Alcohol-Related Risky Behavior Patterns and Their Association With Alcohol Use and Perceived Alcohol Stigma in Moshi, Tanzania

Abstract

Objective:

The Kilimanjaro region has one of the highest rates of reported alcohol use per capita in Tanzania. Alcohol-related risky behaviors pose substantial threats to the health and well-being of alcohol users and the people around them. This study seeks to understand how alcohol-related risky behaviors co-occur with other risky behaviors.

Method:

Latent class analysis (LCA) was applied to examine alcohol-related risky behaviors. The optimal number of latent classes was confirmed by using model fit indices. Negative binomial models were used to test latent classes and their association with harmful and hazardous drinking and perceived alcohol stigma. With the model defined, we explored each class’s drinking patterns and risky behavior patterns.

Results:

A total of 622 (60% male) of 841 participants were included in these analyses because they drank alcohol at least once in their lifetime. Three classes of risky behavior patterns were identified: Class 1, “Limited risk behaviors” (59.7%); Class 2, “Primarily foolish behaviors” (25.6%); and Class 3, “Pervasive risk behaviors” (13.1%). Class 3 had the most alcohol use quantity and frequency. No association between classes and alcohol stigma was found. Compared with males, females are less likely to be classified in Class 2 and 3.

Conclusions:

Three different classes of risky behaviors became apparent and were distinguished by gender, age, and personal alcohol use. Our findings suggest a potential role for personalized interventions based on latent classes specifically to reduce risk behaviors.

Alcohol consumption is embedded in many cultures and is commonly a part of social interactions and events. Although alcohol has at times played an important role in society and culture, it can lead to serious health and social consequences. In the World Health Organization (WHO) Africa Region, the age-standardized alcohol-attributable burden of disease and injury was highest among all WHO regions (70.6 deaths and 3,044 disability-adjusted life years [DALYs] per 100,000 people) (WHO, 2018). In 2016, Tanzania reported average alcohol consumption of 9.4 liters per capita per year (persons ≥15 years old), whereas the corresponding average for the WHO Africa Region was 6.3 liters (WHO, 2018). The prevalence of alcohol use disorders in Tanzania was 6.8% in 2016, which is 1.84 times higher than the WHO Africa Region (3.7%) (WHO, 2018). The Kilimanjaro region, which includes Moshi, has one of the highest reported rates of alcohol consumption per capita in Tanzania (Francis et al., 2015; Mitsunaga & Larsen, 2008). Studies also found that alcohol consumption has been increasing quickly in Moshi (Castens et al., 2012).

Numerous studies have reported the association between alcohol use and consequences such as drink driving, academic difficulties, unprotected sexual behaviors, unintentional injury, violence, and death (Hingson et al., 2005, 2009; Mundt et al., 2009). Alcohol-related risky behaviors pose substantial threats to the health and well-being of alcohol users and people around them as well as society. Considering the large amount of alcohol consumed in Moshi, alcohol-related risky behaviors greatly affect local people’s health. Data from the Kilimanjaro Christian Medical Center (KCMC) showed that about 30% of the injury patients consumed alcohol at the time of injury and approximately 17% of injury patients self-reported driving after consuming three or more standard drinks (El-Gabri, 2017; Staton et al., 2018). Therefore, the current research focuses on alcohol-related risk behaviors in the Kilimanjaro region.

Previous research has shown that risky behaviors are associated and tend to co-occur (Hair et al., 2009; Jessor, 1998). Studies of the co-occurrence of risky behaviors adopt two approaches—a person-centered approach and a variable-centered approach (Bámaca-Colbert & Gayles, 2010). Risk behaviors have usually been explored using variable-centered approaches in which associations between risk variables and both predictors and outcomes were studied (Ray et al., 2012). Although such analyses are important to understand the relationships between risky behaviors and associated factors, they fail to describe individual-level factors. This study seeks to provide a better understanding of how risky behaviors co-occur or how certain types of individuals tend to have risky behaviors. Therefore, we chose to take a person-centered approach.

