Differences between socioeconomic status (SES) groups in alcohol-related mortality and hospitalisations are large in Finland1 as well as elsewhere.2^–4 However, the causes of these differences have not been well established, because the differences between SES groups in drinking patterns, as reflected in surveys, are too small to explain the larger differences in mortality and hospitalisations.4^–7 Because alcohol-related mortality is a major cause of premature death, particularly among men and in lower SES groups,8 9 and because reducing the socioeconomic differences in mortality has been recognised as a major goal,10 it is important to find out whether people with low SES actually end up with more alcohol-related deaths than expected on the basis of their consumption and why.

Previously, it has been shown that there were no socioeconomic differences in all-cause mortality following hospitalisation from alcohol-related causes in Finland,6 which rules out differential quality of treatment as a major explanatory factor for the observed discrepancy. Two major alternative hypotheses remain that could explain why socioeconomic differences in severe alcohol-related harm are so much larger than the differences in drinking patterns. First, it is possible that the effects of similar drinking patterns are more harmful in lower socioeconomic groups. Second, the finding could be artefactual, as data on drinking patterns are derived from surveys, which are capable of capturing only the part of the population dutiful enough to respond; this is one reason why typically only 40–70% of total per capita consumption is covered in surveys.11 12 In contrast, mortality and hospitalisation registers cover the whole population, including even skid-row alcoholics, which magnifies estimates of socioeconomic differences. Therefore, it is important to compare drinking patterns and health outcomes across SES groups using the same data source.

Our aim is to examine whether differential drinking patterns between SES groups explain the observed differences in alcohol-related mortality and hospitalisations within a cohort, ie using one common data source. Our study population included Finnish people who participated in cross-sectional surveys on drinking habits in 1969, 1976 or 1984 and who were followed up for mortality and hospitalisations. Because the baseline data come from a drinking habit survey, they include good measures of patterns of drinking. There are two alternative hypotheses that suggest two different types of analysis which address the same question from two different angles.

• Hypothesis 1. Effects of drinking on alcohol-related mortality and hospitalisations are similar irrespective of SES. In this case, estimated differences in alcohol-related mortality across SES groups should disappear when adjusting for total consumption and drinking patterns. In order to assess whether the findings are specific to SES groups, the same is checked for gender: do differences between genders disappear when controlling for drinking patterns?

• Hypothesis 2. Effects of drinking are different in different SES groups. This indicates effect modification: the estimated effect of a given amount and pattern of drinking is greater in some SES groups than others.

The data at hand are not optimal for studying mediating mechanisms, but the effect of differences in marital status on the socioeconomic differences in alcohol-related outcomes can and will be studied.

METHOD

Subjects

The baseline data come from cross-sectional Finnish Drinking Habits Surveys of Finns aged 15-69, conducted in September 1969, 1976 and 1984 using face-to-face interviews. In the first period, females were undersampled; otherwise, the samples were representative of the general Finnish population. The number of respondents was 1740 in 1969 (91% of the original sample), 2835 in 1976 (response rate 96%) and 3624 in 1984 (response rate 94%) – a total of 8199. Details of the samples are available elsewhere.13 14 We excluded individuals with missing information on alcohol measures (n = 70) or follow-up (n = 14). Young people (<25 years) were also excluded, because SES among youth is not stabilised. The remaining sample size was 6406.

Follow-up

Mortality data were obtained through the cause-of-death register of Statistics Finland, and the data on hospital admissions were obtained from the hospital discharge register of the National Research and Development Centre for Welfare and Health, STAKES. The ethical research committee of STAKES has reviewed and accepted the study design. The linkage to the pooled Drinking Habit Survey data was carried out using personal identification numbers. The respondents were followed up for 16.3 years beginning at each baseline survey, until death or until the first hospital admission for the cause under consideration, whichever came first.

