Beverage-specific mortality relationships in US population data

Abstract

Nearly all countries regulate distilled spirits differently from beer and wine. However, the scientific basis for this differential treatment has not been clearly established. This review covers beverage-specific relationships with alcohol-related mortality causes at the population level in the US and includes a new analysis of motor vehicle accident mortality. Spirits were found to be more strongly related to, or the only beverage type with a significant positive association with, cirrhosis, IHD and head and neck cancer mortality. Spirits sales, along with beer, were also implcated in motor vehicle accident mortality and homicde. While all three beverage types were found to be associated with suicide mortality, only spirits had a significant unlagged association. These associations were found under policy arrangements where spirits were discouraged with relatively higher taxes, lower availability and more restricted marketing opportunities. This suggests that the differential policy treatment found in the US and most other countries is justified on the basis of differential harm relationships.

Introduction

Nearly all countries regulate distilled spirits differently from beer and wine. Differences include higher tax rates, more restricted availability and more extensive advertising restictions. However, the scientific basis for this differential treatment has not been clearly established. Spirits, when drunk at high alcohol concentration levels as measured by the percentage alcohol by volume (%ABV), result in quicker intoxication (except on a full stomach) and can be drunk to achieve higher blood alcohol concentrations (BAC) than possible with wine or beer. (Roine et al., 1993) These higher BAC levels are associated with accidents and other acute health risks and can result in death from alcohol poisoning. Certainly wine and beer can be drunk to achive intoxication and their use is associated with both acute and chronic alcohol related health risks as well as with alcohol dependence. There is no question that all beverage types are potentialy harmful. The question is to what degree do harms increase with the %ABV of the beverage as sold (or as drunk)? While we cannot directly address this with aggregate data, we can examine population-level asociations between per capita alcohol consumption sold in the form of beer, wine and spirits and alcohol-related mortality causes. This review considers studies from the United States, where federal and state-specific policies favor beer and wine over spirits in a variety of ways. We also contribute to this literature with new analyses of beverage-specific per capita consumption and motor vehicle mortality rates.

Taxation and Availability

The US has relatively low taxes on all beverage types and prices are low compared to most countires. Spirits are taxed at considerably higher rates than beer or wine by the federal government and in all states. (Kerr, 2010b) However, because of the generally low level of taxation, prices for spirits are not currently higher in terms of the comparable cost per standard drink and are typically lowest for large bottles of vodka (a July 12, 2011 drugstore ad included 1.75L 40%ABV vodka for US$9.99, about US$0.25 per US standard drink (14g ethanol)) or 5 liter boxes of wine. Tax rates are assessed in US dollars per beverage unit (eg. gallon of spirits) for nearly all states and beverage types, resulting in the amount of tax per standard drink being lower on higher %ABV brands within types. Rates are not indexed for inflation or updated often, so taxes have declined in real value over time from relatively high levels in the 1950’s (The federal spirits tax in 2011 was US$13.50 per proof gallon and in 1952 it was US$10.50, more than 6 times higher in real terms). Beer and wine are also favored in terms of availability in many US states where sales are allowed in supermarkets, drug stores and convienience stores, while spirits are sold only in liquor stores or state-run stores. This varies widely and some states do allow all beverage types in all store categories. Some states also allow “beer and wine only” on-premise licences, which are lower cost and allowed in much higher numbers than full licences covering spirits as well.

Consumption Trends

Beer is currently the most popular beverage with about 55% of sales, spirits is second with about 30% and wine lowest with about 15%. Per capita spirits sales declined dramatically from 1974 to 1994 and have since risen slowly each year. Beer sales declined more slowly, gaining market share from 1980 to around 1995 and have been flat since. Per capita consumption of wine has been rising steadily since the early 1990’s. (Kerr, Greenfield, & Tujague, 2006) These significant changes in sales and relative market share over time, along with between state variation, facilitate the estimation of seperate effects on mortality by beverage type.

Survey-based Associations

US survey research where beverage type is assessed indicates that spirits is most associated with heavy and dependent drinking, followed by beer and that often both are used by heavy drinkers. Both beer and spirits have been associated with binge drinking and beer has been especially associated with binge drinking for younger and under-age (<21) drinkers. (Rogers &Greenfield, 1999) Wine only drinkers tend to be the lightest drinkers and to have the fewest problems. (Kerr, Greenfield, Bond, Ye, & Rehm, 2009) In understanding beverage-specific assocaitions with mortality it is important to consider shifting underlying associations with population sub-groups, particularly birth cohorts. Analyses of the US National Alcohol Surveys (NAS) have found that pre-1945 birth cohorts show a preference for spirits, while the baby boom cohort (1946–64) appears to prefer beer. (Kerr, Greenfield, Bond, Ye, & Rehm, 2004) Our most recent studies have shown younger drinkers shifting to spirits and wine, although beer is still popular in this group. Wine has been generally rising across all ages and both genders (a period effect) and wine drinking appears to be higher among older drinkers. (Kerr, et al., 2009) News accounts of industry research suggest that marketing studies have found similar trends.