Because prior research has shown that isolated unidimensional measures fail to capture problematic drinking, we used latent class analysis (LCA) to identify classes with distinct risky behavior patterns (Stewart & Power, 2002; Townshend & Duka, 2002). LCA not only has many methodological advantages—including control of type I error rates, high statistical power, and the ability to examine higher order interactions—but it also has the capacity to detect underlying structures of a particular set of variables (Lanza & Rhoades, 2013). Previous studies using LCA examined risky behavior patterns and associated behaviors (Hair et al., 2009; Klein et al., 1993), as well as drinking patterns and associated behaviors (Beseler et al., 2012; Bohnert et al., 2014; Chiauzzi et al., 2013; Ray et al., 2012; Reboussin et al., 2006). Ray et al. (2012) identified three classes of risky drinkers by assessing alcohol-related protective and risk behaviors. Ten percent of students showed high protective behaviors and low-risk behaviors, 30% reported pervasive risk behaviors (e.g., social partying) and less protective behaviors (walking home with a friend), and the remaining 60% reported similar frequencies of alcohol-related risk and protective behaviors.

We included perceived alcohol stigma because alcohol-related stigma is both prevalent and pervasive among people in Moshi and can influence people’s alcohol use, risky behavior, and help-seeking behavior (El-Gabri, 2017). Stigma is defined as the negative perception of the act of drinking, alcohol-related risky behaviors, or the person who drinks or persons that conduct alcohol-related risky behaviors. Perceived stigma is defined as one’s awareness of the discrimination and devaluation that is directed at those whose conditions are considered unfavorable (Link, 1987). Public stigma affects alcohol-dependent people at both the social and societal levels (Corrigan & Watson, 2002; Rüsch et al., 2005). People with alcohol use disorders may be discriminated against and/or devalued because of stigma; thus, a burden is also added at the individual level. Self-stigma arises when a patient internalizes a negative view of alcohol use toward him- or herself. As a result, perceived stigmas are related to many negative outcomes among people with alcohol disorders, including poorer psychological functioning (Smith et al., 2010), poorer physical health (Ahern et al., 2007), higher depression scores (Luoma et al., 2010), lower perceived social support, and lower treatment and care-seeking (Keyes et al., 2010; Parcesepe & Cabassa, 2013; Room, 2005; Vogel et al., 2006). Stigma may influence risky behaviors both positively and negatively. Studies have found that moderate stigma surrounding risky behaviors may reduce the spread of harmful behavior (Bayer, 2008; Livingston et al., 2012; Stuber et al., 2008). On the other hand, alcohol stigma is found to exacerbate negative alcohol consequences: it may delay help-seeking behaviors, because people fear being labeled as alcoholics and subsequently incur discrimination (Room, 2005). But, to our knowledge, the association between alcohol stigma and risky behaviors has not been studied.

This study is the first attempt to explore the alcohol-related risky behavior patterns in a low- and middle-income country. The specific goals of this study are to (a) identify profiles of alcohol-related risky behaviors in a mixed sample of injury patients and the general population; (b) explore the characteristics of each profile group; and (c) examine how these profiles are associated with alcohol use patterns and perceived alcohol stigma in our sample. With the model defined, we explore each profile’s drinking patterns and risky behavior patterns. Our findings may help in the development of interventions that can be tailored to reflect the actual need of a particular patient and eventually reduce harmful alcohol use.

Method

Study setting

Moshi is located in the Kilimanjaro Region of northern Tanzania with more than 180,000 people (Tanzania National Bureau of Statistics, 2012). The majority of people in Moshi are members of the Chagga, Pare, and Masai ethnic groups (Tanzania National Bureau of Statistics, 2012). The Kilimanjaro Christian Medical Center (KCMC) is located in Moshi and is a referral hospital for more than 15 million urban and rural people in northern Tanzania.

Sample and procedures

Participants were made up of two populations: 341 injury patients identified on arrival to the KCMC Emergency Department who suffered an injury and a convenience sample of 500 adults selected from Moshi, Tanzania. The general validation population was chosen from different random public locations in downtown Moshi and was included in the study if they were at least 18 years of age, spoke Swahili, and provided informed consent. The public locations selected included markets, stores, and bus stations in urban Moshi. The markets and bus stations are common places for urban and rural residents to visit. The general population may represent the people living near town areas in the Kilimanjaro region. The general validation sample size of 500 people was determined by calculating the statistical power and the local alcohol consumption rates.

