A latent class analysis of alcohol-related problems among adults who drank in the past year

Abstract

Background:

Research on alcohol-related problems often examines individual problem types in isolation or uses scales that provide a single cumulative severity score for alcohol-related harms. This study aims to assess the patterns of seventeen distinct alcohol-related problems and how they co-occur.

Methods:

The East Bay Neighborhood Study surveyed a community sample of 864 adults who drank in the past year in Alameda County, California. Participants reported if they experienced each of seventeen alcohol-related problems in the last year. Latent class analysis assessed subgroups of problems. Logistic regression models examined associations between class membership, sociodemographics, and alcohol use.

Results:

A two-class model best fit the data. The multiple problems class (18% of respondents) was characterized by experiencing problems of all types and almost all experiences of legal, violence, and risky sex-related problems. The none/few problems class (82%) was characterized by a low prevalence of all problem types, with only a small proportion experiencing hangovers. In adjusted models, only older age (AOR=0.90, 95% CI=0.88–0.92) had lower odds of multiple problems class membership.

Conclusions:

Numerous alcohol-related problems co-occurred within a small subgroup of people who drank in the last year, while the majority experienced few problems. Results suggest that focusing on singular alcohol-related problems may overlook patterns of concurrent problems in high-risk groups.

Introduction

Heavy alcohol use (defined as 15+ and 8+ drinks per week for men and women, respectively) and binge drinking (defined as 5+ and 4+ drinks on a single drinking occasion for men and women, respectively) are associated with a variety of physical, psychological, social, and community problems (Cerdá et al., 2016; Guirguis et al., 2015; Moss, 2013; Snowden, 2015). These alcohol-related problems can have deleterious and long-term consequences for the individual and result in undue social and economic burden for the community. In the United States, excessive alcohol use and related problems are associated with an annual cost of $249 billion (Sacks et al., 2015) and approximately five million emergency department visits per year (White et al., 2018). Thus, control policies and interventions often aim to reduce alcohol consumption and prevent associated harms.

Traditionally, alcohol-related problems are divided by problem type and commonly addressed in isolation. For instance, associations of alcohol consumption with specific types of problems, such as physiological and health problems (e.g., hangovers, liver inflammation, cardiovascular disease), social problems (e.g., interpersonal relationships, violence), sexual risk-taking, impaired driving, and crime have all been explored separately in research and intervention development (Abbey et al., 2014; Devries et al., 2014; Guirguis et al., 2015; Jayasekara et al., 2014; Martin et al., 2013; Scott-Sheldon et al., 2016). The DSM criteria for Alcohol Use Disorder (AUD) and standardized screening measures, such as the Alcohol Use Disorders Identification Test (AUDIT), have also been used to measure a range of problems and provide an overall severity score. However, these scales do not examine or reflect patterns of individual types of problems and whether specific problems tend to be experienced together (Devos-Comby & Lange, 2008). Overall, a dearth of research has examined whether specific types of alcohol-related problems co-occur in individuals (Skogen et al., 2019; Stiles & Rice, 2019).

Within a community sample of adults who drank in the last year, this study addresses these research gaps by using Latent Class Analysis (LCA) to describe patterns of seventeen alcohol-related problems and assess how problems co-occur within individuals. We hypothesize that distinct subgroups will be identified by types of alcohol-related problems. The person-centered approach of LCA is well-suited to test this hypothesis and examine homogenous subgroups of substance use and related harms (Masyn, 2013; Oberski, 2016). As it is unlikely alcohol problems occur in isolation, this will provide a better understanding of the prevalence and nature of co-occurring problems experienced in the community. By identifying how problems co-occur, we can better inform intervention and prevention efforts to be comprehensive and address the range of ongoing problems that occur among adults who drink. We then assess the associations of alcohol-related problem class membership with sociodemographics and alcohol use. Based on literature examining individual problem types (Cerdá et al., 2016; Hughes et al., 2016; Karriker-Jaffe et al., 2013; Sumetsky et al., 2022), we hypothesize that membership in class(es) experiencing more alcohol-related problems will be associated with being male, being younger, having a lower income, and heavier drinking.

Material and Methods

The present study used data from the East Bay Neighborhoods Study. Using point-based sampling methods, 72 microecological neighborhoods from Alameda County, California were chosen for recruitment, representing high and low median household income and off-premise alcohol outlet densities. In total, 1,124 respondents completed the survey with an average of 15.6 (Range: 7–26) respondents per neighborhood. Details regarding the sampling design and recruitment are in a previous publication (Mair et al., 2020). Study protocols were approved by the Institutional Review Board at the University of Pittsburgh.

