1.1. Heavy drinking

Heavy drinking patterns (commonly termed ‘binge’, ‘risky single occasion’, ‘heavy episodic’ or ‘short-term risky’ drinking) involve consuming a relatively high amount of alcohol in a relatively short period of time [3, 11, 12]. Heavy drinking is most common on Friday and Saturday nights when young people go out and have few work or study responsibilities the following day [13, 14]. Short term consequences of heavy drinking include blackout, memory loss, nausea, vomiting, hangovers, alcohol poisoning, unintended and/or unprotected sexual activity, injury, traffic accidents, and death [2, 3, 11, 15–17]. Those who engage in heavy drinking may also be more likely to harm others, for example via vandalization, inter-personal violence and aggression, and traffic accidents due to drink-driving [3, 4, 18–22]. While heavy drinking is typically discussed in terms of immediate harms, consequences can also be long-term (e.g., permanent disability from an injury sustained during a traffic accident). Furthermore, there is evidence indicating heavy drinking in early life is linked to numerous long-term negative consequences [3, 11, 17, 23, 24].

Event-level alcohol consumption refers to an individuals’ drinking pattern during a given occasion. An occasion typically refers to a day or evening, but may be more specific (e.g., during a visit to a venue). Event-level alcohol consumption is distinct from measures of alcohol consumption across longer time-periods (e.g., average daily alcohol consumption in the last 12 months, usual alcohol consumption per drinking occasion in last week). Unless otherwise stated, hereafter ‘drinking’ refers to event-level alcohol consumption, ‘heavy’ or ‘heavier’ drinking refers to higher event-level alcohol consumption, and ‘light’ or lighter’ drinking refers to lower event-level alcohol consumption.

1.2. Drinking occurs in complex contexts that influence drinking behaviour

According to a social ecological perspective of human behaviour [5, 6, 8–10], the immediate drinking context can be characterised by the physical and social attributes of the setting, the characteristics and state of individuals, and the interactions of these components. This model implies that contextual factors influence whether an individual engages in heavy drinking and whether alcohol-related consequences are experienced. Further, associations between individual-level factors (e.g., gender, age, personality and motives) and heavy drinking may be altered by contextual factors (e.g., time, place, occasion and presence of others).

1.3. The broader drinking context: a sequence of immediate drinking contexts

Previous studies have focused on the independent effects of factors on drinking occasions and consequences (e.g., [25–28]. However, in the real world, factors may combine or co-occur during a drinking occasion in a specific sequential manner that shapes the broader drinking context and influences an individuals’ drinking. Fig 1 provides a visual depiction of the broader drinking context. As shown, the broader drinking context comprises a sequence of immediate drinking contexts that are described by combinations and sequences of factors related to the characteristics and state of individuals, the physical environment and the social environment. Pre-drinking or pre-loading provides an example of a risky sequential combination of context-related factors—the act of drinking alcohol, usually at a domestic residence, prior to attending a social event, typically at a bar or nightclub [29, 30]. Occasions that include pre-drinking are associated with heavier drinking than those that do not [29, 31–34]. Moving between several outlets (e.g., pub crawling) provides another example of a risky combination of sequential factors.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. The broader drinking context: Comprising a sequence of immediate drinking contexts which are described by combinations of factors related to the characteristics and state of individuals, the physical environment and the social environment ¹.

¹ As per a social ecological perspective of human behaviour [5, 6, 8–10].

https://doi.org/10.1371/journal.pone.0218465.g001

Individual-level factors may also moderate associations between sequences of factors and drinking; for example, the positive association between pre-drinking and likelihood of heavy drinking is stronger among men with lower conformity drinking motives than men with higher conformity motives [35].

The complexity of the broader drinking context (Fig 1) means that it is imperative for research to consider the interplay (i.e., interaction or combined effect) of multiple factors relating to the context and the individual (rather than simply estimating the independent effect of one factor), as well as the specific sequence of these factors. This is necessary to comprehensively explain associations between drinking contexts and heavy drinking and alcohol-related harms.