Injury participants were included if they were at least 18 years old, seeking care at the KCMC Emergency Department for an injury of any severity, were determined to be clinically sober and medically stable by the treating physician, were able to communicate in fluent Swahili, and consented to participate before discharge from the hospital. In prior studies of injury patients from the KCMC with the same inclusion criteria as our stated population, 30% tested positive for alcohol use at the time of injury (Staton et al., 2018). In effect, we oversampled injury patients to 40.5% of the total sample because they generally have more alcohol-related consequences than the normal population. The selection of this mixed population increased the sensitivity of our analyses and allowed us to focus on our planned future intervention population.

Measures

Indicator variables.

Alcohol-related risky behaviors were assessed through 10 questions from the impulse control consequences subscale of the Drinker Inventory of Consequences (DrInC). The DrInC is a 50-item harm assessment questionnaire that is used specifically for assessing adverse consequences of alcohol abuse. The DrInC measures five categories: Interpersonal, Physical, Social, Impulsive, and Intrapersonal aspects (Miller et al., 1995). The impulse control subscale uses a binary yes/no response format to assess alcohol-related risk behaviors. We also examined the gender and age of participants but not types of injuries and characteristics such as education, income, and employment. The Swahili version of the DrInC has been cross-culturally validated in the local Tanzanian Swahili culture (Zhao et al., 2018).

This study excluded 2 questions from the original 12-question impulse control subscale because of their similarities to other questions, indicated by the face validation and strong phi-coefficients (r > .5): “I have broken things or damaged property while drinking or intoxicated” and “While drinking or intoxicated, I have injured someone else.” The first removed question is very similar to the Foolish risk question and Impulsive things that I regret question. The second removed question is similar to the Physical fight question. The questions and variables used are summarized in Appendix 1. (The appendix appears as an online-only addendum to this article on the journal’s website.)

Alcohol use.

We used latent classes to predict three questions from the Alcohol Use Disorders Identification Test (AUDIT) as well as the AUDIT full scale. The AUDIT is an instrument used to identify people with problem drinking patterns (Babor et al., 2001). The first three items in the AUDIT were used to assesses alcohol intake quantity and frequency with a Likert scale format, as shown in Appendix A. The AUDIT’s psychometric properties have been validated in many different regions, including Tanzania (Claussen & Aasland, 1993; Gache et al., 2005; Piccinelli et al., 1997; Skipsey et al., 1997; Steinbauer et al., 1998; Vissoci et al., 2018).

The AUDIT score is a continuous variable that ranges from 0 to 40, but we treated it as a dichotomous variable in this study because studies have shown that the standard cut-off score of 8 provides adequate specificity and sensitivity and yielded the best discrimination when used to detect alcohol-related social and medical problems (Conigrave et al., 1995; Saunders et al., 1993; Vissoci et al., 2018). This dichotomization also makes interpretation of the AUDIT score easier.

Alcohol stigma.

Alcohol stigma was assessed with the Alcohol-Adapted Perceived Discrimination-Devaluation (PDD) scale. The alcohol-adapted PDD is an instrument to assess an individual’s perceived alcohol stigma (PAS) toward drinkers (Glass et al., 2013). The PDD contains 12 six-point Likert scale questions and has shown good psychometric properties in various settings (Luoma et al., 2010, 2013; Ruan et al., 2008). Seven PDD questions assessed perceived discrimination of heavy drinkers and five questions assessed perceived devaluation (Glass et al., 2013). Perceived discrimination is when people feel they are purposefully being mistreated because of their alcohol use. Perceived devaluation is when an individual perceives oneself to be less valued because of alcohol use.

Six items used reverse wording to avoid response biases. Items with reverse wording were recoded so that the high PAS score indicates higher stigma. The maximum score is 84 (6 points × 12 questions). The PDD has been used to assess stigma on both high-risk drinkers and abstainers. A high PAS score is associated with poorer mental health and a lower chance of alcohol treatment in high-income settings (Glass et al., 2013; Keyes et al., 2010; Smith et al., 2010).

Statistical analysis

Data analysis and the generation of figures were performed with R software (Version 3.3.3), and a significance level was set at .05.

Descriptive statistics were initially performed to evaluate data quality and characteristics. Results were reported with imputed data to avoid the listwise deletion of cases that have missing values. Missing data in the DrInC, AUDIT, and PDD were imputed through multiple imputation using the mice package in R, and a sensitivity analysis did not find any significant differences in the model with or without the imputed data (Buuren & Groothuis-Oudshoorn, 2010). Multiple imputation was selected to avoid uncertainty brought by single imputation, which could lead to errors in inferences made (Graham, 2009).