Consenting participants were given the option to complete the survey in English or Spanish via phone or internet. The current analysis was restricted to 864 people who drank in the past year. Analyses examining correlates with alcohol-related problems were restricted to 774 participants for whom complete data on all covariates were available. Ninety participants were excluded due to missing data on a combination of the following variables: gender (n=15), race (n=43), sexual orientation (n=28), marital status (n=12), education (n=7), and drinking frequency and volume (n=8).

Measures

Respondents indicated if they had ever experienced each of 17 alcohol-related problems within 24 hours of having any alcoholic beverage in the past year (response: yes/no). This measure was broadly based on Wechsler et al.’s (1994) list of problems and behavioral consequences after binge drinking. As this list was originally developed for college students, items were adapted for a broader community, adult population (e.g., ‘trouble with campus police’ was changed to ‘trouble with police’). Problems included: physiological problems (experiencing hangovers the next day, passing out, forgetting where one was/what one was doing, becoming nauseated or vomiting, needing medical treatment because of drinking, being hurt or injured), social problems (arguing with a friend/relative, being criticized by someone because of drinking, missing or being late for school/work), legal or violence-related problems (getting in trouble with the police, getting in a physical fight, damaging one’s own or someone else’s property), driving after having too much to drink, and risky-sex-related problems (having sex when did not plan to, having sex when did not want to, being pressured or forced to have sex, and pressuring or forcing someone else to have sex).

Drinking patterns were estimated using the average quantity consumed per drinking occasion and the volume of alcohol consumed per 28 days. Drinking frequency was measured over the past year, with those indicating at least monthly use considered to have past 28-day use. Frequency was then rescaled to indicate the number of days a respondent drank alcohol within the last 28 days. Continued drinking volume, representing the number of drinks beyond one per drinking occasion, was calculated as 28-day drinking volume minus drinking frequency (Gruenewald, Johnson, Light, Lipton, et al., 2003; Gruenewald, Johnson, Light, & Saltz, 2003). The 10-item Alcohol Use Disorders Identification Test (AUDIT) screened participants for unhealthy drinking in the past 12-months by providing a score from 0 to 40, categorized into three groups (Saunders et al., 1993). A score of 8+ indicates hazardous drinking, while 13+ (women) and 15+ (men) indicate likely alcohol dependence. Scores lower than 8 indicate low-risk consumption.

Demographic characteristics assessed included age (continuous), gender (male, female), race/ethnicity (Asian, Black, Hispanic/Latinx, white), sexual orientation (heterosexual/straight, sexual minority), marital status (married or living together, single or divorced/separated/widowed), education (no college degree, college or graduate degree), and annual income (less than $20,000, $20,000 to $60,000, $60,000 to $100,000, more than $100,000). An additional indicator variable was included for missing income data (n=49).

Analysis

LCA was conducted to characterize patterns of alcohol-related problems experienced by people who drank in the past year. Using Mplus, classes were fit using the seventeen alcohol-related problems. Model enumeration followed a stepwise approach, in which one to six class models were fit. Model fit statistics, including the Akaike Information Criterion (AIC), Bayesian Information Criterion (BIC), sample size-adjusted Bayesian Information Criterion (SSA-BIC), Bayes factor (BF), Correct model probability (cmP), and Approximate Weight of Evidence Criterion (AWE), were used to select the best fitting class structure (Kass & Raftery, 1995; Masyn, 2013; Nylund et al., 2007). All models accounted for the clustering of people within neighborhoods (recruitment sites) using the Complex analysis procedure (Muthén & Muthén, 2017).

After selection of the final class structure, descriptive statistics were used to characterize each class. Logistic regression models were fit using the R3Step method in Mplus to account for uncertainty in class assignment (Muthén & Muthén, 2017). Models examined how sociodemographics and alcohol use are associated with class membership. The first model adjusted for demographics and neighborhood types and a second model additionally accounted for 28-day continued drinking volume and frequency. In supplementary bivariate analyses, individuals were assigned to their most likely class based on posterior probabilities, and the demographic composition of classes was compared.

Results

Among the 864 participants who drank in the last year, the average age was 52.6 (SD=16.6, Range=21–87). Fifty-eight percent were female, 8.9% identified as gay/lesbian or bisexual, and 51.6% were married or living with a partner. The majority of the sample identified as non-Hispanic White (66.4%), 17.3% non-Hispanic Black, 9.3% Hispanic/Latino, and 7.1% non-Hispanic Asian. Three-quarters of the sample had a college degree. The sample had an average of 8.8 (SD=8.7) drinking days in the last 28 days, and 8.4% were considered to likely have hazardous drinking or alcohol dependence by the AUDIT. Other sample characteristics can be found in Supplemental Table 3.