1.4. Studying drinking contexts and event-level associations

Two main types of studies have been used to investigate simultaneous and/or prospective relationships of context-related factors and individual-level factors with drinking. The first is event-level studies, which collect data during the event(s) using methods such as ecological momentary assessment (EMA). The second is event-based studies, which collect data retrospectively about an event (e.g., retrospective survey) or events (e.g., timeline follow back; TLFB). In theory, both designs allow the exploration of event-level associations between specific contexts and heavy drinking. However, their ability to comprehensively describe contexts and explore event-level associations between contexts and drinking varies.

Some event-level and event-based studies have identified specific combinations of factors related to drinking contexts and/or individuals that are linked to heavier drinking occasions among adolescents and young adults. For example, Thrul and Kuntsche found that the positive association between drinking with friends and heavier drinking was stronger among males than females [36]. Lau-Barraco and colleagues report that drinking occasions were particularly heavy on weekend days for those with higher social alcohol expectancies [37]. Another study reported that occasions spent in public locations with few intoxicated people are associated with lighter drinking [38].

1.5. The need for a review of the literature on the contexts of heavy drinking

To our knowledge, no review has comprehensively summarised the evidence on the specific contexts that are associated with heavy drinking. Further, we found no studies that comprehensively reviewed and discussed the design and methodology of these studies. A comprehensive review of the literature is needed to compile and summarise the specific contexts associated with heavy drinking patterns.

To address these gaps, we conducted a systematic review which:

1. Summarises the immediate contexts (described by specific combinations of context-related and individual-level factors) and broader contexts (described by specific sequences of immediate contexts) that are associated with heavier drinking occasions. It provides a more comprehensive understanding of how contexts influence drinking, highlights gaps in understanding of contexts influencing heavy drinking and suggests opportunities for effective environmental approaches to reduce alcohol-related harms due to heavy drinking; and
2. Summarises, critiques and proposes ways to improve the design and analytical methodology of the scientific literature that investigates event-level associations between contexts and drinking.

2.1. Study identification, eligibility and screening

A systematic review of English articles using MEDLINE, Embase and the Cumulative Index to Nursing and Allied Health Literature (CINAHL) databases was conducted to identify eligible articles. The review protocol was registered with PROSPERO (registration number: CRD42018089500). The review was conducted in line with the PRISMA Statement [39] (see S1 Table for completed PRISMA checklist). Given the complexity of drinking contexts, this review targeted studies that provided descriptions of drinking contexts via sequences or combinations of multiple factors related to the context. Articles eligible for inclusion in data extraction and synthesis were those which: (a) included a quantitative analysis of event-level data; (b) estimated associations of sequences or combinations of two or more factors (at least one context-level variable) with event-level alcohol consumption of the individual; and (c) reported a combination or sequence that was directly associated with significantly increased (‘risky’) or decreased (‘protective’) event-level alcohol consumption, and (d) sampled from broadly Western countries (this criterion was added after the full-text screening phase and prior to data extraction because the majority of eligible studies sampled from Western countries). To maximise the comparability of the findings with general populations, studies with samples comprised only of those with alcohol use disorder (for example) were ineligible. Associations eligible for extraction (Table 1) were those which:

1. statistically tested for differences in effect between groups of respondents described by combinations of factors (e.g., via t-test, odds ratio, beta coefficient, interaction), or it was possible to conservatively test via 95% confidence intervals, or reported the level of endorsement of responses in latent class analysis; and
2. observed a statistically significant difference (increase or decrease) in the outcome at the 95% confidence level (e.g., p < 0.05 or, when p was not reported, 95% confidence intervals that do not include null or do not overlap), or heavy or lighter drinking endorsed by the majority of a class (>0.50 probability) in latent class analysis.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Criteria for judging eligibility of retrieved articles for inclusion in data extraction and synthesis.

https://doi.org/10.1371/journal.pone.0218465.t001

Where studies presented both bivariate and multivariate estimates of a given association, only the effect estimates that are adjusted for potential confounding variables were extracted. For practical and conceptual reasons, qualitative studies and associations of factors with alcohol-related harms (but not event-level alcohol consumption) were not eligible for inclusion in this review. Further details about the eligibility criteria are described in Table 1.