LCA was applied to examine alcohol-related risky behaviors. First, the optimal number of latent classes was confirmed by using model fit indices. The global fit indices were informed by the Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC), taking into account the goodness of fit and parsimony. Lower values of the BIC and AIC indicate better fit (Muthén, 2004). Likelihood ratio tests were used to examine whether a model with multiple profile groups fit the data better than a model with one profile group. Entropy is a measure of aggregated classification uncertainty (Celeux & Soromenho, 1996). A higher entropy value represents a better fit. The normalized entropy value ranges from 0 to 1; a value greater than .8 indicates the latent classes are highly discriminating (Tein et al., 2013).

Model fitness was assessed based on the model fit indices as well as the research purpose and model interpretability as recommended in the literature (Jung & Wickrama, 2008; Kriston et al., 2011). Models with Profile Groups 1–6 were tested by the poLCA package in R software (Linzer & Lewis, 2011). Negative binomial models were used to test latent classes and their association with the AUDIT score and PDD score.

Results

Demographics

Table 1 provides an overview of the sample characteristics. Of the 841 patients surveyed, this study included only those who consumed alcohol at least once in their lifetime: 242 injury patients and 380 general population (n = 622). Of the 622 participants, most were male (60%) and had an average age of 41.73 years (SD = 23.86). Of the participants, 72.2% (n = 452) reported consuming alcohol in the past 12 months. Among them, 53.10% consumed alcohol at least two times a week, and the majority (85.62%) reported consuming no more than four drinks per drinking day, as shown in Table 1.

Table 1.

Sociodemographic profile of the validation sample

Variables
Age, in years, M (SD)	41.73 (23.86)
Male, n (%)	375 (60%)
Consumed alcohol in the last year, n (%)	452 (72.20%)
Drinking frequency, n (%)
Monthly or less	107 (23.67%)
2–4 times a month	105 (23.23%)
2 or 3 times a week	139 (30.75%)
≥4 times a week	101 (22.34%)
Drinking quantity (per drinking day, n (%)
1 or 2	214 (47.35%)
3 or 4	173 (38.27%)
5 or 6	36 (7.96%)
7–9	16 (3.54%)
≥10	13 (2.88%)

Number of latent classes

Three different models were suggested by the fit indices (Table 2). A three-class model and four-class model were suggested by the consistent AIC. A four-class model and five-class model were suggested by BIC. The six-class model was suggested by the adjusted BIC. Because different statistical fit indices suggested different models, we decided to give research objective and practical interpretability more weight as suggested by the literature (Jung & Wickrama, 2008; Kriston et al., 2011). The three-class model and four-class model were selected because the sample distribution in the three-class model (59.67%, 25.57%, and 14.76%) and four-class model (54.79%, 23.48%, 14.54%, and 7.17%) are relatively even and they are the only classes that meet the standard that the size of all subgroups more than 5% of the whole cohort (Nielsen et al., 2016). Ultimately, we selected the three-class model because the highly contrasted risky behavior patterns in the three-class model are easier to interpret than the four-class model, and many items among classes in the four-class model are indistinguishable. For example, 6 of 10 risky behavior items (drink drive, use other drugs, overweight, had accident, legal trouble, smoke more) have a similar pattern between Class 3 and Class 4 in the four-class model, limiting the interpretability of the four-class model.

Table 2.

Fit indices for the models of latent class analysis

Model	Log likelihood	Resid. df	BIC	aBIC	cAIC	Likelihood ratio	Entropy	Smallest class proportion
1-class solution	11,192.38	582	22,668.09	22,528.40	22,712.09	1,7149.06	–	–
2-class solution	-7,464.49	537	15,502.09	15,219.53	15,591.09	9,693.29	.975	25.4%
3-class solution	-6,799.74	492	14,462.36	14,036.93	14,596.36	8,363.79	.956	14.8%
4-class solution	-6,599.55	447	14,351.74	13,783.44	14,530.74	7,963.39	.874	7.17%
5-class solution	-6,470.50	402	14,383.42	13,672.25	14,607.42	7,705.31	NaN	3.65%
6-class solution	-6,385.20	357	14,502.58	13,648.55	14,771.58	7,534.70	NaN	3.47%

Notes: Resid. = residual; BIC = Bayesian Information Criterion; aBIC = adjusted BIC; cAIC = consistent Akaike Information Criterion; NaN = not a number.

