi, Moodie, Auger, & Kaufman, 2014). The SES score was, thus, re-categorized into 3 groups: (1) no sexual assault (score of 0), (2) moderate severity (score of 1 to 6), and (3) high severity (score of 7 or higher). These categories were selected in order to ensure that there were sufficient numbers within each category and that the size of the two highest bins was similar. Because of the categorical nature of this measure we used a cumulative probability (ordinal logistic) regression model to regress the sexual assault category at each relevant exposure time point (12- and 24-month assessment) on covariates reported prior to this time point. Model-predicted probabilities were used to derive the probability of one’s observed category of exposure (e.g., the predicted probability of being in the highest category of sexual assault severity for a woman who was actually in the highest category of severity) at each time point. These predicted probabilities served as the denominator of the IPWs. To improve precision of estimates, we used stabilized IPWs such that the numerator of the IPWs was the predicted probability of one’s own level of sexual assault severity according to time-fixed covariates only (e.g., race, baseline age). Thus, the stabilized IPW (SW) for subject i at follow-up visit j is defined as

$$S{W}_{ij}(t)={\displaystyle \prod _{j=1}^t\frac{p(A_{ij}=k_{ij}{|\overline{A}}_{i,j-1},V_i)}{p(A_{ij}=k_{ij}{|\overline{A}}_{i,j-1},V_i,{\overline{L}}_{i,j-1})}}$$

where A is the category of sexual assault victimization, k is the level of exposure, ${\overline{A}}_{i,j-1}$ and ${\overline{L}}_{i,j-1}$ represent the exposure and covariate history up to time j, and V represents a vector of time-fixed covariates. Further, we truncated IPWs such that extreme low or high values were recoded to the 1^st or 99^th percentile in order to increase precision of estimates (Cole & Hernan, 2008).

A first set of models examined lagged effects of sexual assault severity at wave j-1 on alcohol outcomes assessed at the following wave, j. For this set of models, the weights were applied to Generalized Estimating Equations (GEE) models with robust standard errors and a working independence correlation in order to account for nesting of observations within individuals (Diggle, Heagerty, Liang, & Zeger, 2002). The alcohol outcome (typical drinks per week or alcohol-related consequences) at study time point j was regressed on sexual assault severity at the previous time point, j-1. Because both alcohol outcomes were discrete counts that showed evidence of over-dispersion, a negative binomial rather that Poisson form of the model was used. In negative binomial models, covariates are connected to the outcome via a log link. Coefficients for covariates are often exponentiated (e^β) to yield count ratios (CRs; also referred to as rate ratios) that describe the proportional change in the count associated with a one-unit increase in the covariate (Atkins & Gallop, 2007; Hilbe, 2014). Baseline time-fixed covariates were included in the GEE model to improve precision of parameter estimates (Cole & Hernan, 2008). The second set of models examined the effects of cumulative sexual assault severity over two years on the alcohol outcomes at the 36-month assessment. Cumulative severity was defined as the sum of SES categories at 12- and 24-month assessments (possible range: 0 to 4). Because only the 36-month outcomes were examined for these models, a single-level (non-GEE) negative binomial regression model was performed. The IPWs for this set of models was the product of the two wave-specific (12- and 24-month) IPWs. The same baseline time-fixed covariates were included in these models. For comparison, we also ran “traditional” non-weighted models that included the same time-fixed covariates for both the lagged and cumulative models.

As post-hoc analyses, we also examined whether effects of sexual assault differed by baseline sexual orientation (lesbian vs. bisexual) by including sexual assault by sexual orientation interaction terms into the models.

There was 20%, 28%, and 30% of the sample missing data at 12-month 24-month and 36-month assessments, respectively. To account for missingness, we used multiple imputation where missing values were imputed according to various covariates including earlier and later measures of the variable (Graham, 2009). Assuming data are missing at random (MAR) such that missingness is not associated with unmeasured variables, parameter estimates using multiple imputation should be unbiased. At any given study visit, missingness of typical weekly drinking and alcohol-related consequences were not statistically significantly associated with those same measures from earlier or later assessments. Although not offering definitive proof, this is consistent with the MAR assumption. Imputation was performed using the multiple imputation chained equations (MICE) approach and 20 imputed datasets were created (Azur, Stuart, Frangakis, & Leaf, 2011; White, Royston, & Wood, 2011). Within each imputed dataset, the IPWs were calculated and then the weighted model was run. Parameter estimates were combined across the datasets and standard errors were calculated to account for the uncertainty of imputed values according to Rubin’s rules (Rubin, 2004). All analyses were performed using Stata 14 (StataCorp, College Station, TX).

