A Multilevel Model of Alcohol Outlet Density, Individual Characteristics and Alcohol-Related Injury in Argentinean Young Adults

Abstract

Objectives.

Previous research from high-income countries has consistently shown an association between alcohol-related harms and neighborhood characteristics such as alcohol outlet density, but this research has not been extended to middle- and low-income countries. We assessed the role of neighborhood characteristics such as alcohol outlet density, overcrowding and crime rates, and individual characteristics including gender, age, alcohol and marijuana use, and geographic mobility associated with alcohol-related injuries in university students in Argentina.

Methods.

Data were collected from a randomized sample of students attending a national public university (n = 1346). Descriptive, bivariable, and multilevel logistic regression analyses were performed.

Results.

In the final model, on-premises alcohol outlet density—but not off-premises outlet density, overcrowding or crime—was associated with past-year and lifetime alcohol-related injury (median odds ratio=1.16). At the individual level, quantity (odds ratio (OR)=1.05, 95% CI=(1.01, 1.10)) and frequency (OR=1.66, 95% CI=(1.41,1.97)) of alcohol consumption and age (OR=0.81, 95% CI=(0.74, 0.88)) were associated with past-year and lifetime alcohol-related injury.

Conclusions.

This study contributes to an area with a paucity of information from non-high-income countries, finding differences with previous literature.

INTRODUCTION

Fatal and non-fatal injuries are among the most serious consequences of alcohol drinking in the Americas. Non-fatal injuries account for more years lived with disability than non-communicable diseases, and fatal injuries account for the highest percentage of alcohol-attributable deaths.¹ Countries in the Southern Cone of South America have one of the highest alcohol consumption levels worldwide, with Argentina, Chile, and Uruguay having low abstention rates and high heavy episodic drinking prevalence among young people.¹ These countries also face a higher rate of harm per liter (eg, more alcohol related injuries at a given consumption level) compared to high-income countries.² While in Argentina the number of drinkers is not much higher than the United States,¹ among alcohol users, Argentinean drinkers are 40% more likely to experience an alcohol-related injury compared to their United States or Canadian counterparts.²

An effective strategy to reduce alcohol consumption and related problems is limiting the physical availability of alcohol. Increased alcohol outlet density has been associated with lower prices of alcoholic beverages because of increased competition and increased positive perceptions of alcohol use.³ In combination with other factors, a higher alcohol outlet density is accompanied by a raise in alcohol-related problems^4–6 and alcohol-related harms (eg, crime, violence, motor vehicle crashes).^7–12 Longitudinal studies have shown an increase in alcohol consumption in youth as the number of alcohol outlets increases,¹³ and drinking among college students (an at-risk population for alcohol related harms) and older adolescents increases with closer proximity to alcohol outlets.^14,15 Furthermore, a recent meta-analysis¹⁶ showed that decreasing the physical availability of take away alcohol results in a reduction of per capita consumption. Despite poorer countries exhibiting higher alcohol related harms, evidence on the association of alcohol outlet density and related harms comes almost exclusively from high-income countries.^16–18

Additionally, neighborhood characteristics other than alcohol outlet density that act as environmental stressors and prevalent in middle-income countries might influence drinking or exacerbate related harms. Among these are poor living conditions such as overcrowding, and a higher likelihood of social conflict, often in the form of aggression or crime.^19–21 Crime weakens community ties and reduced social control could raise deviant behavior; likewise, drinking could be a way of coping with hostile living conditions, particularly were opportunities for education and employment are scarce.³

Despite the wealth of evidence regarding the role of alcohol outlet density on risk for alcohol-related harms in high income countries, results have not always been unanimous,²² nor exempt from criticism.¹⁷ Among concerns is the neglect of possible confounders, such as co-occurring drugs use, particularly marijuana use which may increase potential harms from both substances due to a greater impairment in cognitive and psychomotor abilities.^23,24

In sum, despite high consumption levels and high rates of alcohol-related injuries, there is a paucity of evidence regarding the association of alcohol outlets and alcohol related injuries in middle-income countries; this relationship has not taken into account neighborhood characteristics or individual confounders such as marijuana use. Furthermore, there are important cultural and policy differences in the way young adults in Argentina acquire and consume alcohol. In Argentina, alcohol drinking is widely accepted. Alcoholic beverages, including beer, wine, and hard spirits, are commonly purchased along with other items at groceries stores-compared to the United States, where many states do not allow the sale of alcohol in food stores. Specialized stores such as liquor stores account for less than 2% of the market share,²⁵ and policy measures to control density and location of alcohol outlets are scarce.²⁶