Latent class characteristics

The three class profile groups comprise sufficient numbers of participants in each class. The “Limited risk behaviors” class has 374 (59.67%) participants, the “Moderate risky behavior” class has 160 (25.57%) participants, and the “Pervasive risk behaviors” class has 92 (14.76%) participants. The risky behavior patterns are shown in Figure 1 and Table 3.

Figure 1.

Class characteristics regarding different risky behaviors

Table 3.

Characteristics of the identified classes

Class	Drink drive	Use other drugs	Foolish risk	Impulsive things that regret	Physical fight	Smoke more	Over- weight	DUI	Legal trouble (other than drink drive)	Had accident
1. Limited risk behaviors (n = 374)	3.67%	0.80%	2.66%	0.08%	0.56%	2.93%	1.37%	0.00%	0.00%	1.35%
2. Primarily foolish behaviors (n = 160)	15.15%	3.12%	41.43%	27.11%	11.67%	17.57%	8.04%	3.99%	2.48%	10.52%
3. Pervasive risk behaviors (n = 82)	30.33%	16.24%	91.74%	85.84%	58.71%	46.45%	18.41%	22.31%	31.43%	42.36%

Note: DUI = driving under the influence.

Participants in our “Limited risk behaviors” class exhibited low alcohol-related risky behaviors. Only 3.7% of the participants in this class had drink driving behavior, 2.9% had smoked more because of drinking, and 2.7% had taken foolish risks after drinking. Less than 2% of the participants in this class participated in other risky behaviors.

Participants in the “Primarily foolish behaviors” class showed a moderate amount of alcohol-related risky behaviors. In this class, 41.4% of participants had taken foolish risks and 27.1% had done impulsive things that they regretted later after drinking. Less than 20% of the individuals in this class participated in other risky behaviors.

As expected, participants in the “Pervasive risk behaviors” class showed very high numbers of alcohol-related risky behaviors. In this class, 91.7% of participants had taken foolish risks and 85.5% had done impulsive things that they regretted later after drinking, 58.7% of participants had engaged in an alcohol-related physical fight with others, and 46.4% had smoked more because of alcohol use. All risky behaviors in Class 3 occur at least two times more frequently than in the “Moderate risky behavior” class, with certain risky behaviors such as had an accident, legal trouble, and smoke more occurring three or more times.

Cross-class comparisons

In the third step, we compared classes on (a) sociodemographic variables, (b) alcohol use, and (c) alcohol stigma. Table 4 shows the detailed results of the analysis. To better understand the associations between the risky behavior profiles and drinking patterns and alcohol stigma, we converted the coefficients derived from models to the odds ratio. We compared the mixed sample results and each population’s results (injury patients or the general population) and found no significant difference in profile group, alcohol use, and perceived stigma in either population.

Table 4.

Cross-class comparisons on demographics, alcohol use, and alcohol stigma

Variable	Gender (% female)	Age M (SD)	AUDIT > 8%	Alcohol Use quantity M (SD)a	Alcohol use frequency M (SD)b	Binge drinking frequency M (SD)c	Alcohol stigma, PDD M (SD)
Class
1. Limited risk behaviors (n = 374)	51.3%	42.7 (16.1)	36.4%	0.52 (0.75)	2.31 (1.09)	0.41 (0.94)	38.8 (8.7)
2. Primarily foolish behaviors (n = 160)	30.2%	42.4 (39.0)	59.4%	0.91 (0.94)	2.58 (1.06)	0.86 (1.23)	39.1 (8.3)
3. Pervasive risk behaviors (n = 92)	10.9%	36.7 (12.7)	76.1%	1.37 (1.11)	3.07 (1.03)	1.52 (1.41)	38.3 (7.4)
Odds ratio
Class 2 compared with Class 1	0.29***	0.90	2.47***	2.31***	1.42*	2.45***	1.01
Class 3 compared with Class 1	0.07***	0.23***	5.36***	5.24***	3.53***	6.37***	0.997

Notes: With the exception of gender (% female) and Alcohol Use Disorder Identification Test (AUDIT) > 8, class data are means (SD). PDD = Perceived Discrimination-Devaluation.