Reviews of Aggregate Mortality Studies

Cirrhosis of the Liver

Research utilizing US data has long noted the close association between per capita spirits consumption and cirrhosis mortality rates. (Terris, 1967) This relationship was confirmed in a fixed-effects analysis on a panel of US states (Gruenewald & Ponicki, 1995a) and in an autoregressive integrated moving average (ARIMA) analysis of a longer time-series of the whole US (Roizen, Kerr, & Fillmore, 1999). In the most recent (Ye & Kerr, 2011), using series of liver cirrhosis mortality rates from 1950 to 2002 for 47 U.S. states (4 states were dropped to achieve a balanced panel), the effects of alcohol consumption were estimated from pooled ARIMA models and four types of panel data models: generalized estimating equations, generalized least squares (GLS), fixed effects and multilevel models. Various specifications of the error term structure under each type of model were also examined. Different approaches to controlling for time trends and for using concurrent or accumulated consumption as predictors were also evaluated. When cirrhosis mortality was predicted by total alcohol, highly consistent estimates were found between ARIMA and panel data analyses, with an average overall effect size of 7% to 9% per liter of ethanol. However, less consistent estimates were derived using spirits, beer, and wine consumption as predictors, suggesting cautious interpretation when beverage-specific volumes are used as predictors, particularly for wine as consumption of this beverage type is low in the US and correlations with total and other beverage type sales may bias estimates. As in past analyses, spirits consumption was found to have the strongest relationship with cirrhosis rates. A further analysis examined racial differences in cirrhosis mortality rates in the Eastern US states and relationships with alcohol sales (Kerr & Ye, 2008), building upon previous work by Herd (1985), which indicated that African Americans experienced dramatically higher cirrhosis mortality than whites in states where a high proportion of the black population lived in urban areas. A positive association between cirrhosis mortality and spirits sales, but not beer or wine, was found for both race groups with larger effect sizes on African American mortality rates. We also showed that the mortality rate differential declined in the 1980’s and 1990’s, until white and African American cirrhosis rates no longer differed, even in states where the largest differences were seen.

Motor Vehicle Accidents

Only one previous US study has address this outcome. Utilzing fixed effects models on a 12-year and 38-state panel Gruenewald et al. (1995b) found that motor vehicle accident (MVA) mortality rates were most strongly related to beer. A later analyses by the same group suggested that the main explanation for this beverage-specific risk was that young people in the US tended to binge drink with beer. (Gruenewald, Johnson, Millar, & Mitchell, 2000) We now present a new analysis of MVA mortality in the US covering a longer time period and more states, which finds significant relationships with both beer and spirits.

Methods

The relationship between MVA mortality rates and beverage-specific alcohol consumption was examined using cross-sectional time-series data of 48 US states or state groups from 1957–2002. MVA mortality rates were taken from the National Center for Health Statistics Compressed Mortality Files series (National Center for Health Statistics, 2000, 2003, 2004) for years 1968–2002 and from the Vital Statistics of the United States for each year from 1950 to 1967 (National Center for Health Statistics, 1961–1969; National Office of Vital Statistics, 1954–1960). Data from 1957 was used in current analysis since a key control variable, vehicle miles traveled, was only available from that year. MVA was defined across several revisions of the ICD using codes E810-E835 for ICD 6 and 7, codes E810-E823 for ICD 8, codes E810-E825 for ICD9 and codes V02-04, V09, V12-V14, V19-V79 and V86-V89 for ICD 10. Data for per capita consumption of ethanol were derived for spirits, beer and wine (in liters of ethanol), described in detail in Kerr et al. (Kerr, et al., 2006). Because of substantial cross-border alcohol purchases between New Hampshire and Massachusetts and between the District of Columbia, Maryland and Virginia, two state groups were created with population averaged variables utilized in each. State sub-groups based on recent drinking levels and patterns are analyzed separately to assess differences in relationships between relatively dry states and relatively wet states. Dry states are defined by higher proportions of abstainers, relatively low population proportions of heavy occasion drinkers and lower per capita consumption, while wet states are defined by high proportions of drinkers and heavy occasion drinkers and high per capita consumption (Kerr, 2010a).