Therefore, the purpose of our study was to assess the association of alcohol outlet density on alcohol-related injuries in Argentina. We assessed the relationship between density of on-premises and off-premises alcohol outlets and alcohol-related injuries in university students in Argentina, with the addition of other neighborhood and individual characteristics that may help to understand the relationship between alcohol outlets and related harms. A synthesis of the conceptual framework of this study is provided in the Supplementary Fig. S1, http://links.lww.com/CJA/A10.

METHOD

Participants and study procedure

We collected data from a probabilistic sample of undergraduate students in all departments of the National University of Mar del Plata in Argentina from April to November 2014. The National University of Mar del Plata in Argentina is a large, public university serving almost 24,000 undergraduate and graduate students. As in other Argentinean public universities, access is unrestricted and free of charge, ensuring a wide sociodemographic composition of students. We randomly selected one course of each year in each department and asked the professors to allow data collection, with no refusals. Within the selected courses, criteria for the selection of participants were as follows: current students who had at least one drink in the past year, under age 35 years, and had permanent residence in Mar del Plata. Of the eligible sample (n=1512), 89% participated (n=1346), 6% (n=90) declined to participate, 4% (n=61) left the class, and 1% (n=15) provided mostly incomplete questionnaires.

Data collection took place in classrooms during class hours with the professors’ consent. Time of administration was about 20 minutes, and trained researchers were present while questionnaires were completed in order to answer questions. Students were supplied with informed consent documents detailing the study’s purpose and outlined procedures for assuring data confidentiality; participation in the study did not affect their class grade, and no compensation was offered. The distribution of genders among the sample was similar to the university’s records for the same year.²⁷ This study was approved by the Ethics Committee of the National Institute of Epidemiology in Argentina.

Measures

Alcohol Outlets, Overcrowding, and Crime Rates

Of the 2309 licensed alcohol outlets in the city of Mar del Plata in 2014 (total population=614,350), 1872 were classified as on-premises according to their licensing registration (restaurants=32% and bars=51%), and 437 as off-premises (wine, beer, and liquor stores=11%, super and minimarkets=4%). According to their addresses, they were distributed across Mar del Plata’s 57 neighborhoods (a neighborhood is a municipal division based on distribution of civil services). We estimated the density of alcohol outlets per 1000 inhabitants for each neighborhood. We also calculated the overcrowding and crime rate for each neighborhood. Overcrowding is a measure of housing inadequacy and an indicator of poverty.²⁸ Although it is usually calculated by dividing the number of people living in a home by the number of rooms available, the standard for considering a home overcrowded varies across regions or countries.²⁹ While in the United States, a home may be overcrowded if there is more than 1 person per room,²⁶ in Latin America overcrowded is usually more than 3 per room.³⁰ Here, the overcrowding rate was obtained by dividing the number of inhabitants in each neighborhood by the number of housing units; a higher overcrowding rate reflects poorer household conditions. We also calculated the rate of crime (threats, violent robberies, nonviolent robberies, attempted robbery, illegal gun holding, sexual assault, and homicides) reported per 1000 inhabitants for each neighborhood, according to official police and security camera surveillance reports.³¹

Each study participant was asked to name the neighborhood they lived in, and this information was linked to each neighborhood’s area-level variables.

Quantity and Frequency of Alcohol Consumption and Marijuana Use

Quantity of alcohol consumption in the past 12 months was assessed by asking the usual number of standard drinks consumed in one episode of drinking. One unit of alcohol was defined as any beverage with 11 gr of pure alcohol; participants were shown pictures of different alcohol volumes to define equivalent alcohol units. Frequency of alcohol consumption was categorized as follows: 1 to 5 times a year; less than once a month; once a month; 2 to 3 times a month; 1 to 2 times a week; 3 to 4 times a week; almost every day; every day. We also asked if participants had used marijuana at least once in the past year (yes or no).