The search strategy was designed in direct consultation with La Trobe University library staff with expertise in conducting systematic review literature searches. To ensure the search was highly sensitive to retrieving eligible records, we used a detailed list of search terms that describe alcohol drinking, event-level or event-based study design, and combinations, interactions or sequences. The search was conducted on the 29^th of January 2018. A copy of the full search strategy is provided in S2 Table. Additional relevant articles were identified by contacting experts in this topic of research who suggested articles which were then screened for eligibility.

The process for screening records for eligibility is described in Fig 2. Titles and abstracts, then full texts of all retrieved records were screened by an independent researcher (OS) using the criteria described in Table 1. Records that OS was initially unsure whether to include or exclude were screened for eligibility by a second researcher (FL, CW or AP), and differences in opinions of eligibility were resolved via the majority opinion of a group of five researchers (OS, FL, CW, EK and AP).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Flow diagram of identification of eligible articles for review.

¹ Of these, 15 were duplicates, 12 were excluded based on full text and 11 were included in quantitative synthesis; Records eligible for full text screening with no full text published in English were translated to English then screened; Date searched: 29 January 2018; Fig 2 adapted from: [39].

https://doi.org/10.1371/journal.pone.0218465.g002

2.2. Data extraction and synthesis

The following information was extracted by OS from the full text of each article included for review:

1. Type of study design: e.g., EMA, daily diary study, TLFB survey, intercept survey, retrospective survey.
2. Sample description: e.g., N; adult, student, patron of night-time precinct, drinker; country; age range.
3. Outcome variable description: e.g., number of drinks consumed, breath alcohol concentration (BrAC), heavy drinking episode; continuous or categorical.
4. Predictor variable(s) description: e.g., pre-drank, number of friends present, gender; continuous or categorical; event-level (e.g., today) or individual-level (generally); sequence or combination.
5. Type and magnitude of effect estimate(s) and comparison statistic(s): e.g., difference in means or proportions between groups described by combinations of factors, odds ratio, beta-coefficient, interaction coefficient; p-value, 95% confidence interval, probability.
6. Analytical approach used: e.g., multi-level modelling, person-mean centering of event-level predictors, latent class analysis, factor analysis or principal components analysis.
7. Direction of association(s) (for each valid association): e.g., heavier drinking (i.e., risky context) or lighter drinking (i.e., protective context).

The Quality Assessment Tool for Observational Cohort and Cross-Sectional Studies [40] was used by one researcher (OS) to identify potential sources of bias in the design or conduct of the study.

Combinations and sequences that were consistently associated with heavier drinking were compiled into a list summarising ‘risky contexts’ for men, for women and for either gender. Similarly, combinations and sequences associated with lighter drinking were compiled into a list of ‘protective contexts’ for men, women and either. Contexts were divided into categories that denoted whether the context was described by factors relating to the characteristics or state of the individual, physical environments and/or social environments. Specific combinations or sequences that were not unidirectionally associated with the outcome across studies (e.g., a combination was associated with heavier drinking in one study, but lighter drinking in another) were dropped (see S3 Table for dropped associations). A meta-analysis was not conducted because few studies investigated comparable combinations and sequences, thus yielding effect estimates that were not collapsible across a sufficient number of studies.