GLS models estimated the alcohol and MVA mortality relationship utilizing cross-sectional and time-series data. GLS is a generalized approach that specifies the variance-covariance matrix of the error structure and allows panel-specific variance (heteroskedasticity) and first-order autoregressive (AR) error terms. The analysis of time series data requires that the series be stationary, in order to avoid obtaining spurious results. (Greene, 2000) Differencing is the preferred method for achieving stationarity (Raffalovich, 1994) and preliminary results found that when first-differenced data was used GLS and fixed-effects models gave similar results, while inconsistent estimates were generated using the raw series.

Results

Table 1 presents coefficient estimates from GLS models of the relationship between alcohol consumption (in liters of pure ethanol) and MVA mortality rates (per 100,000). Consumption of spirits, beer and wine, as well as vehicle miles traveled (not presented), were entered simultaneously in the model. For the combined 48 states (or state groups), spirits consumption significantly predicted MVA mortality, with positive changes in annual spirits consumption being strongly associating with increases in MVA mortality rates. Beer consumption, though having a smaller effect than spirits, was also significantly related with MVA mortality. When data was divided into two separate periods before and after 1980, the year when US total alcohol consumption peaked, similar estimates were found for spirits and beer consumption from 1981–2002 and both were statistically significant. In comparison, none of beverages were significantly related to MVA mortality during the 1957–1980 period, although a positive effect was suggested for spirits.

Table 1.

Coefficients (standard errors) for Spirits, Beer and Wine Consumption from Generalized Least Squares Models of Mortality Rates from Motor Vehicle Accidents. Based on 48 US States (or state groups) from 1957–2002 and Controlling for Vehicle Miles Traveled¹

	Spirits	Beer	Wine
All states (48 states)	1.455 (0.301)***	0.661 (0.294)*	−0.009 (0.667)
1957–1980	0.746 (0.497)	−0.322 (0.506)	−0.572 (1.096)
1981–2002	1.812 (0.348)***	1.806 (0.311)***	−0.461 (0.709)
Dry states (11 states)	0.474 (0.853)	2.060 (0.843)*	1.774 (2.674)
1957–1980	0.058 (1.370)	0.883 (1.592)	6.057 (3.971)
1981–2002	0.869 (1.023)	3.068 (0.828)***	−6.266 (3.159)*
Moderate states (18 states)	1.640 (0.448)***	0.638 (0.406)	0.250 (0.876)
1957–1980	0.965 (0.709)	0.274 (0.709)	−0.398 (1.478)
1981–2002	2.586 (0.623)***	0.988 (0.434)*	−0.525 (1.003)
Wet states (19 states)	1.578 (0.462)**	0.326 (0.504)	−0.831 (1.128)
1957–1980	0.746 (0.830)	−1.253 (0.838)	−1.991 (1.840)
1981–2002	1.696 (0.467)***	2.241 (0.539)***	−0.331 (1.144)

¹GLS model allowing for heteroskedasticity and panel-specific AR(1). Dependent and predictor variables are all transformed using first-order difference. Spirits, beer and wine consumption are entered simultaneously in each model.

^***p<0.001,

^**p<0.01,

^*p<0.05

The same models were then estimated within the groups of Dry, Moderate and Wet states. Using the whole 1957–2002 range, only beer consumption was significantly associated with MVA mortality in Dry states, while only spirits consumption was significantly predictive in Moderate and Wet states. Across all three regions, significant associations of alcohol consumption with MVA mortality were only observed from 1981–2002, with no significant effects seen from 1957–1980, as found for all 48 states combined. The pattern of coefficient sizes was found to differ across the three regions for the 1981–2002 period with a much larger effect from beer in the Dry states and from spirits in the Moderate states, while strong effects from both beverage types were found in the Wet states. Results for wine were generally negative and not significant.

A separate analysis was performed using data only from 1975–1986 for comparison to findings from Gruenewald and Ponicki (1995b) using the same time period. This analysis was not an exact replication, as the earlier analysis was restricted to 38 states and estimated fixed and random effects models on log-transformed levels data for single vehicle nighttime crashes. We found a somewhat stronger relationship for beer (coef 2.14, p<0.001) than for spirits (coef 1.34, p<0.01). Although a much stronger effect of beer was found in the earlier study, our results generally confirm the larger impact of beer during this specific period.