In the last year, 366 (42.4%) participants experienced at least one alcohol-related problem. Among those who experienced at least one of the 17 assessed problems, there was an average of 2.2 experienced problems (SD 1.8, Median: 2, Range: 1–14; data not shown). Hangovers were experienced by the largest proportion of the sample (34.2%), followed by arguing with a friend or relative (13.1%) and becoming nauseated or vomiting (13.0%). The prevalence of each of the 17 alcohol-related problems experienced by the sample is presented in Supplemental Table 1.

Fit statistics indicated that the two-class model best fit the data (Supplemental Table 2). The multiple problems class (class 1) represented 18.4% of the sample and experienced an average of 3.6 problem types (SD 1.9) in the last year. This class was characterized by experiencing multiple physiological (e.g., hangover, passing out) and social (e.g., arguing with a friend/relative, missing work/school) problems (Figure 1). For instance, 91.4% of this class experienced a hangover the day after drinking, 63.5% nausea or vomiting, 21.5% passing out, and 19.1% forgetting where they were or what they were doing after drinking. This class also included almost all participants who experienced any violence, legal, and risky sex-related problems, driving after drinking, and being hurt/injured. The none/few problems class (class 2) represented 81.6% of the sample and experienced an average of 0.4 problems (SD 0.6) in the last year. This class was characterized by a low number of problems experienced overall and a majority who experienced no problems (70.5%). The remaining minority of this class primarily experienced hangovers (21.4%) and/or arguing with friends or relatives (6.1%). In supplementary bivariate analyses, the multiple problems class had higher drinking frequency (10.2 vs. 8.7 per 28 days) and continued volume (14.8 vs. 6.1 drinks per 29 days) than the none/few problems class and was more likely to be classified by the AUDIT as having hazardous (23.7% vs. 4.4%) or dependent (5.8% vs. 0%) drinking (Supplemental Table 3).

Figure 1.

Alcohol-related Problems of the Multiple Problems (n=159, 18.4%) and None/Few Problems (n=705, 81.6%) Latent Classes

In multivariable model A, age, gender, and being a sexual minority were significantly associated with multiple problems class membership (Table 1). Females had half the odds compared to males (AOR=0.49, 95% CI=0.31–0.79) and those identifying as gay, lesbian, or bisexual had more than two times the odds compared to heterosexuals (AOR=2.37, 95% CI=1.16–4.83) of being in the multiple problems class vs. no/few problems. However, when additionally adjusting for drinking frequency and continued volume in model B, the associations between class membership and both gender (AOR=0.72, 95% CI=0.41–1.29) and sexual minority status (AOR=2.44, 95% CI=0.91–6.55) were no longer statistically significant. Moreover, neither continued drinking volume (AOR=1.07, 95% CI=0.98–1.16) nor frequency (AOR=1.02, 95% CI=0.95–1.09) were significantly associated with multiple problems class membership, and only age remained independently associated (AOR=0.90, 95% CI=0.88–0.92).

Table 1.

Adjusted Odds Ratios of being in Multiple Alcohol-related Problems Latent Class (n=774)¹

	Model A.	Model B.
	Membership in the Multiple Problems Class versus No/Few Problems Class AOR (95% CI)	Membership in the Multiple Problems Class versus No/Few Problems Class AOR (95% CI)
Demographics
Age (years)	0.92 (0.90, 0.93)*	0.90 (0.88, 0.92)*
Gender (Ref: Male)
Female	0.49 (0.31, 0.79)*	0.72 (0.41, 1.29)
Race/Ethnicity (Ref: White)
Asian	0.71 (0.37, 1.36)	.69 (0.35, 1.39)
Black	1.01 (0.57, 1.77)	.92 (0.46, 1.86)
Hispanic/Latinx	0.61 (0.26, 1.43)	.75 (0.34, 1.62)
Sexual Identity (Ref: Heterosexual)
Gay, Lesbian or Bisexual	2.37 (1.16, 4.83)*	2.44 (0.91, 6.55)
Marital Status (Ref: Single)
Married or Living Together	0.76 (0.52, 1.11)	0.86 (0.49, 1.50)
Education (Ref: Less than College)
College or Graduate Degree	1.12 (0.66, 1.90)	1.06 (0.60, 1.85)
Income (Ref: Less than $20,000)
$20,000 to $60,000	1.45 (0.65, 3.27)	1.53 (0.63, 3.71)
$60,000 to $100,000	1.39 (0.53, 3.70)	1.43 (0.54, 3.76)
More than $100,000	1.10 (0.43, 2.83)	0.84 (0.30, 2.37)
Missing Income Variable	0.82 (0.29, 2.33)	0.65 (0.22, 1.95)
Neighborhood Site Type
High Income Neighborhood	0.67 (0.40, 1.13)	0.72 (0.37, 1.39)
High Outlet Density Neighborhood	0.91 (0.56, 1.47)	0.95 (0.51, 1.78)
Alcohol Use (per 28 days)
Continued Volume		1.07 (0.98, 1.16)
Drinking Frequency		1.02 (0.95, 1.09)

¹All models account for recruitment site clusters.

