Design
Online survey data were collected in Australia during October 2017, as part of a broader study of attitudes towards alcohol consumption in the presence of children [24]. Eligible participants were adults aged 18–59; across Australia, the legal age for alcohol purchase or consumption on licensed premises is 18 years. Potential participants across each Australian state and territory who were already registered with a survey provider (i.e., who had previously been recruited for survey research, and had provided basic demographic details to facilitate invitation to other research for which they may be eligible) were recruited via an email invitation. Initial screening items were used to confirm participant eligibility and representation by key demographic quotas (i.e., approximately equal male–female ratio, ≥ 50% parents) within a total pre-specified sample of N = 1,000. Once a quota had sufficient representation, additional participants within the quota could not proceed with completing the full survey.
Only data from those who regularly consumed alcohol were retained for the present analyses; of the 1,000 initial respondents, 197 (19.7%) who did not report past-year alcohol consumption at least monthly were excluded. Open-text comments from a further two (0.2%) suggested non-genuine responses, so all data from these respondents were also excluded. Therefore, the final sample comprised 801 participants.
Procedure
Participants were offered redeemable points from the survey provider (e.g., for gift cards or charitable donations) equivalent to < AUD$5. The provider satisfied International Organization for Standardization standards (AS ISO 26362).
Measures
Demographic and health-related characteristics
Participants reported gender, age, marital status, completed education, current paid employment status, annual household income (AUD$, pre-tax), and parental status (living with child/children under 18 years). On the basis of respondent postcode, remoteness (major cities vs regional/remote areas, according to Australian Statistical Geography Standards [25]) and general relative socioeconomic disadvantage (collapsed to most disadvantaged [quintiles 1–2] vs least disadvantaged [quintiles 3–5] using the Index of Relative Socioeconomic Disadvantage 2016, a composite area-level measure based on 16 individual- and household-level variables in population data, e.g., household internet connectivity and occupation classifications [26]) were also derived.
BMI (kg/m2) was calculated from height and weight, reported in participants’ preferred (metric/imperial) units; 153 (19.1%) reported they ‘did not know’ or ‘could not say’ their height (n = 119, 14.9%) and/or weight (n = 125, 15.6%). Improbable low outlier BMI values more than three times below interquartile range (≤ 11.1, n = 6) were excluded. Participants also reported perceived weight on a single item (“Do you consider yourself to be…?”) with response options ‘Underweight’, ‘An acceptable weight’, or ‘Overweight’, plus ‘Don’t know/Can’t say’.
Alcohol consumption characteristics
Per population surveys [18], standard graduated frequency items accompanied by an image of the number of standard drinks in typical servings of various beer, wine, and spirit beverages were used to assess alcohol use [27]. Past-year average daily consumption of > 2 standard drinks (each containing 10 g/12.7 mL pure alcohol) indicated long-term high-risk consumption above the threshold in national (NHMRC) guidelines [27]. Revised NHMRC guidelines were released in December 2020. However, the 2009 long-term guideline (“For healthy men and women, drinking no more than two standard drinks on any day reduces the lifetime risk of harm…”) was current during data collection, and therefore used to assess patterns of drinking known to put health at risk at this time ([27] p. 3). As opposed to the 2009 single occasion/short-term risk guideline, the long-term guideline acknowledged cumulative longer-term effects of alcohol consumption, including a potential association with overweight and obesity.
Past-year frequency of alcohol consumption was based on single item “In the last 12 months, how often did you have an alcoholic drink of any kind?”, with six frequency levels (e.g., Every day, 5 to 6 days a week) collapsed to ‘Daily or weekly’ vs ‘Monthly’.
To measure current alcohol-related risk perceptions, participants estimated low-risk levels using a single item for each of adult males and females (i.e., How many “standard drinks” do you believe an adult male/female could drink every day for many years without adversely affecting his/her health?) [18], with a ‘Don’t know’ option (47.1% and 53.2% for males and females respectively). Although Australian guidelines present a singular recommendation for adults, only own-gender estimates were used as considerable gender differences persist in alcohol-related risk perceptions [28]. In analyses, we distinguished estimates of ≤ 2 standard drinks (i.e., below or consistent with the NHMRC long-term high-risk guidelines and therefore considered to not over-estimate alcohol-related risk) [27] from combined over-estimate/don’t know responses (i.e., over-estimated risk).