3.1. Study selection

The flow of records throughout the screening process is presented in Fig 2. The literature database searches retrieved a total of 1,953 records. A further 38 records were retrieved via consultation with experts (‘other sources’). After the removal of 89 duplicate records, the titles and abstracts of 1,902 records were screened according to the eligibility criteria (Table 1). A further 1,661 records were excluded during the screening of titles and abstracts. Of the remaining 241 records which underwent full text screening, 163 were excluded during the full text screening and a further 13 were excluded during critical appraisal or data extraction. The reasons for exclusion during title and abstract screening, full text screening, and critical appraisal and data extraction are listed in Fig 2. The remaining 65 articles were included in the review.

3.2. Study descriptions and methodologies

3.2.1. Study sample.

Characteristics of the studies included in the review are described in Table 2. The sample size ranged from 47 to 60,215 individuals. However, it is important to note that EMA, daily diary and TLFB [41] designs produce observations for multiple occasions per individual, and EMA designs produce observations for multiple timepoints per occasion per individual. Thus, despite having a low sample size, the data from the EMA, daily diary and TLFB studies included in this review typically encapsulated hundreds of occasions or more. The majority (52/65; 80%) of the studies used data from samples comprised mostly or entirely of adolescents or young adults: students (36/65; 55%); nightlife precinct patrons (12/65; 18%); and people aged < 30 years who were not specifically students or nightlife precinct patrons (5/65; 8%). Twenty-nine studies (45%) restricted their sample to drinkers. Almost all of the studies drew their sample from populations in North America (42/65; 65%) and Europe (17/65; 26%). Participants were primarily of white/Caucasian ethnicity.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Study characteristics and risk of bias within studies.

https://doi.org/10.1371/journal.pone.0218465.t002

3.3. Synthesis of results: Contexts that encourage and discourage heavy drinking

The contexts that were found to be associated with heavier drinking are summarised in Table 3 (denoted via ↑). Contexts associated with lighter drinking are denoted via ↓.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Contexts–described by combinations and sequences of factors related to the characteristics or state of the individual, the physical environment and the social environment ¹ –associated with heavier drinking (↑) and lighter drinking (↓) ².

https://doi.org/10.1371/journal.pone.0218465.t003

4.1. Associations between context and heavy drinking, policy implications, and opportunities for research

Elements of contexts that were commonly associated with heavier or lighter drinking occasions included a person’s mood throughout the day, the size, gender and expectations of the social group, the location where drinking takes place, and whether certain events or parties are attended. However, the strength and direction of associations between context-level factors and drinking differ according to the characteristics of the individual (e.g., their gender, age, personality and motives). The variety of interactions among the elements of a context demonstrate the complexity of relationships between contexts and drinking behaviours. In fact, in a given social or physical context, some people may feel compelled to continue or accelerate their drinking, whereas others may be influenced to slow down, drink less, or choose not to drink at all. Therefore, it is important to consider the physical, social and individual elements of a context when endeavouring to understand contextual influences on people’s drinking. This is also an important consideration for interventions designed at the individual-level. Identification of contexts that are risky or protective for people of a particular gender and age and with particular personality traits and drinking motives may help to design and implement effective policies for reducing heavy drinking occasions and related harms among specific sub-populations.

Identifying contexts that are associated with heavy drinking and contexts associated with lighter drinking is useful because it enables targeted interventions and policies, thus potentially reducing heavy drinking and alcohol-related harms. Concurrent discouragement of risky contexts (via policies, targeted interventions, health promotion and education) and encouragement of protective contexts (via the same avenues) may further reduce heavy drinking and alcohol-related harms.

The search strategy retrieved 31 studies that are well-suited to measuring sequences of factors across an occasion (17 EMA studies and 14 intercept surveys). However, very few studies investigated links between specific sequences during an occasion and event-level drinking, aside from pre-drinking. It is also possible that some eligible studies that investigate sequences of factors during an occasion and event-level drinking were not captured by the literature search. One of the advantages of EMA studies is they can record the time of specific occurrences across the course of a drinking occasion [99]. Therefore, new EMA studies or analysis of existing EMA data are needed to identify specific sequences that influence young adults to accelerate, maintain, decelerate, or cease drinking on a given occasion.