^a0 = never; 1 = monthly or less; 2 = 2–4 times a month; 3 = 2 or 3 times a week; 4 = 4 or more times a week.

^b0 = 1 or 2; 1 = 3 or 4; 2 = 5 or 6; 3 = 7–9; 4 = 10 or more.

^c0 = never; 1 = less than monthly; 2 = monthly; 3 = weekly; 4 = daily or almost daily.

^*p < .05;

^***p < .001.

Sociodemographics.

Gender proportion differed significantly between risky behavior classes. Females comprised 51.3% of Class 1 but only 30.2% of Class 2 and 10.9% of Class 3. The age distribution differed significantly between Class 1 and 3 but not between 1 and 2. The average age of Class 3 was 6 years younger than it was in Class 1 and 2.

Alcohol use.

Significant cross-class differences were observed for all alcohol use items after controlling for age and gender (Table 4). Members in three classes were found to have increasingly higher values across items including alcohol use quantity, frequency, and binge drinking. Class 2 was 2.47 and Class 3 was 5.36 times more likely to be classified as having an AUDIT score above 8 than was Class 1. The box plot between the AUDIT score and profile groups also suggested this association.

Perceived alcohol-related stigma.

The alcohol-adapted PDD revealed no significant differences between profile groups (Table 4). The average PDD scores ranged between 38.3 and 39.1 among the different classes.

Discussion

This study is a first attempt at categorizing alcohol-related risky behaviors and exploring drinking patterns and alcohol stigma in a low- to middle-income country like Tanzania. Three profiles for risky behavior emerged from our LCA: “Limited risk behaviors,” “Primarily foolish behaviors,” and “Pervasive risk behaviors.” Overall, our findings indicate that (a) a three-class model fits our population, with the severity of the behavior describing the differences between these classes; (b) the volume of drinking is associated with the severity of a broad array of risky behaviors; and (c) we did not find an association of perceived alcohol-related stigma with risk behavior.

Three adequately large classes of participants with distinct risky behavior patterns were identified. A total of 59.7% of the sample reported very low risky behaviors and were classified as a limited risk behaviors class (Class 1). Coinciding with previous literature, this class contains far more women (51.3%) and has significantly lower alcohol use quantities and frequencies compared with the other two classes (Chiauzzi et al., 2013; Hair et al., 2009; Reboussin et al., 2006). The mean age of participants in Class 3 is 6 years older than those in Classes 1 and 2, indicating that those who have the most-risky behaviors are younger than those who have medium or less risky behaviors.

Three classes showed obvious differences in the AUDIT score: alcohol use quantity, frequency, and binge drinking frequency. Those with more risky behaviors drink more and more often. These relationships are not surprising since those who drink more are more likely to suffer from alcohol-related consequences (Cooper, 2002; Swahn & Bossarte, 2007). However, it should be noted that alcohol use quantity and binge drinking frequency are more diverse among three classes than alcohol use frequency. This can be explained by the pervasive habitual drinking behavior among people in Moshi (Castens et al., 2012).

In contrast, we expected to find classes with different types of risk behavior (foolish vs. legal), but instead we found that the broad array of risk behaviors increased with alcohol use. The most obvious contrast of items between three classes were foolish risks, impulsive things that were regretted, and physical fights. This suggested that these three items were among the most common risks and much more likely to happen with increased intake of alcohol, although foolish risks and impulsive things usually did not come with severe consequences. In contrast, although only 22.3% and 31.4% of participants in Class 3 had driving under the influence and legal troubles, respectively, they are likely to receive special attention since they have the most serious risky behaviors. Previous qualitative research of the same population found that alcohol use is only deemed a problem by the population when it causes serious consequences like physically harming others (Meier et al., 2019). Instead of broadly focusing interventions on both Class 2 and Class 3 drinkers, with health promotion and harm reduction strategies, respectively, we can reduce the overall risk for alcohol-related injuries. Similarly, the topics to include, in an educational strategy or otherwise, might also cover this large diverse array of consequences.