^*p<0.05

Discussion

Among adults who drank in the last year, we identified two distinct classes of alcohol-related problems. It is noteworthy that only two classes were identified among the sample, generally representing a small subgroup that experiences several problems and a large majority of people who drink who experience none or few problems. The none/few problems class, consisting of over 80% of the sample, experienced a limited number of problems and a large portion experienced no problems at all. Almost all of the problems experienced by this class were hangovers, with a smaller proportion reporting arguing with a friend or relative after drinking. Some research has referred to these as “common” problems (Courtney et al., 2018; Stiles & Rice, 2019).

In comparison, the multiple problems class experienced numerous problems, as well as problems generally considered more severe, including driving after drinking, risky sexual behaviors, violence, and legal problems. Unlike the none/few problems class, this class also experienced multiple forms of physiological and social problems, such as passing out, needing medical treatment for drinking, and missing school or work. These results indicate that a multitude of problems occur within the same subgroup, and if an individual experiences physiological problems other than hangovers, they may be at risk for all other problem types. Thus, focusing on specific types of problems in isolation may overlook larger patterns of co-occurring alcohol-related problems in high-risk groups.

Our findings are however unclear in regard to the relationship between alcohol use and alcohol-related problems. While the multiple problems class was associated with heavier drinking and frequency in the bivariate supplemental analysis, these associations did not remain after adjusting for sociodemographics and the study design, though the direction of association remained for continued volumes. Moreover, 28 individuals identified by the AUDIT with hazardous drinking in the last year had the highest probability of being in the none/few problems class, and a sizable proportion of the multiple problems class (70.5%) were considered to have low-risk consumption. While it is possible that the 28 people considered to have hazardous drinking in the low/few problems may underreport or be in denial about the consequences of their drinking, it is clear that our problem classes are not simply representations of harmful or unhealthy drinking. Instead, they highlight the unique epidemiology of alcohol-related problems versus alcohol use. These problems, for instance, may occur after a single night of drinking and do not necessarily indicate a pattern of frequent or heavy drinking, as is likely the case among those considered to have low risk drinking in the multiple problems class. This requires further investigation into the pathway between alcohol use and problems as well as the identification of modifiable mediators and contexts that can be targeted to prevent problems among those who drink at any level of use.

We also found that heavier drinking attenuated the association between problems and gender. Although females were significantly less likely to be in the multiple problems class than males, this gender differential was not evident after adjusting for alcohol use, suggesting that alcohol use may contribute to the relationship between problems and gender. Along with evidence that the gender-gap for alcohol use is decreasing (Slade et al., 2016), our findings underscore the need to understand gender-specific risks and contexts for alcohol problems and investigate the causal pathway between gender, consumption, and problems. Comparatively, while the association between class membership and sexual minority status was no longer statistically significant after adjusting for alcohol use, the effect size remained similar. While sexual minorities in the sample did not have substantially different drinking patterns than heterosexual individuals (e.g., 8.76 vs. 8.84 drinking days per 28-days; data not shown), this disparity in alcohol problems require further investigation.

Moreover, older age was negatively associated with the odds of multiple problems class membership regardless of consumption, indicating that younger individuals are at increased risk of multiple and concurrent alcohol-related problems. However, this sample skews slightly older (mean 52.9) and it is possible this may account for the age association found in this community sample of largely middle-aged and older adults. As some problems may be more prevalent in college-aged populations, further research is needed to examine and compare our classes of co-occurring problems in samples of specifically middle-age and college populations.

In interpreting these findings, the study’s limitations should be considered. Alcohol-related problems were retrospectively self-reported for the past year, and thus may be suspectable to recall bias. Problems were also assessed as binary variables, indicating if each problem was experienced ever in the previous year, and did not measure the frequency of each problem occurring. As mentioned, this sample is older and is highly educated, and it is possible this may skew study results. Lastly, the sample was recruited in the California East Bay, in which there are distinct alcohol-related policies and alcohol availability policies. Alcohol-related problems in this sample may not resemble those in more restrictive alcohol environments and should be considered with caution in generalizing these results.

Overall, we found two distinct patterns of alcohol-related problems among people who drank in the past year. The multiple problems class represents a high-risk group in the community that experience a disproportionate burden of alcohol-related problems. Prevention and harm reduction efforts targeting this group should consider that alcohol-related problems often occur together, and interventions may not be as effective by addressing problems in isolation. Identifying patterns of alcohol-related harms and those at risk can enable the development and improvement of interventions and policies to better address consequences of alcohol use.