【latex版】水贴

Results We fitted an ARIMA (autoregressive integrated moving average) model to test for a change following the increased ‘alcopops’ tax in April 2008. There was no significant decrease in alcohol-related ED presentations in 15–29-year-olds compared to any of the controls. We found similar results for males and females, narrow and broad definitions of alcohol-related harms, under-19s and ED presentations at night-time and weekends.

Conclusions The increase in tax on ‘alcopops’ did not result in any reduction in alcohol-related harms in this population. Targeting particular alcoholic drinks may therefore not be as effective as more comprehensive policies such as minimum unit pricing for alcohol.

Young adults are vulnerable to alcohol-related harm and so drinks specifically marketed at them are of particular concern.1 For instance, up to 62% of 15–16-year-olds in Europe reported recently consuming alcopops, premixed spirit-based beverages that are highly sweetened and modelled on non-alcoholic or energy drinks.2

In terms of health outcomes, most injuries, suicides and drownings in young people are associated with alcohol intoxication.3–5 This is compounded by the adverse effects of alcohol on development and the fact that alcohol use in youth predicts problematic use in adulthood.6

Measures to reduce high-risk alcohol consumption have included minimum unit pricing,7 advertising bans and random breath testing.8 Whereas there is evidence that raising alcohol duty across the board can reduce alcohol-related harms,9 there is less for measures that target specific types of alcohol beverage in isolation. An example was the 70% increase in excise duty on ‘premixed’ alcoholic beverages (‘alcopops’) implemented in Australia on 27 April 2008. The rationale was that this targeted increase would reduce alcohol consumption among young people of both sexes10 given their preference for premixed spirits and spirits over other forms of alcohol.5

The effectiveness of this measure has been extensively debated.11 ,12 In particular, it is unclear whether young people reduced harmful consumption, absorbed the price increase or changed to other alcoholic drinks with no effect on alcohol-related harm. Alcohol sales data reported a substantial fall in the sales of ready-to-drink beverages in the 3 months following introduction of the tax, with a smaller shift to other beverages (beer and spirits) and a net reduction in overall sales.13 However, sales data can be contradictory with reports from Europe that increasing the tax on alcopops did not influence total alcohol consumption when the price of other beverages remained unchanged.2 Moreover, sales data reflect overall consumption, not the amount of risky drinking in certain population groups.14 ,15 Neither do they take into account changes in the alcohol content of drinks over time.16 Indicators of health outcomes, such as health service use, may therefore be more appropriate to assess the success of such policies.

An initial study of hospital admissions and emergency department (ED) presentations in young people for alcohol-related incidents across Australia found no decrease in alcohol-related harms following the tax increase,17 while another restricted to New South Wales (NSW) reported a reduction.18 However, health outcomes in both studies were limited to blood alcohol levels, or to alcohol-attributable mental health conditions such as intoxication, dependence and abuse. These form the minority of alcohol-attributable conditions for which young people are admitted, unintentional and intentional injuries being the most common causes in both males (66%) and females (59%).11 In addition, the Australia-wide study did not apply time series techniques to adjust for underlying secular trends.17

Two further studies included a wider range of alcohol-related harms, including trauma and used interrupted time series to adjust for seasonal and secular trends.19 ,20 Both found no change in either ED presentations or hospital admissions.19 ,20 However, they also had limitations. One was confined to the Gold Coast, a popular tourist destination in Queensland for end-of-school celebrations, which may affect generalisability to elsewhere in Australia or overseas.19 The other used a more representative sample from wider Queensland, including all hospital admissions in the state, as identified by both hospital administrative data and the Queensland Trauma Registry, but only covered 29% of the jurisdiction's EDs.20 In addition, there were only data for 1 year following the tax increase and alcohol-related presentations were not compared to a control group. It is therefore possible that the results could have been confounded by changes in population or catchments of the emergency departments included in the study. We therefore undertook a study that included controls, covered more EDs and extended follow-up to 2 years following the increase in the tax. Our hypothesis was that the tax increase was not associated with any change in alcohol-related harms in spite of the documented fall in the sale of alcopops immediately following implementation.13