From a systems standpoint, having myriad alcohol-related harm reduction strategies (i.e., violence reduction, alcohol reduction, drink driving reduction) available to be differentially deployed for these classes could markedly reduce risk. A risky behavior–based classification system may be more appropriate, as it is the alcohol-related consequences that have been known to cause major harm, thus potentially identifying the highest risk groups amenable to interventions. We understand that the health system in Tanzania is already overburdened and adding an additional layer of screening could be challenging. But ultimately, this helps declutter the system. This system can identify who is low risk for harm and focus our otherwise overburdened system on those most at risk for alcohol-related harms.

Our results showed that perceived alcohol-related stigma, although high (range: 38.3–39.1), was similar within all groups. To our knowledge, no previous literature has investigated the relationship between perceived alcohol stigma and risky behavior patterns. Some studies have shown that disproportionately high perceived stigma among affected drinkers may prevent them from seeking treatment (Smith et al., 2010). As the relatively strong perceived alcohol stigma is prevalent in this population and stigma does not differ between the three classes, we feel our data show that stigma does not deter excessive alcohol use or risky behaviors. Alternatively, given that the PAS scale was constructed from a U.S. English–speaking context, the concepts of stigma and alcohol-related stigma might not be the same. Although we were able to adapt and demonstrate the psychometric properties of the PAS in our population, the construction of a stigma scale de novo from the Tanzanian population might highlight different important concepts.

Limitations

Our results should be considered in light of three main limitations. First, our mixed sample was drawn from injury patients presenting to a hospital and a population-based cohort. The injury population is made up of patients who survived the injury with relatively good physical function and were able to provide informed consent and answer our questionnaires; this is representative of the overall injury population in this setting (Staton et al., 2017). We purposely oversampled injury patients because they experienced more alcohol-related consequences than the general population. Also, the population-based cohort was recruited in random public places in town using a convenience sampling method; thus, they might not best represent the general population. Therefore, our mixed participants are unlikely to represent the whole of the Tanzanian population and it is possible that different drinking patterns would have been found in other regions and populations within Tanzania. Second, self-reported data on the standard drinks of alcohol consumed may be imprecise due to the common drinking practice of sharing one cup of homebrew among family members or friends. Drinkers may not be aware of the quantity or percentage of alcohol contained in a cup, and it might be hard for them to estimate the number of drinks consumed. However, to increase our data quality, our Tanzanian research team adapted to the local culture, providing examples and pictures to help participants estimate the number of drinks they consumed. Third, we did not measure variables such as marital status, education, or tobacco use, which could be useful when identifying the characteristics of different profile groups. Fourth, the stigma scale we used only assessed external perceived stigma. We hope to incorporate self-perceived stigma into the analysis in the future by using other scales such as the Self-Stigma in Alcohol Dependence Scale (Schomerus et al., 2011).

Implications for prevention and intervention

Identification of these three risky behavior patterns offers valuable information for clinicians and harm reduction policymakers. Previous literature has recommended incorporating information on alcohol-related risky behaviors when designing prevention and intervention programs (Borsari, 2004; Borsari et al., 2007; Ray et al., 2012). Our classification system may help the individualization of alcohol interventions to focus on behaviors specifically considering alcohol use patterns. A brief intervention such as health education might be suitable for people classified in the class “Limited risk behaviors,” whereas those in the “Primarily foolish behaviors” or “Pervasive risk behaviors” groups might need further intervention adaptations, like the inclusion of other more advanced and specific harm reduction strategies (i.e., a violence, drink-driving educational strategy) (Choo et al., 2013). For example, after detecting a physical fight risk behavior, health personnel may provide intervention to the person targeted on drink driving or other legal troubles, depending on his or her situation.

Future research directions

As the first analysis of alcohol-related risky behavior patterns in a sub-Saharan African country with an LCA, validation of our findings in other sites is necessary. In addition, longitudinal studies concerning the course of development of risky behavior patterns are desired to determine where the transition from one risky behavior pattern to another is. Future research should also explore whether the intervention effectiveness varies among classes and whether we should apply different alcohol use intervention methods to different classes.

Conclusion

To our knowledge, this is the first study to investigate alcohol-related risky behaviors in a mixed sample from a sub-Saharan African country with an LCA. Our findings provide insights into the sizes and characteristics of three classes with different levels of risky behaviors and associated alcohol use patterns but all with similar levels of perceived alcohol-related stigma. Further research about our two higher risk classes and their alcohol-related harm reduction best practice strategies is warranted.

Acknowledgments

The authors acknowledge our KCMC/Duke ED Research Team, without whom our research would be impossible.