Causes of death and hospitalisations were classified using the Finnish versions of the International Classification of Diseases, FICD, 8th revision for 1969 to 1986; 9th revision for 1987–1995; and 10th revision for 1996–2000. Follow-up for deaths can be considered 100% complete. As alcohol-related causes, we selected the following when they appeared as either the contributory (available for FICD 9th and 10th revisions) or the underlying causes of death or the main diagnosis at hospitalisation: alcoholic psychosis, alcoholism and alcohol abuse (FICD8: 291, 303; FICD9: 291, 303, 3050A; FICD10: F10), diseases of the liver (FICD8: 571), alcoholic diseases of the liver (FICD9: 5710A–5713X, ICD10: K70), alcoholic diseases of the pancreas (FICD9: 5770D–5770F, 5771C–D; FICD10: K860), alcoholic beriberi disease (FICD9: 2650A), alcoholic polyneuropathy, degeneration of the nervous system due to alcohol and alcoholic myopathy (FICD9: 3575A; FICD10: G621, G312, G721), alcoholic cardiomyopathy (FICD9: 4255A; FICD10: I426), alcoholic gastritis (FICD9: 5353A; FICD10: K292), toxic effects of alcohol and accidental poisoning by alcohol (FICD8: 980, E860; FICD9: 980, E851A; FICD10: T51, X45). In the follow-up time, there were 180 such alcohol outcomes, of which 22 were deaths.

Measurement

Total annual consumption was calculated on the basis of the amounts consumed on each of the drinking occasions within a period of time preceding the interview that varied between 1 week and 12 months (“survey period”). The length of the period was chosen so that, with the given drinking frequency, this period should theoretically cover four drinking occasions (for details, see Mäkelä 197115). The annual estimate was obtained by extrapolation, ie multiplying the period estimate by a constant to extend the period to 12 months. The coverage rate, ie the ratio between the average consumption of the respondents and total per capita consumption (registered sales + estimated unregistered consumption) was approximately 50% in 1969 and 1984 and a good 60% in 1976.

Drinking patterns were measured in three ways. First, we transformed the survey period data into measures of the annual frequency (in five categories) of drinking 1–4, 5–7, 8–12 and 13 or more drinks. These 20 dummy variables together roughly correspond to a total consumption measure in the sense that, by adding up the products of frequencies and quantities, an estimate of total consumption could be derived, but they additionally account for the pattern of drinking.

The second way of measuring drinking patterns was that we divided the total annual consumption into volume of consumption that came from heavy episodic drinking (HED; estimated blood alcohol concentration (BAC) more than 1%) and the volume of consumption that came from non-HED occasions. BAC was estimated for each drinking occasion on the basis of the amount of alcohol drunk, duration of drinking and weight.14

The third measure of drinking patterns was a subjective measure of intoxication frequency (“How often do you drink enough really to feel it?”). This measure was only available for the baseline years 1969 and 1984.

Total consumption was categorised into approximate quartiles, with the highest category split into two, and with further two categories for abstainers: (1) lifetime abstainers; (2) current abstainers; (3) 1–26; (4) 27–116; (5) 117–364; (6) 365–999; and (7) more than 1000 centilitres of 100% alcohol per year. HED and non-HED volume were categorised using these same categories.

SES was based on self-reported occupation and divided into manual workers, non-manual workers and a heterogeneous category of “others” (farmers, entrepreneurs, unknown occupation). We focus on manual and non-manual workers. Sex, period and age (25–34, 35–44, 45–54 and 55–69 years) were considered possible confounders in the analyses. Marital status was divided into married or cohabiting; single; divorced or widowed.

Statistical analyses

The Cox proportional hazard model was used in analysing survival time data.16 Results are expressed in terms of hazard ratios (HRs). The assumption of proportional hazards was examined visually for different alcohol consumption categories. No violations in the proportional hazard assumption were observed.

The models are only shown for men and women combined, because there were no significant interactions between gender and the SES measures, or between gender and drinking patterns, on alcohol-related mortality and hospitalisations. This may be due to the small number of alcohol-related outcomes among women.

RESULTS

The differences between manual and non-manual workers’ drinking patterns were typically not large (table 1), particularly when allowing for the fact that men made up a larger proportion of manual workers than of non-manual workers: some of the differences in drinking patterns between manual and non-manual workers in table 1 disappeared when stratified by gender (not shown). The types of heavy drinking that were more common in the lower SES categories were also the most hazardous ones (table 1, last column).

After controlling for sex, age and period, the hazard of an alcohol-related death or hospitalisation was 2.06-fold among manual workers compared with non-manual workers in the three cohorts combined (table 2/model 1); without control for sociodemographic factors, the hazard ratio was 2.35. When total consumption was additionally adjusted for, the ratio did not decrease but increased (table 2/model 2). This result indicates that, if the distribution of annual consumption among the SES groups was similar, the differences in alcohol-related hospitalisations and mortality would be even higher than they are now.