Results

Table 1 displays the distribution of selected demographics and other characteristics of the study sample. Nearly 60% of the sample identified as bisexual. The proportion of participants who had clinically elevated scores for depression, GAD, or PTSD was notably elevated compared to general population and primary care samples (Lowe et al., 2008; Mair et al., 2009; Stein, McQuaid, Pedrelli, Lenox, & McCahill, 2000; Walker, Newman, Dobie, Ciechanowski, & Katon, 2002). Consistent with extant literature, history of sexual assault was common with 39% of the sample reporting sexual assault during childhood (before age 13) and 35% reporting sexual assault during adolescence (between ages 13 and 18).

Table 1.

Distribution of baseline characteristics

Characteristic	% or mean (SD) N = 1,057
Age (years)	20.9 (2.1)
Bisexual vs. lesbian sexual identity	59.5
Race/ethnicity
White, non-Hispanic	63.4
African American, non-Hispanic	8.5
Asian American, non-Hispanic	2.5
Multiracial, non-Hispanic	14.3
Other race, non-Hispanic	2.2
Hispanic, any race	11.3
Elevated depression symptoms (CES-D ≥16)	69.5
Elevated generalized anxiety symptoms (GAD-7 ≥10)	69.5
Elevated PTSD symptoms (PCL ≥50)	25.7
Any sexual assault during adolescence (age 13–18)	35.2
Any sexual assault during childhood (before age 13)	38.9

Abbreviations: CES-D: Center for Epidemiologic Studies-Depression scale; GAD-7: Generalized Anxiety Disorder-7 scale; PCL: Post-traumatic stress disorder checklist.

Reports of sexual assault in adulthood were also common. As shown in Table 2, more than one-half of the sample reported experiencing moderate or severe sexual assault between their 18^th birthday and the baseline survey. Further, at each annual follow-up assessment more than 20% of the sample reported some form of sexual assault in the previous year. Based on the continuous version of the sexual assault severity scale (range: 0 to 30), the mean score was 2.0 at the 12- (SD = 5.2) and 24-month (SD = 5.4) follow-up visits and 1.6 (SD = 5.0) at the 36-month visit. Table 2 also shows levels of typical weekly drinking and drinking-related consequences at each of the study assessments.

Table 2.

Levels of alcohol use and consequences and sexual assault severity across study assessments.

	Baseline	12-months	24-months	36-months
Typical drinks per week, mean (SD)	8.2 (10.8)	6.6 (11.6)	6.6 (12.3)	6.2 (10.5)
Alcohol-related consequences, mean (SD)	8.0 (9.3)	6.1 (9.2)	5.4 (8.4)	4.8 (8.6)
Sexual assault severitya, %
None reported	47.5	71.8	74.9	80.0
Moderate	19.9	15.5	11.1	8.4
Severe	32.6	12.7	14.0	11.7

^aFor baseline prevalence, this measure was asked in relation to the period between age 18 and the baseline assessment. Subsequent follow-up measures asked about the previous 12 months.

Table 3 presents results from the first set of models that examined the effects of prior year sexual assault severity on typical level of drinking. According to findings from the marginal structural model, past year severe sexual assault was associated with a 71% higher count of typical drinks per week compared to no sexual assault (CR = 1.71; 95% confidence interval [CI]: 1.27, 2.31). However, when comparing moderate to no sexual assault, there was no statistically significant association (CR = 1.13; 95% CI: 0.85, 1.52). The linear trend as assessed when modeling the SES as a single ordinal variable was statistically significant (p = .023). Notably, when using a “traditional” unweighted GEE model that adjusted only for the time-fixed covariates, the magnitude of associations was stronger compared to the weighted model. This suggests that effect sizes from the traditional model may be upwardly biased due to time-varying confounders not accounted for in the model.

Table 3.