Alcohol-Related Injury

We asked if the students had experienced an alcohol-related injury ever in their lifetime. We also asked if the injury occurred in the past year or earlier. An alcohol-related injury was defined as getting hurt while under the influence of alcohol in a risky situation, for example getting wounded in a fall, or a car crash or “anything else.” This question was adapted from the WHO Composite International Diagnostic Interview for use in Spanish.³²

Data analyses

We performed descriptive analyses of survey data using R 3.3.3. Distribution of the density of on-premises and off-premises alcohol outlets in each neighbourhood of the city was mapped with ggmap.³³ Preliminary bivariable analyses were conducted to evaluate the inclusion of variables in the multilevel model. We used Spearman’s method for non-parametric data for continuous measures and simple logistic regression for dichotomous measures.

Because missing data in some variables may bias the model solution,³⁴ we imputed missing values for the model test with the Multivariate Imputation by Chained Equations (mice) package.³⁵ A description of this process is provided in online supplementary Table S1, http://links.lww.com/CJA/A12. If a student did not report neighborhood of residence—whether because the student did not know the name of the neighborhood or because of confidentiality concerns—this resulted in missing values for 4 neighborhood-level variables: on- and off-premises alcohol outlet density rates, crime rate, and overcrowding rate. Almost 20% of students did not report their neighborhood of residence (n=250); therefore, we imputed these 4 neighborhood-level variables. We also imputed missing values for past year injury (n=7, 1%), lifetime injury (n=7, 1%), quantity of alcohol consumed (n=23, 2%), marijuana use (n=123, 9%), and gender (n=4,2%). Significant differences were found in marijuana use between those who had an alcohol related injury and those who did not, the former less likely to report marijuana use (P < 0.05) (supplementary Table S2, http://links.lww.com/CJA/A12). We created 5 datasets and created a complete version combining all of them with the mice package function. Data distribution were similar before and after imputation according to graphical analyses, and model relationships were constant before and after imputation.

We used multilevel logistic regression analyses to assess differences in past-year alcohol-related injuries according to the neighborhood’s alcohol outlet density. These models allow the estimation of fixed effects, or individual-level variables, along with random effects, or neighborhood-level or grouping variables, suggesting a non-independence of the relationships between the fixed effects and the outcome. To test this hypothesis, we first calculated null models (M_o, simple logistic regression of alcohol-related injury) and compared them with models with neighborhood-level variables added one by one (M₁, density of on-premises and off-premises alcohol outlets, overcrowding, and crime rate). Each neighborhood-level variable was judged by the median odds ratio (MOR), and the akaike information criterion (AIC) and Bayesian information criterion (BIC). A preferable model has a MOR greater than 1, which indicates clustering among neighborhoods, and a lower AIC and BIC. After evaluation of the best null model, individual-level variables significantly related to past-year injury in bivariable analysis were added to the model. Each variable was estimated separately, and then included in the model without the neighborhood-level predictors (M₂). Because some of the variance differences in the neighborhood-level variables may be explained by individual variables, we estimated each neighborhood-level variable’s proportional change in variance (PCV) after the addition of individual-level variables (M₃) in comparison with the model without them (M₁). The final model was replicated for injuries prior to the past year (lifetime injuries, M₄). Models were fitted by lme4 package of R 3.3.3³⁶ with the generalized linear mixed model function.

RESULTS

More than 1 in 10 students reported an alcohol-related injury in the past year (n=149, 11%), and a quarter reported at least one in their lifetime (n=296, 22%) (Table 1). Fifty-seven percent of the sample were women (n=775), and 73% were under age 21 years (n=793). Over half of participants (n=577) lived in neighborhoods with 5 or more alcohol outlets per 1000 inhabitants and 6 or more crime acts per 1000 inhabitants (n=424). Usual quantity of alcohol consumption was almost 4 standard units (mean=3.85, SD=3.52). Alcohol consumption was most commonly reported 1 to 2 times per week (n=494, 37%) and 2 to 3 times per month (n=346, 26%). Roughly one-third (n=429) of respondents used marijuana in the past year (n=370) (Table 1).

