Study design and participants
This study uses the Health Survey for England (HSE) 2005 to 2015, accessed via he UK Data Service, subject to their end user license [19]. The Health Survey for England is a nationally-representative annual cross-sectional survey of the population living in private households in England [19]. Participants were selected using multi-stage stratified-sampling; selecting participants within primary sampling unit (PSUs). Household response rates declined somewhat over the period, ranging from 74% in 2005 to 60% in 2015 [20]. Non-response weights have been calculated and were included in the datasets. Ethical approval for the HSE was obtained ahead of the data collection from the relevant ethics committee [21], data was anonymised and further ethical approval was not needed. In our study, the sample was limited to the participants aged 16 to 24 years, who answered questions about drinking status (N = 9699) in surveys between 2005 and 2015. Non-response to the drinking status question accounted for 1.5% of all 16 to 24-year olds. Information was collected via trained interviewers who administered the interview face-to-face in participants’ households using CAPI and a self-completion booklet.
Variables
Non-drinking
Non-drinkers were defined as participants who reported ‘no’ to the question on drinking status: “Do you ever drink alcohol nowadays, including drinks you brew or make at home?” Lifetime abstainers; non-drinkers who reported they had always been a non-drinker, and former drinkers; non-drinkers who reported they had not always been a non-drinker, were derived from a follow-up question specific to non-drinkers.. Non-drinkers were also asked if they drank occasionally, who we refer to as ‘occasional drinkers’. Non-drinkers have been found to be a heterogeneous group, consisting of lifetime abstainers, former drinkers and occasional drinkers [22]. In addition, to explore periodic abstinences, we also examined changes in the prevalence of not having an alcoholic drink in the past week.
Drinking patterns
Drinking patterns were identified based on alcohol units drank on the heaviest drinking day in the past week. These questions were asked consistently across the survey years between 2005 and 2015. A category for drinkers, drinking alcohol within limits at the time of the survey (not exceeding 4 units for men, and 3 units for women on any day [23]) were created. Binge drinking was defined as drinking twice the recommended daily limits on the heaviest drinking day.
Social and demographic variables
The following variables were considered as sub-groups; sex, broad ethnicity (white/non-white), full-time education versus not in full-time education, north and south regions of England, area-deprivation, measured by the Index of Multiple Deprivation (IMD) in quintiles dichotomised (three least deprived versus two most deprived area), urban location (urban/town/village), household level national-statistics socio-economic classification (NS-SEC) (managerial professional/intermediate/routine manual) and individual employment status (employed/non-employed).
Health and health behaviours
Positive health behaviours and health statuses were considered including non-smokers (versus smokers), eating five or more portions of fruit and vegetable a day (versus 3–4 or 0–2 portions), high physical activity (versus medium or low level), and up to normal Body Mass Index (BMI) category (underweight/normal (up to 24.9 kg/m2), overweight or over (25 kg/m2 or over). Apart from objectively collected data on BMI, all information was self-reported. Physical activity was measured using the short-form International Physical Activity Questionnaire (IPAQ) [24], which has been asked annually since 2013. Questions on fruit and vegetable consumption were not asked in 2012 and 2014; all other years were presented. The proportion of non-drinkers among those with no longstanding illness (versus those with a longstanding illness or limiting longstanding illness) were also explored. Mental health and wellbeing was measured through the 12 item General Health Questionnaire (GHQ-12), and the Warwick-Edinburgh Mental Wellbeing scale (WEMWBS), respectively. Total GHQ-12 scores were calculated by assigning values of 0 if symptoms were not present, or 1 if symptoms were present on each of the 12 items, and summing scores on the items together (maximum score 12). We dichotomised total GHQ-12 scores into zero (no evidence of mental illness), or 1 or more (less than optimal mental health including probable mental ill health) [25]. GHQ-12 scores were not collected in 2007, 2011, 2013 and 2015. Participants with total scores on the 14-item WEMWBS with five response categories (scored zero to five), ranging from 14 to 70 were dichotomised. Participants with scores one standard deviation below the mean were categorised as having low mental wellbeing (14–42), versus above one standard deviation from the mean (mid to high wellbeing; 43 or higher) [17]. Questions from the WEMWBS scale have been asked annually since 2010. The GHQ-12 and WEMWBS were administered via a self-completion booklet, which has a higher non-response rate.
Statistical analyses
All analyses applied complex survey design and non-response weighting. The proportion of non-drinkers among the population and corresponding confidence intervals were calculated for each year from 2005 to 2015. Significant differences were highlighted when proportions differed from the 2005 start year. Tests for linear trends in the level of non-drinking over time, were examined for each sub-group using regression analyses, modelling year as an independent variable and non-drinking as the dependent variable and adjusting for age. Trends were illustrated in charts using three-year moving averages. The same analyses were repeated among different social-demographic and health sub-groups. Information for variables with missing year’s data, were modelled as consecutive years, observing whether a significant linear increase was found among the years that data was collected.
In pooled analyses of all datasets, we examined whether the chances of being a non-drinker increased greater by year for certain sub-groups, by conducting logistic regression on the odds of being a non-drinker versus drinker, modelling an interaction effect between each sub-group and year, adjusting for age and sex. These analyses were limited to variables that had information on all years; urban area, IMD, educational, employment, household social class, smoking status, limiting longstanding illness statuses which was dichotomised (BMI was not included due to a relatively high proportion of missing BMI measurements (14%)). In preliminary analyses, the interaction effect between broad ethnic groups (white vs. non-white) and year was significant (OR = 1.06 (95%CI 1.01–1.11) p = 0.03), suggesting that the odds of being a non-drinker have increased faster for the white than non-white population. However, in models there were large effect sizes, due to sparse data problems [26]. Therefore we limited these logistic regression models to white-participants only (N = 7934).
We examined whether increases in non-drinking were related to changes in drinking patterns among young people by undertaking ecological analyses. Spearman correlation co-efficient were calculated between the proportion of non-drinkers by year and the proportion binge drinking, and mean units consumed on the heaviest drinking day. Ordinary least squares regression analyses were used to test the strength and direction of the relationship between the proportion non-drinking (independent variable) and the proportion binge/mean units (dependent variable), over time. The relationship is illustrated using scatter diagrams. As a sensitivity analyses we also examined the relationship between the proportion of non-drinkers and the proportion binge drinking and mean units consumed on the heaviest drinking among drinkers only, which does not include the numbers of non-drinkers in its calculation.