Count ratios for typical drinks per week according to sexual assault severity and other time-fixed covariates from marginal structural and unweighted models

Characteristic	Weighted		Unweighted
	CR	95% CI	CR	95% CI
Past year sexual assault severity
None reported (reference)
Moderate	1.13	0.85, 1.52	1.26	0.95, 1.68
Severe	1.71	1.27, 2.31	2.23	1.65, 3.01
Baseline Age	1.03	0.98, 1.09	1.04	0.99, 1.09
Parent’s education
Low (reference)
Moderate	1.07	0.80, 1.43	1.02	0.76, 1.36
High	1.20	0.90, 1.59	1.21	0.93, 1.58
Race
White (reference)
African American	1.40	1.02, 1.92	1.40	1.04, 1.90
Other	1.31	0.90, 1.92	1.32	0.93, 1.88
Sexual assault prior to age 18
None reported (reference)
Child (<13 years)	1.22	0.91, 1.64	1.17	0.89, 1.54
Adolescent only (13–17 years)	0.98	0.72, 1.34	1.01	0.75, 1.35
Bisexual identity	0.84	0.68, 1.05	0.81	0.66, 1.00

Note. CR = count ratio. CI = confidence interval.

Sexual assault severity also appeared to have effects on alcohol-related consequences (Table 4). Similar to findings of typical drinking, past year severe sexual assault compared to no sexual assault was associated with a 63% higher level of alcohol consequences at the subsequent assessment (CR = 1.63; 95% CI: 1.21, 2.20). Although the association was not as strong, moderately severe compared to no sexual assault was also related to higher levels of alcohol-related consequences the following year (CR = 1.42; 95% CI: 1.10, 1.84). Further, the linear trend was statistically significant (p = .004). Again, non-weighted effect estimates were stronger than the weighted estimates.

Table 4.

Count ratios for alcohol related consequences according to earlier sexual assault severity and other time-fixed covariates from marginal structural and unweighted models

Characteristic	Weighted		Unweighted
	CR	95% CI	CR	95% CI
Past year sexual assault severity
None reported (reference)	–	–	–	–
Moderate	1.42	1.10, 1.84	1.84	1.61, 2.09
Severe	1.63	1.21, 2.20	2.61	2.25, 3.02
Baseline age	1.04	0.98, 1.09	1.05	1.01, 1.08
Parent’s education
Low (reference)	–	–	–	–
Moderate	1.02	0.73, 1.41	0.92	0.75, 1.13
High	1.26	0.96, 1.67	1.11	0.93, 1.33
Race
White (reference)	–	–	–	–
African American	1.11	0.80, 1.55	1.04	0.84, 1.28
Other	1.13	0.77, 1.66	1.08	0.84, 1.39
Sexual assault prior to age 18
None reported (reference)	–	–
Child (<13 years)	1.35	1.03, 1.77	1.19	1.00, 1.42
Adolescent only (13–17 years)	1.00	0.75, 1.34	1.04	0.86, 1.26
Bisexual identity	1.11	0.88, 1.40	1.01	0.88, 1.16

Note. CR = count ratio. CI = confidence interval.

Effects of cumulative sexual assault severity over two years on alcohol outcomes at the 36-month follow-up assessment were also examined (Table 5). Examining typical drinking as the outcome, findings from the marginal structural model showed that a one-unit increase in two-year cumulative sexual assault severity categorical score was associated with a 27% increase in count of drinks per week at the final follow-up visit (CR = 1.27; 95% CI: 1.14, 1.42). There was also a strong association between two-year cumulative sexual assault severity and drinking related consequences (CR = 1.27; 95% CI: 1.12, 1.43). Similar to analyses examining the one-year lagged effects, findings showed stronger associations for both typical drinking and drinking consequences when using unweighted models. Parameters for other covariates (not shown) were similar to those from corresponding models shown in Tables 4 and 5. For the lagged and cumulative models, there was no evidence of moderation of sexual assault severity effects by sexual orientation for either alcohol outcome.

Table 5.