Harm minimisation strategies used when consuming alcohol
Frequency of use for four harm minimisation strategies when usually consuming alcohol were assessed (three existing items “Count the number of drinks you have”, “Deliberately alternate between alcoholic and non-alcoholic drinks”, “Limit the number of drinks you have in an evening due to driving” [18], in addition to original item, “Limit the number of drinks you have in an evening (e.g., to get up early for children, sport etc.)”). Responses were required on a 5-point scale, and collapsed to indicate strategies used at least sometimes (‘Always/Most of the time/Sometimes’ = “Yes” vs ‘Rarely/Never’ = “No”).
Reductions in alcohol consumption
Past-year reductions in alcohol consumption were reported in the survey using five dichotomous items (e.g., In the last 12 months have you… “Reduced the amount of alcohol you drink at any one time”), which represented reductions in both the volume and frequency of alcohol consumption, temporary and more permanent cessation, as well as substitution with lower-alcohol beverages. Participants were instructed to ‘tick all that apply’, with each item checked (as opposed to unchecked) in the survey form included as a positive response for analysis. Additional options ‘None of the above’ (selected by 40.0% of participants), ‘Don’t know/ unsure’ (4.0%), and ‘Prefer not to answer’ (0.2%) were provided; a response was required on at least one option of the measure.
Primary outcome: changing alcohol consumption behaviours because of energy-related concerns
Due to a dearth of existing research on changes in alcohol consumption because of energy-related concerns, two single-item outcome measures were developed to assess the usual frequency of this behaviour: When you have an alcoholic drink, how often do you… “Limit the number of drinks because you are concerned about the calories/kilojoules/effects on body weight”, and “Drink lower carb [carbohydrate] alcohol because you are concerned about the calories/kilojoules”. The latter item included reference to only energy content, as there is little evidence that links lower-carbohydrate (‘low-carb’) drink consumption with reduced body weight. Specific health claims such as weight loss cannot be made for alcoholic beverages under Australian law, except for nutrition content claims including on the basis of energy or carbohydrate [29]. As such, energy content is often emphasised in lower-carb alcohol promotions [30], and was assumed to be more salient to respondents than effects on body weight. Responses to the two items were required on a 5-point scale. They were strongly positively correlated (Pearson r = 0.68, p < 0.001, n = 801) and demonstrated good internal consistency (α = 0.81), so to distinguish participants who reported one or both of the behaviours at least sometimes (collapsed from ‘Always/Most of the time/Sometimes’ = “Yes”) from those who reported neither of the behaviours (both ‘Rarely/Never’ = “No”), a combined variable was used for analyses.
Analyses
Frequency analyses explored sample characteristics. Harm minimisation strategies used when consuming alcohol, reductions in alcohol consumption, and motivations for reduction were compared by participant subgroups (those who had and had not changed alcohol consumption behaviours because of energy-related concerns) with Pearson’s chi-squared tests and Fisher’s Exact tests as appropriate. These analyses were stratified by gender, as gender differences are well-established in reducing (i.e., more frequently reported by men) and ceasing consumption behaviour (more frequently reported by women) [31].
To inform the predictor variables used in the modelling approach for each of the overall and gender-stratified samples, chi-squared tests were also used to assess the association between demographic and alcohol-related predictors, and the outcome measure of changing alcohol consumption behaviours because of energy-related concerns. Only significant predictors from univariate analyses in each of the respective samples were then entered simultaneously in multivariable binary logistic regression models (with those ‘Rarely/Never’ changing alcohol consumption behaviours because of energy-related concern as the reference category), which were used to determine the association between demographic and alcohol-related characteristics and the likelihood that participants changed alcohol consumption behaviour because of energy-related concerns. The multivariable model in the total sample included six demographic characteristics (gender, age group, remoteness, current employment status, household income, and parent status), in addition to risky alcohol consumption and frequency of alcohol consumption. The model for males included three demographic characteristics (age group, remoteness, current employment status) and frequency of alcohol consumption; these variables were also included for females, in addition to marital and parental statuses and risky alcohol consumption. All analyses were conducted in IBM SPSS Statistics Version 21 (IBM Corp, Armonk, NY, USA), and a two-sided Type I error rate of α = 0.05 was assumed for all significance tests.
Missing data
Unless otherwise specified, ‘Don’t know’ or similar response options were not used for other measures. Listwise deletion was used where relevant; in addition to BMI data (described above), remoteness and socioeconomic status could not be derived from non-valid postcodes (n = 4, 0.5%). All other data were complete.