Conclusion

As can be seen in Table 1, both spirits and beer have been found to be significantly related to MVA mortality, with the strength of these relationships varying across state groupings and time periods. While spirits were found to have a larger impact overall, results from the period since 1981 indicate a roughly equal impact of spirits and beer. This more recent period was also associated with the largest impact of beer and spirits on mortality. This larger impact is likely related to declining beer and spirits sales in the context of policy effects from increased minimum drinking age and reductions in the legal BAC limits in a number of states. (Wagenaar, Maldonado-Molina, Ma, Tobler, & Komro, 2007; Wagenaar & Toomey, 2002) We conclude that both beer and spirits have had significant impacts on MVA mortality in the US.

Ishaemic heart Disease (IHD)

While speculation over beverage-specific relationships with heart disease morbidity and mortality has suggested that wine is especially protective against heart disease and related problems, in practice it is quite difficult to separate beverage effects from drinking patterns and the socio-demographic characteristics of drinkers. (Klatsky, Friedman, Armstrong, & Kipp, 2003) Individual-level research distinguishing drinking patterns is becoming more common and suggests that protective effects may be limited to regular drinkers, while intermittant heavy drinking is especialy harmful. (Roerecke & Rehm, 2010) Even studies based only on usual quantity and frequency have been able to identify aspects of this effect. (Breslow & Graubard, 2008) There have now been two studies of IHD mortality at the population level in the US and both indicate beverage-specific associations. First, a time-series analysis of the US utilized both ARIMA and vector error correction models finding a positive (harmful) association with spirits and a negative (protective) association with beer. (Kerr & Ye, 2007a) These analyses were extended to a 53-year panel of US states in a recent analysis focusing on GLS models utilizing first differenced data and distributed lag specifications of both alcohol and tobacco variables. (Kerr, Karriker-Jaffe, Subbaraman, & Ye, 2011) Results indicated both protective effects of alcohol associated with changes in beer and wine consumption and harmful effects associated with changes in spirits consumption. We interpret these countervailing effects as likely being related to different patterns of consumption and argue that the question of the overall effect of per capita alcohol consumption on IHD mortality rates can only be answered through population-level analyses. We found that for all US states combined there was an overall harmful effect indicating an increase in IHD mortality rates of 1% for each liter of ethanol increase in consumption. Analyses by wetness regions (Kerr, 2010a) indicate that the overall balance of protective and harmful effects differs by region, with no significant overall effect found in the Wet and Moderate regions and a significant harmful effect of 2.7% per liter of ethanol found in the Dry region. These findings suggest that the regular light to moderate consumption patterns likely associated with protection from IHD mortality are much less common in the Dry states relative to harmful heavy and binge patterns likely associated with increased IHD mortality risk, which appear to be common throughout the US. Beverage-specific findings also suggest that spirits have been most closely associated with harmful drinking patterns among older Americans.

Head and Neck Cancers

US time-series analyses estimated ARIMA models on data from 1950 to 2002 to examine relationships between per capita alcohol consumption, tobacco use and oral cancers (Kerr & Ye, 2007b). We found significant positive (harmful) effects on oral cancers from cigarettes and total alcohol consumption, with spirits being the only beverage to show a significant positive effect in multivariate models also including beer and wine. The association with spirits may indicate the importance of alcohol concentration or, alternatively, may be linked to birth cohort beverage preferences, drinking patterns or the use of alcohol and tobacco together.

Stomach Cancer

A US time-series analysis of data from 1950–2002 found negative coefficient for spirits (implying protection) and no association with other beverage types with stomach cancer mortality rates. (Kerr & Ye, 2007b) We suggest in the paper that a potential mechanism for such an effect where the expansion of spirits drinking in the 1960’s and 70’s suppressed the activity of Helicobacter pylori bacteria. These bacteria are a major cause of stomach cancers and are known to be negatively affected by alcohol consumption. (Brenner, Rothenbacher, Bode, & Adler, 1999)

Homicide

Parker and Cartmill (Parker & Cartmill, 1998) conducted a US time-series analyses of homicide data from 1934–1995 with attention to beverage-specific effects. Spirits were found to be significantly and positively predictive of white homicide rates while wine was negatively related and no effect was found for beer. Beer, but not wine or spirits, was found to be significantly and positively predictive of non-white homicide rates. The authors conclude that alcohol is an important predictor of homicide rates in the US but do not emphasize beverage-specific relationsips.