Table 1:

Sample Characteristics of Neighborhoods and University Students in Mar del Plata, Argentina, 2014 (n=1346)

Measures	n(%)	Mean(SD)	95% CI1
Neighborhood characteristics (n=57)
Alcohol outlets
On-premise (per 1000)	—	5.29 (5.07)	5.02–5.58
Off-premise (per 1000)	—	0.93 (0.70)	0.88–0.97
Overcrowding rate (per 1000)	—	1.92 (0.86)	1.87–1.96
Crime rate (per 1000)	—	7.19 (4.68)	6.94–7.47
Individual characteristics (n=1346)
Alcohol related injury
Past year injury (yes)	149 (11)	—	9–13
Lifetime injury (yes)	296 (22)	—	20–24
Alcohol consumption
Quantity (in standard units2)	—	3.85 (3.52)	3.64–4.05
Frequency	—	—	—
1–5 times a year	140 (10)	—	9–12
less than once a month	90 (7)	—	5–8
Once a month	174 (13)	—	11–15
2–3 times a month	346 (26)	—	23–28
1–2 times a week	494 (37)	—	34–39
3–4 times a week	>77 (6)	—	4–7
Almost every day	17 (1)	—	1–2
Every day	4 (1)	—	0–1
Marijuana use (yes)	429 (32)	—	29–34
Sex
Female	775 (57)	—	55–60
Male	571 (43)	—	40–45
Age	—	20.76 (2.99)	20.58–20.96
Geographic mobility (yes)	370 (28)	—	25–30

¹Before multiple imputation.

²Defined as any beverage with 11 gr of pure alcohol (range: 0-32).

CI, confidence interval.

While most of the on-premises alcohol outlets were concentrated downtown, off-premises outlets were more distributed throughout the city (See Supplementary Fig. S3, http://links.lww.com/CJA/A11).

In bivariable analysis, alcohol-related injury was related to the density of on-premises alcohol outlets (past-year injury OR=1.04, 95% CI=(1.01, 1.08), P < 0.05, and lifetime injury OR=1.05, 95% CI=(1.02, 1.08), P < 0.001); quantity (past year OR=1.10, 95% CI=(1.06, 1.14), P < 0.001 and lifetime OR=1.13, 95% CI=(1.09, 1.17), P < 0.001) and frequency of alcohol consumption (past year OR=1.63, 95% CI=(1.40, 1.89) P < 0.001, and lifetime OR=1.61, 95% CI=(1.44, 1.80), P < 0.001); marijuana use (past year OR=1.93, 95% CI=(1.36, 2.75), P < 0.001, and lifetime OR=2.55, 95% CI=(1.93, 3.36), P< 0.001) and age (past year OR=0.84, 95% CI=(0.77, 0.91) P < 0.001), and lifetime OR=0.91, 95% CI=(0.87, 0.96) P < 0.001). Off-premises alcohol outlets showed no association with past-year injury (P > 0.05), yet there was a moderate relationship with lifetime injury (OR=1.33, 95% CI=(1.09, 1.63), P < 0.01). Overcrowding was only associated with lifetime injury (OR=0.76, 95%CI=(0.64, 0.90), P < 0.01). Crime and gender were not related to past year or lifetime alcohol-related injury, and, therefore, were not included the multilevel model (Table 2).

Table 2:

Bivariate Relationships Between Alcohol-Related Injury and Neighbourhood-Level and Individual Characteristics, Mar del Plata, Argentina, 2014

Measures	21	311	41	51	61	72	81	92	102	112	122
Neighborhood characteristics (n=57)
1. On-premises alcohol outlets (per 1000)	0.74**	−0.83**	0.57**	0.12**	0.14**	1.02	0.01	1.02	1.12**	1.04*	1.05**
2. Off-premises alcohol outlets (per 1000)	—	−0.49**	0.25**	0.05	0.09**	1.31**	0.00	1.08	1.30**	1.25	1.33**
3. Overcrowding rate (per 1000)	—	—	−0.60**	−0.10**	−0.11**	0.93	−0.03	0.92	0.39**	0.86	0.76**
4. Crime rate (per 1000)	—	—	—	0.08**	0.06*	1.03	0.05	1.01	1.07**	1.02	1.02
Individual characteristics n=1346
5. Quantity of alcohol consumption	—	—	—	—	0.42**	1.17**	−0.10**	1.19**	1.04*	1.10**	1.13**
6. Frequency of alcohol consumption	—	—	—	—	—	1.71**	0.08**	1.14**	1.16**	1.63**	1.61**
7. Marijuana Use	—	—	—	—	—	—	1.01	1.80**	0.91	1.93**	2.55**
8. Age	—	—	—	—	—	—	—	1.05**	1.01	0.84**	0.91**
9. Sex	—	—	—	—	—	—	—	—	1.06	.72	1.01
10. Geographic mobility	—	—	—	—	—	—	—	—	—	0.82	0.76
11. Past year injury	—	—	—	—	—	—	—	—	—	—	—
12. Lifetime injury	—	—	—	—	—	—	—	—	—	—	—

Performed before multiple imputation.