Two further studies included a wider range of alcohol-related harms, including trauma and used interrupted time series to adjust for seasonal and secular trends.19 ,20 Both found no change in either ED presentations or hospital admissions.19 ,20 However, they also had limitations. One was confined to the Gold Coast, a popular tourist destination in Queensland for end-of-school celebrations, which may affect generalisability to elsewhere in Australia or overseas.19 The other used a more representative sample from wider Queensland, including all hospital admissions in the state, as identified by both hospital administrative data and the Queensland Trauma Registry, but only covered 29% of the jurisdiction's EDs.20 In addition, there were only data for 1 year following the tax increase and alcohol-related presentations were not compared to a control group. It is therefore possible that the results could have been confounded by changes in population or catchments of the emergency departments included in the study. We therefore undertook a study that included controls, covered more EDs and extended follow-up to 2 years following the increase in the tax. Our hypothesis was that the tax increase was not associated with any change in alcohol-related harms in spite of the documented fall in the sale of alcopops immediately following implementation.13

Method
This was a quasi-experimental design using an interrupted time series (ITS) analysis with multiple controls of the effect of the increase in alcopops tax covering 3 years before, and 2 years after, the change. Where applicable, we followed the STROBE guidelines (STrengthening the Reporting of OBservational studies in Epidemiology).21 However, given the limited applicability to interrupted time series we also followed recommendations of the Cochrane Collaboration such as a minimum of 20 data points prior to the intervention.22 It was approved by the relevant University and Queensland Health Human Research Ethics Committees.

Subjects
We used the Emergency Department Information System (EDIS) to measure presentations of 15–29-year-olds from 28 April 2005 to 28 April 2010. We were limited to 11 hospitals (out of a possible 28) that contributed data to EDIS for the duration of the study. However, importantly, these hospitals saw 60% of all ED presentations in Queensland (see Results). We used the following International Classification of Diseases 10th Edition (ICD-10) codes in the principal diagnosis field: F10 codes for mental and behavioural disorders due to alcohol; S00-T18 codes for injury (excluding superficial injury S codes); Y90–91.9 for evidence of alcohol involvement by level of intoxication or blood alcohol level; R78.0 for a finding of alcohol in blood; and Z04.0–0.5 for blood-alcohol and blood-drug test, or examination and observation following injury. Where present, we also obtained external cause of injury codes as follows: V01–94 for transport accidents; W00–X59 for other accidental injuries except W20–W21 and W35-W64; X60-X84 for intentional self-harm; and X93-Y34 for assault and events of undetermined intent. We used narrow and broad definitions of alcohol-related harm. The former was restricted to codes that are solely associated with alcohol: F10, Y90–91.9, R78.0 and Z04.0–0.5. The latter included the narrow definition plus all the injury codes. We used alcohol-attributable fractions (AAFs) to adjust for the fact that not all injures are due to alcohol. AAFs assign the likelihood that any given condition has an association with alcohol using previously published clinical data. Alcoholic cirrhosis, for example, has an AAF of 1.0, while road accidents (V01-V89) have a value of 0.4 for males and 0.31 for females. The reported prevalence is multiplied by the AAF to estimate the morbidity due to alcohol. We used AAFs derived from data from Australia or Britain. 23 ,24 Where cases did not have F, V, W, X or Y codes, we were unable to apply cause-specific AAFs. We therefore applied an average across all injuries of that type using AAFs appropriate to the relevant gender and age group.

We compared the narrow and broad definitions of alcohol-related harm in 15–29-year-olds with the following ED controls: (1) 30–49 year-olds with alcohol-related harms; and (2) 15–29 year-olds with asthma (J45) or appendicitis (K35.9). These two diagnoses were chosen given neither was associated with alcohol or substance use. We used two control groups to check if results were consistent across all comparisons in spite of any differences in either age or gender between the cases and controls. We did not use the Queensland population as a denominator as the opening of new EDs during the study period could alter the catchments of departments included in the study and so lead to misleading results.

Analysis
We calculated the ratio of 15–29-year-olds presenting with alcohol-related harms to numbers in both the control groups. As presentations for alcohol-related harms, among others, may be subject to seasonal fluctuations, time series analyses were employed to test for any significant change in the ratio of presentations among 15–29-year-olds for alcohol-related harms before and after the ‘alcopops’ tax increase to those of the other two control groups. We used the X11 procedure to identify and adjust the series for trend, seasonality and autocorrelated data errors.25 This technique identifies seasonal factors and decomposes the original series into seasonal, irregular and trend components. Examination of the underlying trend can provide a more useful indication of the overall direction of a time series with significant seasonality.