Relatively few contexts were described in relation to the physical environment. Research investigating whether and how physical contexts are associated with an individual’s drinking behaviour is needed because, in a practical sense, modification of the physical environment may be relatively feasible (e.g. luminosity, noise level and density restrictions, location-specific text message interventions) [100]. Technologies such as smartphone-based environmental measurement tools may be useful for measuring physical environments and investigating associations with heavy drinking behaviours [101–103].

A minority of the studies included in this review described contexts via a combination of factors related to the individual, the social environment and the physical environment. Future event-level studies that consider all the main elements that comprise a context have the potential to improve understandings of how and why specific contexts can influence drinking behaviour.

4.2. Study design and analytical methodology: implications and opportunities for studying drinking contexts

Numerous types of event-level or event-based designs were used by the studies included in this review. However, their ability to comprehensively describe contexts and explore their associations with event-level drinking is varied.

To accurately distinguish between the contextual influences at the event-level and the individual-level, studies which capture information about multiple occasions, such as EMA and daily diary studies are advantageous because they can properly account for both inter- and intra-individual variations using multi-level modelling [104, 105]. Studies that account for inter- and intra-individual variations using multi-level modelling can explore the true effects of contexts as drivers of within-person variation in event-level drinking [104, 105]. Multi-level modelling also enables event-level predictor variables to be centered according to each individual’s mean measurement across all occasions (called person-mean centering). Person-mean centering uses the person’s usual behaviour as its own baseline, and standardizes the momentary behaviour by considering only the deviation from the usual behaviour [105].

Studies with longer recall periods, such as retrospective surveys about single occasions and TLFB surveys, are more susceptible to recall bias [106].

Event-level studies may harness new technologies such as smartphone environmental measurement tools and continuous objective monitoring of alcohol use to improve measurement of contextual factors and behaviours. Continuous objective monitoring of blood alcohol concentration in real time via objective measures such as transdermal sensors may reduce the risk of self-reporting biasing measurements of alcohol consumption and intoxication levels [107]. Smartphones may provide useful tools for objectively measuring other contextual factors and behaviours [100, 101, 103, 108]. Although, data gathered via built-in sensors, camera, microphone and other features are not without limitation and subjective self-report questionnaires may temporarily remain the most practical method for measuring many contextual factors and behaviours.

This review found that most studies that explore event-level associations between contexts and drinking were conducted among samples of adolescents and young adults, and often students. Therefore, many of the risky and protective contexts identified may generalise to younger populations, but not to older populations. The scope of this review was widened from young adults to all ages during the project conception and design phase, to attain a more-complete review of the relevant literature and to compare the representation of young samples and older samples in the literature on drinking contexts. The high representation of young samples in the literature appears warranted, given that heavy drinking occasions are most frequent and intense during late adolescence and early adulthood [3, 109, 110]. Although, a proportion of older adults are heavy drinkers and, recently in many countries, alcohol consumption levels have decreased in younger age groups but have been maintained or increased in some older age groups [1, 111–114]. Thus, further exploration of event-level associations between contexts and drinking among older populations may be warranted to identify and discourage contexts associated with heavy drinking in middle-late adulthood. Further, as all studies included in the review sampled from Western countries, the results may generalisable to younger populations in Western countries only.

Similarly, the results of this review may only generalise to broad Western populations, not to socio-demographically-defined sub-populations, which are known to vary in alcohol use and problems. Therefore, research investigating event-level associations between contexts and heavy drinking in specific sub-populations of both Western and non-Western countries is necessary.