¹Spearman’s correlation (for comparisons involving continuous variables 1–6, 8).

²Logistic regressions odds ratio (for comparisons involving dichotomous variables 7, 9–12).

^*P < 0.05.

^**P < 0.01.

We tested empty models (without individual-level variables), testing on-premises alcohol outlets, off-premises alcohol outlets, and overcrowding rate, with past-year alcohol-related injury as the outcome of interest. Variability between neighborhoods was only found when on-premises alcohol outlets were included. The individual probability of having an alcohol-related injury in the past year increased (PCV=29%) when on-premises alcohol outlets were considered. No differences between M₀ and M₁ were found with off-premises alcohol outlets and overcrowding rate; therefore, they were not included in the multilevel model (Table 3).

Table 3:

Multilevel Logistic Regression Models for Alcohol-Related Injury Associated with Neighborhood-Level and Individual-Level Variables in University Students. Mar del Plata, Argentina, 2014 (n=1346)

	Neighborhood-Level Model Only1	Individual-Level Model Only1	Neighborhood- and Individual-Level Model1	Final Model Replication2
Fixed effects (OR, 95% CI)
Quantity of alcohol (in standard units3)	—	1.05 (1.01–1.10)**	1.05 (1.01–1.10)**	1.07 (1.04–1.12)*
Frequency of alcohol4	—	1.66 (1.41–1.96)**	1.66 (1.41–1.97)**	1.57 (1.39–1.77)**
Age (in years)	—	0.81 (0.75-0.89)**	0.81 (0.74-0.88)**	0.90 (0.85-0.95)**
Random effects (variance, SD)
On-premise alcohol outlet density (per 1000)	0.07 (0.27)	—	0.02 (0.15)	0.07 (0.26)
Measures of cluster
AIC	940	—	860	1307
BIC	950	—	884	1333
MOR	1.29	—	1.16	1.28

¹Outcome=past year alcohol-related injury (yes or no).

²Outcome=lifetime alcohol-related injury (yes or no).

³Defined as any beverage with 11 gr of pure alcohol (range 0–32).

⁴From (1) 1–5 times a year to (8) every day.

^*P < 0.05.

^**P < 0.01.

AIC, akaike information criteria; BIC, Bayesian information criteria; CI, confidence interval; MOR, median odds ratio; OR, odds ratio.

When included as covariates, marijuana use showed no association with past year alcohol-related injury (Table 3). Frequency and quantity of alcohol consumption and age were strongly associated with past-year injury. When individual-level variables were added to on-premises outlet density (neighborhood-level variables), the model showed a better fit according to AIC and BIC (AIC=860, BIC=884), but variance related to on-premises alcohol outlet density exhibited a reduction (PCV = −66%). Increasing density of on-premises outlets was accompanied by a 16% (MOR=1.16) increase in the odds of past year alcohol-related injury. At the individual-level, increasing quantity (OR=1.05, 95% CI=(1.01, 1.10)) and frequency of alcohol consumption (OR=1.66, 95% CI=(1.41,1.97)) was associated with significantly increased odds of past-year alcohol-related injury. As age increased, the odds of a past-year alcohol-related injury decreased (OR=0.81, 95% CI=(0.74, 0.88). The model replication with lifetime alcohol-related injury indicated similar values, with the variance explained by on-premises outlet density higher (MOR=1.28) despite poorer model fit (AIC=1307, BIC=1333).

DISCUSSION

To address the gap in the literature on alcohol-related harms in middle-income countries, we evaluated several neighborhood and individual characteristics that have been shown to impact alcohol-related harms in high-income countries. Specifically, we assessed the association between alcohol-related injury and neighborhood alcohol outlet density, crime rates, and individual age, gender, and alcohol and marijuana use. We also assessed characteristics unique to middle- and low-income countries, such as overcrowding.