Autoregressive integrated moving average (ARIMA) modelling was then used to test for any significant interruption to the time series following the tax increase.26 ,27 This is a regression analysis to test for a break at the time of the tax increase, taking into account any seasonal autocorrelations. Based on evidence of an increasing trend in most of the series prior to the ‘alcopops’ tax increase, we applied one order of differencing to create stationary series for modelling. Differencing takes into account long-term increases in alcohol-related presentations that could mask the effect of the increased tax. We found the ARIMA (1,1,0)(1,0,0)12 model to be the best fit, testing for white noise with the Ljung-Box statistic. We also included standard and seasonal autoregressive components to model the time series structure. We visually inspected residuals and used Q-tests to ensure there were no unexplained patterns over time.

We undertook several sensitivity analyses. First, we applied AAFs for all injuries without taking into account external cause of injury codes. As we could not find Australian AAFs for S and T codes, we used Swiss data.28 We then did a sensitivity analysis of including S codes for superficial injury. We also stratified the analyses by gender in case the proportion of males to females varied between the cases and any of the controls. Similarly, we assessed if there was any difference between younger and older age groups within the 15–29-year-old sample by looking at 15–19-year-olds only. Next, we investigated if there was any difference when we restricted alcohol-related presentations to the narrow definition. In addition, we only considered presentations between 22:00 and 06:00, or on weekends when alcohol would be most likely to be a factor. Finally, to test whether the introduction of the tax was associated with a change in the seasonal pattern of alcohol presentations or affected the underlying rate of growth in presentations for alcohol-related harms we fitted alternative forms of the overall models (males, females, persons, 15–29 years, broad definition) with dummy terms for each month and also with a deterministic trend without differencing.

Analysis
We calculated the ratio of 15–29-year-olds presenting with alcohol-related harms to numbers in both the control groups. As presentations for alcohol-related harms, among others, may be subject to seasonal fluctuations, time series analyses were employed to test for any significant change in the ratio of presentations among 15–29-year-olds for alcohol-related harms before and after the ‘alcopops’ tax increase to those of the other two control groups. We used the X11 procedure to identify and adjust the series for trend, seasonality and autocorrelated data errors.25 This technique identifies seasonal factors and decomposes the original series into seasonal, irregular and trend components. Examination of the underlying trend can provide a more useful indication of the overall direction of a time series with significant seasonality.

Autoregressive integrated moving average (ARIMA) modelling was then used to test for any significant interruption to the time series following the tax increase.26 ,27 This is a regression analysis to test for a break at the time of the tax increase, taking into account any seasonal autocorrelations. Based on evidence of an increasing trend in most of the series prior to the ‘alcopops’ tax increase, we applied one order of differencing to create stationary series for modelling. Differencing takes into account long-term increases in alcohol-related presentations that could mask the effect of the increased tax. We found the ARIMA (1,1,0)(1,0,0)12 model to be the best fit, testing for white noise with the Ljung-Box statistic. We also included standard and seasonal autoregressive components to model the time series structure. We visually inspected residuals and used Q-tests to ensure there were no unexplained patterns over time.

We undertook several sensitivity analyses. First, we applied AAFs for all injuries without taking into account external cause of injury codes. As we could not find Australian AAFs for S and T codes, we used Swiss data.28 We then did a sensitivity analysis of including S codes for superficial injury. We also stratified the analyses by gender in case the proportion of males to females varied between the cases and any of the controls. Similarly, we assessed if there was any difference between younger and older age groups within the 15–29-year-old sample by looking at 15–19-year-olds only. Next, we investigated if there was any difference when we restricted alcohol-related presentations to the narrow definition. In addition, we only considered presentations between 22:00 and 06:00, or on weekends when alcohol would be most likely to be a factor. Finally, to test whether the introduction of the tax was associated with a change in the seasonal pattern of alcohol presentations or affected the underlying rate of growth in presentations for alcohol-related harms we fitted alternative forms of the overall models (males, females, persons, 15–29 years, broad definition) with dummy terms for each month and also with a deterministic trend without differencing.