Suicide

An ARIMA analysis of US suicide rates (Landberg, 2009) found a significant positive relationship between spirits consumption and suicide mortality for women, but no significant relationship for men. A fixed-effects analysis of a relatively short panel of US states also found an association with spirits only. (Gruenewald, Ponicki, & Mitchell, 1995) Most recently, GLS models on first-differenced data from a 53-year state panel found assocations with spirits for women but with beer and wine for men. Total alcohol was significant in the population model and for both genders using a distributed lag specification for the alcohol measures. The earlier studies had only considered the effects of current alcohol measures. With current measures only, the effect of spirits on women’s mortality was found to be significant. (Kerr, Subbaraman, & Ye, in press) For men, an effect of alcohol was only found using distributed lags, indicating that chronic drinking is the key factor. For women about half of the effect appeared to be related to current drinking and half to chronic, indicating multiple pathways, and only spirits were found to be significant. The prominent effect of lagged measures indicates the importance of the impact of chronic alcohol use on depression, a major cause of suicide, as separate from the effects of intoxication on the decision to attempt suicide and the success rate of attempts. The mixed finding with respect to beverage types is difficult to interpret here, perhaps suggesting that all types are potentially implicated, but again emphasizing that associations with spirits were most commonly found and were the only beverage associated with a contemporaneous effect.

Discussion

US alcohol policy has long favored beer and wine over spirits in terms of taxation, availability and advertising restrictions. Population-level analyses of associations between apparent beverage-specific per capita consumption and mortality rates for several mortality causes appear to support this distinction. Spirits were found to be more strongly related to, or the only beverage type with a significant positive association with, cirrhosis, IHD and head and neck cancer mortality. Spirits sales, along with beer, were also implicated in motor vehicle accident mortality and homicde. All three beverage types were found to be associated with suicide mortality, however, spirits was the only beverage type found to be a significant predictor in models utilizing only current alcohol measures and in three of the four studies reviewed. The harmful effect of spirits on IHD was in contrast to apparent protective effects associated with beer and wine. On the other hand, for stomach cancer, spirits sales at the US population level were found to be negatively associated, suggesting protection. Such an effect is plausable given the known effects of alcohol on the Helicobacter pylori bacteria. Wine in the US makes up only 15% of alcohol sales and wine only drinkers tend to be older and lighter drinkers so it is not surprising that wine would be the beverage least associated with mortality causes.

The US offers unique opportunities for aggregate-level analyses because of good data availablity at the state level and substantial variation in beverage-specific consumption across states and over time. Many beverage-specific analyses have also been conducted for the US, allowing a more detailed picture of associations than is available for any other country. Spirits consumption has been found to be related to aggregate mortality rates in other countries for cirrhosis (Kerr, Fillmore, & Marvy, 2000; Terris, 1967), suicide (Norström, Stickley, & Shibuya, in press; Razvodovsky, 2009b), IHD (Razvodovsky, 2009a) and homicide (Stickley, 2011). However, other beverage types are also implicated. For homicide, a study of from Canada found both beer and spirits to be significantly associated (Mann, Zalcman, Smart, Rush, & Suurvali, 2006) and a study of 14 European countries found significant associations with beer, wine and spirits in different countries (Rossow, 2001). Cirrhosis has also been associated with wine consumption in wine drinking countires like France and Italy and also in the United Kingdom (Mills, 2007).

There are, of course, limitations to inference from aggregate analyses. These population-level studies cannot distinguigh between effects that are truly specific to a beverage type and associations operating through correlations with drinking patterns, demographic groups, birth cohorts or even other risk factors including tobacco or obesity. Interpretations must always consider results from individual-level studies and the complex policy, demographic, economic and generational environment from which the data was generated. Many drinkers do not drink only one beverage complicating the attribution of alcohol-related harms to a specific beverage type. With these cautions, population-level studies can confirm and distinguish between hypotheses generated from individual-level studies and demonstrate population–level effects.

More research is clearly needed to understand potential beverage or %ABV-based differences in alcohol-related harm and the potential for beverage-specific alcohol policies to address these differences. The literature reviewed here has identified spirits sales as being significantly associated with alcohol-related mortality in key areas where beer and wine sales did not have significant associations (cirrhosis and head an neck cancers) or were associated with protective effects (IHD). These associations were found under policy arrangements where spirits were discouraged with relatively higher taxes, lower availability and more restricted marketing opportunities. This suggests that the differential policy treatment found in the US and most other countries is justified on the basis of differential harm relationships and should perhaps be strengthened in the US, particularly through taxes high enough to make the cheapest spirits more expensive than wine or beer. Government control of spirits sales at the wholesale or retail level exists in 19 US states and our results also support this differential treatment.