Research from other high-income countries shows that off-premises outlets are more strongly associated with drinking problems,^4–6,37,38 crime,^10,39 violence,^7,8 and alcohol-related motor vehicle crashes.^9,11,40 A recent meta-analysis found that off-premises outlets were associated with increased per capita alcohol consumption.¹⁶ Our results, however, differ from those in 2 ways. First, we found only on-premises outlets were associated with alcohol-related injuries. Secondly, while on-premises outlet density was associated with alcohol-related injuries, other community stressors were not. We found no difference in alcohol-related injuries by overcrowding or crime rates. On-premises outlets’ business structure in Argentina is similar to those in other high-income countries, while off-premises outlets are different. Most beverages for off-premises consumption are sold in grocery stores or supermarkets, and only to a minor extent by stores selling alcoholic beverages exclusively, such as liquor stores. The discrepancy between our results and results from other countries may also be attributed to broader differences in social norms and drinking cultures. In Argentina, alcohol consumption is more frequent and more tolerated compared to other middle- and high-income countries in the Americas region.¹ In our study, more than 9 out 10 university students drank alcohol, and consumption among women was high. With little differences in quantity and frequency of consumption by gender, there may be similar rates in alcohol-related problems such as alcohol-related injury.⁴¹ In this study, we found no association between gender and alcohol-related injury, which differs from other countries where men’s higher alcohol consumption is associated with a higher rate of alcohol-related injury.⁴⁰

In addition to differing business structures for off-premises outlets in Argentina, it is likely that restricted hours for buying take-away alcohol contributed to our findings.¹⁶ In Mar del Plata, the sale of alcohol for off-premises consumption after 9 p.m. was outlawed in 2009.⁴² There is strong evidence that restricting the temporal availability of take-away alcohol reduces consumption,¹⁶ which may lead to decreases in alcohol-related harms Previous studies have shown that restricted dates and times of sale have correlated with reduced alcohol-related injury in high-income and some Latin American countries,^10,43,44 but as far we know this research has not been extended to South America. In sum, our results suggest restricted hours of sale may decrease alcohol-related injury rates in Argentina, highlighting the need to further examine this issue to advance evidence-based policies in the region.

We found that 3 individual characteristics were significantly associated with alcohol-related injuries in multilevel modeling: age, quantity, and frequency of alcohol consumption. Marijuana use was significantly positively associated with both frequency and quantity of alcohol in bivariable analysis, indicating that heavier and more frequent drinkers were also more likely to consume marijuana; however, marijuana use was not significant when included in the model with frequency of alcohol consumption. Previous studies suggest that marijuana use is a complement to alcohol consumption among young adults,²³ and research has indicated an additive effect when combining marijuana and alcohol use on fatal⁴⁵ and non-fatal injuries.^46,24

Given changes in legislation regarding marijuana use in Argentina, Uruguay, and other countries, research in South American countries on the effects of marijuana on injury and its role when combined with alcohol deserves further advancement.

Limitations of this study merit discussion. First, this study is cross-sectional and, therefore, does not allow for discussion of the causal relationship between alcohol outlet density and alcohol-related injuries. Furthermore, as the city of Mar del Plata has a regulation restricting temporal availability of alcohol, our findings may not be generalizable to other cities in Argentina or other South American countries. Quantity, frequency, and delivery method of marijuana consumption were not assessed in this study; these factors may affect the level and persistence of impairment leading to injury.²⁴ Alcohol consumption behavior, marijuana use and related injury were self-reported; it is possible students underreported their behaviors because of social desirability bias. Because of confidentiality concerns, we did not collect the residential address of participants; consequently, we were unable to measure distance to alcohol outlets. Our sample was composed of young adults who preferred to drink cocktails and beer, which may have affected their behavior regarding outlet preference as those beverages are consumed mainly in bars. Furthermore, our sample had a majority of women, and a previous study of Argentinean youth found different consumer behavior and attitudes regarding perceived alcohol availability between genders.⁴⁷

Despite these limitations, to our knowledge, this is the first study in the Southern Cone that assesses a relationship between alcohol outlets and alcohol-related harms. As such, this study contributes to an area with a paucity of information from non-high-income countries and may facilitate the development of culturally appropriate policies for this at-risk population.