In models 3 and 4, the two measures of drinking patterns available for all three cohorts were adjusted for. Frequency of intoxication was adjusted for in model 5 and compared with models 1 and 2 using data only from years 1969 and 1984 when intoxication frequency was available. For the explanation of SES differences, the biggest impact came from adjusting for total volume of consumption as divided into 20 different dichotomous variables describing the frequency of drinking different quantities (1–4,…, 13+ drinks). However, even after adjustment for this factor, clear differences remained between the SES groups in alcohol-related outcomes (HR 1.91).

Model 6 shows the effect of controlling for marital status, compared with model 3: the excess hazard among manual workers decreased by 14%. This effect varied between 8% and 17% for models 2, 4 and 5, and it was 14% on average. For the “other” social status category, the effect of controlling for marital status on excess risk was 3% on average.

In order to put the results obtained into context, we looked at what happened to gender differences in alcohol-related outcomes, ie to the main effect of gender, after adjustment for drinking patterns. The original HR for men was 3.8 (95% CI 2.6 to 5.5). It diminished considerably after controlling for total consumption (HR 1.56, 95% CI 1.0 to 2.4), and it effectively disappeared when controlling for the frequencies of different amounts per occasion (HR 1.2, 95% CI 0.8 to 1.9). The original HR also decreased considerably when controlling for HED and non-HED volume (HR 1.7, 95% CI 1.1 to 2.7), although not more than when controlling for total consumption only.

Figure 1 illustrates the estimated impact on alcohol-related outcomes of total consumption and of volume of consumption that is drunk in heavy drinking occasions, separately for manual and non-manual workers. The estimates were derived from interaction models, where the interaction terms between consumption and SES were non-significant. In a multiplicative model, this means that the effects of drinking and SES were not more than multiplicative but, as can be seen from figure 1, the additive or absolute effects of a high volume of drinking and a high heavy drinking volume were much greater among non-manual workers. Controlling for marital status diminished the excess mortality of manual workers in each consumption category, but especially in the highest categories, where 17–19% of the excess hazard of manual workers was attributable to marital status.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Risk of alcohol-related outcomes by socioeconomic group in (A) categories of total consumption and (B) in categories of the volume of consumption that was drunk in heavy drinking occasions. Abstainers were excluded from both models. (a) From a model that included as predictor variables: background variables (period, age, gender), SES, total consumption (in five categories, I = 1–26 cl) and interaction (non-significant) between SES and total consumption (in three categories: because of insufficient number of cases, a common interaction estimate was derived for categories I–III). (b) From a model that included as predictor variables the background variables, SES, total consumption divided into volume drunk in heavy and non-heavy drinking occasions and categorised (five categories, I = 0–26 cl), and the interaction (non-significant) between socioeconomic status and categorised heavy drinking volume (in four categories: because of insufficient number of cases, a common interaction estimate was derived for categories II and III).

DISCUSSION

We set out to study the relationship between SES, drinking patterns and severe alcohol-related outcomes in order to shed more light on the previously recognised puzzle of why individuals of low SES suffer more harmful consequences of drinking than would be expected on the basis of their drinking.4^–6 The difference in alcohol-related mortality between manual and non-manual workers in this study was somewhat smaller than observed earlier.1 Dissimilarity in occupational categorisation (self-reports versus register-based reports) is one possible reason for this. Additionally, part of the population was not represented in our study, either because they were not included in the sampling frame (the homeless and the institutionalised) or because they chose not to respond. The former group is especially likely to have low SES and a high rate of alcohol-related problems. Register-based data, which include these population groups, can therefore yield higher estimates of differences by SES in alcohol-related mortality and hospitalisations. In any case, the difference seen in this study was sufficient in that it allowed us to examine whether it could be attributed to differences in drinking.

According to our results, controlling for total alcohol consumption and drinking patterns diminished the difference between manual and non-manual workers in alcohol-related mortality and hospitalisations only marginally, ie the observed difference was only marginally accounted for by differences in drinking patterns. This does not necessarily mean that there would not be any socioeconomic differences in drinking patterns. They were similar to those noted in earlier studies.7 17 However, these differences were generally small and not sufficient to account for the difference in adverse effects.

The corresponding result was seen from models depicting the effects of drinking on alcohol-related hospitalisation and mortality separately in the different SES groups, ie in analyses of effect modification. Even if the proportional, or multiplicative, effect of drinking on mortality and hospitalisations did not differ between manual and non-manual workers, in absolute terms, heavy drinking increased the risk of severe outcomes more among manual workers. Further, the risks were clearly higher among manual than among non-manual workers in all levels of total consumption and in all the different levels of volume drunk in heavy drinking occasions.