Count ratios^a for typical drinks per week and alcohol-related consequences at 36-month follow-up according to two-year cumulative sexual assault severity based on reports from 12- and 24-month follow-up

Alcohol outcome	Weighted		Unweighted
	CR	95% CI	CR	95% CI
Typical drinks per week	1.27	1.14, 1.42	1.34	1.22, 1.47
Alcohol-related consequences	1.27	1.12, 1.43	1.33	1.19, 1.49

^aCovariates in regression models included baseline age, parent’s education, race, history of childhood sexual assault, and sexual orientation

Discussion

Results from marginal structural models indicated that there were deleterious effects of sexual assault on both typical drinking and alcohol-related consequences the following year. These effects were stronger with increasing levels of sexual assault severity. Further, there also appeared to be a cumulative effect of sexual assault on alcohol use where cumulative sexual assault severity over a two-year period also predicted greater levels of drinking and more alcohol-related consequences. These findings point to the importance of prevention programming for SMW to decrease prevalence of sexual assault victimization and re-victimization.

Consistent with other studies of prevalence of sexual assault exposure in SMW across the lifespan, we found high occurrence of exposure among young SMW (Balsam, Rothblum, & Beauchaine, 2005; D’Augelli, Pilkington, & Hershberger, 2002; Hughes et al., 2001; Hughes et al., 2010). In a review of 75 studies that included SMW, the median estimate of lifetime sexual assault was 43% for SMW (Rothman et al., 2011). In our sample, the prevalence of lifetime sexual assault was even higher – over one half experienced an assault as an adult and roughly one third experienced an assault in childhood or adolescence. This study highlights the important role of adult sexual victimization in understanding potential risk for alcohol use and consequences, even after accounting for multiple putative confounders. Sexual assault, as a particular risk for alcohol use among young SMW, has only relatively recently been an area of focus, while much of the earlier research has focused on child sexual victimization.

The minority stress model, where stress mediates relationships between a stigmatized sexual identity and, in this case, alcohol use and consequences (Meyer, 2003; Talley et al., 2016), has some limitations in that it does not incorporate potential psychological mediators between stress and health outcomes (Hatzenbeuhler, 2009). It also fails to incorporate moderators such as gender. The focus specifically on stressors that occur because of sexual orientation may fail to adequately consider issues of intersectionality, such as victimization that may be multiply determined by gender and by sexual orientation. It is also possible that there are cumulative effects of sexual victimization and discrimination among already marginalized and at risk populations where high intensity stressors such as moderate to severe repeated sexual victimization may have a particularly deleterious effect. More recent theoretical models address potential mechanisms of action for the effects of minority stress on health outcomes including coping and emotion regulation skills, social factors, and maladaptive cognitions (Hatzenbeuhler, 2009). Although this study was not situated to test these potential mechanisms, sexual victimization can lead to higher endorsement of coping motives for drinking, increased tension reduction alcohol expectancies, increased emotion dysregulation, and more negative cognitions about self and others, and higher drinking norms, all of which may help explain increased alcohol use and consequences (Bedard-Gilligan, Lee & Kaysen, 2011; Gilmore et al., 2014; Stappenbeck, Bedard-Gilligan, Lee & Kaysen, 2013; Ullman, Filipas, Townsend & Starzynski, 2005). There may be an interaction between victimization and discrimination such that effects of these factors are stronger for women who have experienced both. Similarly, the appraisal of the intention of the assault (e.g., related to one’s sexual identity) may also amplify effects of the assault on alcohol outcomes. Future research should examine the explanatory and moderating roles of these factors among SMW.

The study findings have important public health and clinical implications. The widespread occurrence of sexual assault, both at baseline and across the three follow-up assessments, speaks to the vulnerability of this population and to the need for targeted services aimed at risk reduction. Findings highlight the need for targeted strategies to prevent sexual assault among SMW. Further, although sexual assault is a major public health problem in and of itself, implementation of effective sexual assault prevention strategies could also have substantial effects on reducing the burden of alcohol misuse in this population. As has been frequently noted, SMW are an important health disparities population and initiatives such as Healthy People 2020 (U.S. Department of Health and Human Services, 2016) have called for the reduction of sexual-orientation-related disparities across a range of health and behavioral outcomes. Based on the counterfactual framework and marginal structural model results, in this sample the average number of typical drinks per week among women who reported a severe sexual assault in the prior year would have declined from 10.3 to 6.0 and the average number of past-month alcohol-related consequences would have declined from 5.0 to 3.1 had these women actually not experienced any sexual assault. Further, although the difference would be more modest, had they not experienced any sexual assault, women experiencing a moderate sexual assault would have shown a decline from 4.4 to 3.1. In light of the magnitude of these potential reductions in drinking and alcohol-related consequences as well as the elevated prevalence of sexual assault among SMW, the prevention of sexual assault could yield important reductions in the disparity between SMW and heterosexual women in alcohol misuse.