The intuitive way to construct variables and interpret model estimates may partly explain why fewer protective contexts were identified than risky contexts. The reference category of a categorical variable or the lower value of a continuous variable typically corresponds with a lower level of exposure–e.g., a categorical variable for pre-drinking with reference category ‘no’ and comparison category ‘yes’. Also, the format of effect estimates of regression models (e.g., odds ratios, beta coefficients) make it intuitive to interpret associations as the comparison category (or higher value) versus the reference category (or lower value)–e.g., ‘pre-drinking is associated with heavy drinking’. The same effect estimate could also be interpreted as the reference category versus the comparison category–e.g., ‘not pre-drinking is associated with lighter drinking’. Both interpretations are correct and provide a slightly different interpretation, although the former (e.g., comparison category vs. reference category) is more intuitive and commonly reported. While in some cases it is possible to derive protective contexts by taking the reverse of the reported association (e.g., ‘did not pre-drink’), we chose not to do so because this is generally not how estimates are interpreted and explained in literature. Thus, the tendency for lower values to be a variable’s reference category/value may partly explain why fewer protective contexts were identified than risky contexts.

This review identified many well-designed studies that investigated event-level associations between contexts and drinking using appropriate analytical methodology. However, some studies were highly susceptible to recall bias, and the absence of multi-level modelling in 18 studies meant some studies may not have properly accounted for between-person differences that may bias event-level effects [104, 105].

4.3. Strengths and limitations

Traditionally, most event-level and event-based studies of drinking contexts have focussed on describing independent associations between singular isolated context-related factors and drinking outcomes. This review excludes a section of the literature that does not conceptualise and measure contexts as combinations or sequences of factors. As drinking contexts are described by the complex interaction of factors related to the physical and social environment and the individuals within them [5, 6, 8–10], these associations were not eligible for inclusion in this review. Therefore, there are likely numerous studies not included in this review that take a simpler approach and identify context-related factors which are independently associated with heavy drinking. However, the focus of this review on more complex conceptualisation and measurement of drinking contexts provides a novel and useful review of the cutting-edge of event-level alcohol research. Given the complex conceptualisation of drinking context employed by this review, developing a search strategy that effectively captures the relevant literature was difficult and a potential limitation. The terms used to identify event-level or event-based study design did not list specific study designs (e.g., timeline follow back, ecological momentary assessment) (S2 Table). An objective of this review was to summarise and critique the study design of the event-level drinking literature, and inclusion of such terms would bias the search to capturing study designs that are well known to the researchers a-priori. Experimental studies were not included in this review because they do not represent real-world settings but provide another method for studying drinking contexts. Screening and data extraction were conducted by an independent researcher. To minimise this limitation, records that the researcher was initially unsure whether to include or exclude were screened for eligibility by a second researcher, and differences in opinions of eligibility were resolved via the majority opinion of a group of five researchers. While event-level alcohol consumption and intoxication (the review’s outcome measure) is associated with alcohol-related harms, this review does not explore direct associations with between contexts and alcohol-related harms. While qualitative literature was not included in this review, such studies provide clues as to the types of contexts that may be associated with heavy drinking and related harms. Therefore, there may be some context-related factors associated with occasions involving heavy drinking or harms not captured in this review that warrant further exploration.

4.4. Conclusions

This review found that the contexts an individual encounters on a given occasion are associated with how heavily they drink alcohol during that occasion. The direction and magnitude of these associations differ according to the gender, age, personality, motives and mental state of the individual, such that in a given social or physical context some people may feel compelled to continue or accelerate their drinking whereas others may be influenced to slow down, drink less, or choose not to drink at all. Contexts or factors are experienced in specific sequences that shape the broader drinking context and influence drinking behaviours and alcohol-related consequences across drinking occasions. However, risky contextual sequences are under-studied. Therefore, event-level alcohol research should prioritise improving understandings of the types and mechanisms of contextual sequences that are associated with heavy drinking and alcohol-related harms. New technologies such as smartphone environmental measurement tools and continuous objective monitoring of alcohol use and multi-level analytical methods are recommended to improve understandings of why people engage in heavy drinking. Continued research investigating event-level associations between contexts and heavy drinking will facilitate public health interventions and policies that reduce heavy drinking and alcohol-related harms.