Some limitations of the study should be noted. First, even if the data included a rich set of measures of drinking patterns, socioeconomic differences in some dimensions of drinking may have been left unmeasured. For example, it is possible that drinking in binges, ie for several days in a row, is important in predicting severe alcohol-related harm and differs by SES. Additionally, the study subjects’ alcohol consumption was recorded at one time point only, and the follow-up time was 16 years. Many people change their drinking patterns in this time,18 and the changes may be differential by SES. Also, underreporting of drinking is widespread in surveys11 12 and, theoretically, respondents with low SES could underreport their drinking to a larger extent. Also, even if the coding of causes of death in Finland is relatively reliable, there could be a bias against recording an alcohol-related code for individuals of higher SES.9

Further, two of the three drinking pattern measures were based on the so-called survey period measure, which is based on a time frame varying from only one week, in the case of frequent drinkers, to one year. For some drinkers, the behaviour reported was probably atypical, but a systematic bias by SES is unlikely. However, the measurement error could make it less likely that the adjustment for drinking patterns would successfully make the estimated differences between SES groups disappear, even in the case where this should happen. In order to assess whether measurement error could be too great to invalidate the study design, we looked at the gender difference in alcohol-related outcomes. According to these results, the estimated gender difference disappeared after adjusting for drinking patterns. This increases our confidence that drinking patterns were sufficiently well measured for our purposes.

So we can conclude that the small differences between SES groups in drinking patterns did not suffice to explain the larger differences in mortality and hospitalisation, even when the results on drinking and the consequences were now based on the same set of respondents. This indicates that the consequences of similar drinking patterns are more severe for individuals of lower SES. According to previous studies, this may be true for other risk factors as well, eg for the effect of job strain on blood pressure.19 Also, in Lantz et al’s study,20 controlling for various health behaviours only explained a small part of the socioeconomic differences in overall mortality. Similarly, Pampel and Rogers21 showed with US data that individuals of low SES were more vulnerable to the effects of smoking, inactivity, stress, drinking and body mass index on ill-health, as measured by morbidity and self-rated ill-health in particular. However, vulnerability to the effects of drinking on mortality could not be shown. A major difference between their study and the current study is that we used alcohol-related outcomes, whereas they used overall mortality, a large part of which is from cardiovascular deaths that have a curvilinear relation to alcohol consumption.

Our data were not optimal for explaining the mechanisms of the differential vulnerability to the effects of alcohol. We present some hypotheses, but future studies are needed to investigate their explanatory power. A potentially important explanation relates to social support. Men of higher SES are more often married and have a family,22 and wives are an important agent of social control of excessive drinking.23 For example, in Finland, alcohol-related deaths were found to be 2.6–5.7 times more common among never married, divorced and widowed compared with married men.24 Our data showed that 14% of the excess risk among manual workers compared with non-manual workers could be accounted for by differences in marital status. In addition to social support from a partner, support from employers may play a role, if they are ready to invest more energy in enduring or solving the alcohol problems of higher SES workers.

It can also be assumed that individuals of higher SES have better resources to protect them from the hazards of drinking. They may, for example, be able to choose to drink in safer environments or to take a taxi home instead of driving. Differences in treatment are, however, not likely to be an important explanation for the greater vulnerability of the lower SES groups in Finland, where health care is universally available. Empirically, it has been shown that the difference in survival after hospitalisation was not a critical factor in the greater alcohol-related mortality of manual workers in Finland.6 Similarly, results on differences in outcomes of treatment for alcohol problems have shown that lower SES groups do not fare badly in comparison with higher SES groups after treatment (for a review, see Mäkelä et al6), and also the majority of those treated often come from marginalised groups.25

Childhood environment and accumulation of risk factors from early on may also affect the greater vulnerability of individuals of low SES: low childhood SES has been found to increase the risk of alcohol-related deaths even when own SES has been controlled for,26^–28 although for alcohol use or harmful use, similar intergenerational effects have not been found.29 Additionally, the various predictors of alcohol problems and of other health problems that tend to be more prevalent in the lower social strata20 may interact and enforce each others’ effects to result in higher rates of mortality. One such example of interaction is Hemmingsson and Lundberg’s30 finding that heavy use of alcohol in late adolescence interacted with later low work control in relation to alcoholism. By identifying the central factors by which some groups escape the consequences of drinking better than others, we will be one step closer to being able to help those who cannot escape the consequences of drinking by themselves.