Further, although the blame for victimization lies firmly on the perpetrator, there is clinical utility in implementing effective risk reduction programs to empower potential victims and decrease incidence of victimization among high-risk groups such as SMW. Future studies should seek to explore other factors that may play a role in understanding the relationship between sexual assault and drinking behavior, such as relationship with the perpetrator, the development of posttraumatic stress disorder, and adaptive and maladaptive coping strategies in the aftermath of an assault, to best understand how to improve prevention and intervention efforts for SMW. SMW are at unique increased risk for adverse societal and cultural experiences, such as discrimination and microaggressions, in addition to sexual assault. Thus, better understanding of the ways in which these factors interact and the strategies that can promote recovery and reduce risky behavior, such as drinking, for this population is crucial.

A major strength of this study is its use of marginal structural models to account for time-varying confounding. Using this approach, we were able to isolate and estimate the specific causal pathway from sexual assault to alcohol use adequately accounting for other time-varying factors that can often co-occur with both. The utility of this approach is highlighted when comparing results from the weighted and unweighted models. Results from the unweighted model were consistently stronger than those from the weighted model. This suggests that there are important confounders that were not adequately accounted for in the traditional model. Application of the marginal structural modeling approach in other longitudinal investigations of effects of risk factors on alcohol outcomes could prove beneficial in establishing more accurate effect sizes.

Despite these notable strengths there are limitations that should be considered when evaluating the study outcomes. First, this sample was recruited online and there were differences in racial composition between those who ultimately consented and those who declined participation. It is therefore possible that the sample may not be representative of the general population of young adult SMW in the United States. However, regarding the online sampling, prior studies have shown that online recruitment methods can be reflective of intended populations of interest (Harris, Loxton, Wigginton, & Lucke, 2015). Also, obtaining large samples of hard-to-reach minority populations using traditional sampling methods such as random digit dialing may not be feasible. Finally, as highlighted by recent discussions in epidemiology, although representativeness is necessary for descriptive epidemiologic studies that are intended to report on the health status of a population, it may not be as relevant for studies that are intended to understand causal mechanisms and that have appropriate adjustment for potential confounding (Rothman, Gallacher, & Hatch, 2013). Sexual orientation can change for women over time, yet sexual orientation analytically was treated as a time-fixed covariate. Indeed there was some evidence of sexual orientation change over follow-up. However, the prevalence of change at any follow-up wave was low (<8%). Thus, including this as a time-varying covariate in estimation of IPWs could lead to extreme weights for some participants. Descriptive statistics also suggest that change in orientation showed no significant association with sexual assault at the following wave. Thus, we would not expect changing sexual orientation to bias our findings in any appreciable manner. Another limitation is the use of self-report measures of alcohol use and consequences which may underestimate true levels of consumption and consequences. However, the DDQ and YAACQ, measures used for this study, have been validated against other criterion standards in multiple studies. Further, because of the longitudinal nature of the study, there is a lower likelihood of differential reporting of alcohol use and consequences due to sexual assault history.

To summarize, using a marginal structural modeling approach that accounts for both time-fixed and time-varying confounders, this study found effects of sexual assault on increasing levels of typical drinking and alcohol-related consequences one year later among SMW. Further, two-year accumulation of sexual assault exposure also appeared to have effects on increasing alcohol use and alcohol-related consequences. This evidence highlights the health consequences of sexual assault and the need to identify effective prevention efforts to reduce risk of sexual assault and its long-term sequelae, especially among SMW. As research into this area progresses, understanding mechanisms through which sexual assault leads to increased drinking is necessary to provide clearer targets of intervention. Further, studies that include both sexual minority and heterosexual women may be informative to compare effects between the two groups. Overall, SMW are a highly vulnerable group for both sexual assault victimization and substance use and clearly there is a need for additional research to understand both shared and unique factors between SMW and heterosexual women that contribute to the increased risk.