OUP user menu

Alcohol, Mortality and Cardiovascular Events in a 35 Year Follow-up of a Nationwide Representative Cohort of 50,000 Swedish Conscripts up to Age 55

A. Romelsjö, P. Allebeck, S. Andréasson, A. Leifman
DOI: http://dx.doi.org/10.1093/alcalc/ags021 322-327 First published online: 1 March 2012


Aims: To assess the association between drinking patterns and mortality, and cardiovascular disease in a large cohort of young- and middle-aged men and to assess whether the net balance of harm and protective effect implies protective effect or not. Methods: Information from health examinations, psychological assessments and alcohol use background in a nationally representative birth cohort of 49,411 male military conscripts aged 18–20 years in 1969/1970, were linked to mortality and hospitalization data through 2004. Cox regression analyses were conducted and attributable proportions (APs) calculated. Confounders (baseline social status, intelligence, personality and smoking) were taken into account. Results: Increasing alcohol use was associated with increasing mortality (2614 deceased) and with decreasing risk for myocardial infarction (MI). The hazard ratio (HR) for mortality was 1.42 [95% confidence interval (CI) 1.10–1.82] with a consumption corresponding to 30 g 100% ethanol/day or more in multivariate analysis. The risk for non-fatal MI was significantly reduced at 60 g 100% ethanol/day (HR 0.37, 95% CI 0.15–0.90), not reduced for fatal MI, and non-significantly reduced for total MI. There was a marked association between alcohol use at conscription and mortality and hospitalization with alcohol-related diagnosis. APs indicate that alcohol caused 420 deaths, 61 cases of non-fatal stroke and protected from 154 cases on non-fatal MI. Conclusion: Many more deaths were caused by alcohol than cases of non-fatal MI prevented. From a strict health perspective, we find no support for alcohol use in men below 55 years.


Most studies indicate that high alcohol consumption and binge drinking is associated with increased total mortality and low-to-moderate alcohol consumption with reduced mortality in ischaemic heart disease (IHD) (Corraro et al., 2000; Ronksley et al., 2011). This has been attributed to an increased level of high-density lipoprotein and decreased level of low-density lipoprotein cholesterol, a decrease in the aggregation of thrombocytes (platelets) and a positive impact on the fibrinogen level from alcohol (O'Brien et al., 2011). The current knowledge has been summarized in two papers, which however do not address the possibility of different associations related to age (Ronksley et al., 2011; O'Brien et al., 2011).

A possible preventive effect of alcohol may not start until the age of ∼40–50 years in men because below that age IHD is uncommon and the atherosclerosis process has usually not progressed far (Tolstrup and Grönbeck, 2007; Hvitfeldt et al., 2010). However, there are few studies on the association between alcohol and IHD before the age of ∼55 years. In a follow-up to the age of 45 years of the cohort in this study, Romelsjö and Leifman (1999) found that alcohol consumption over 15 g 100% ethanol/day was related to significantly elevated total mortality in multivariate analysis and with a non-significantly reduced myocardial infarction (MI) incidence. Hvitfeldt et al. (2010) found, in a pooled follow-up analysis of almost 200,000 women and 75,000 men, a negative association between alcohol and coronary heart disease (CHD) incidence but that the reduction was smaller in men aged 39–50 years. White et al. (2002, 2004) found in a study with aggregate data that alcohol caused 2663 non-IHD deaths and protected from 898 IHD deaths in men aged 16–54 in Wales and England in 1997, i.e. an overall negative effect up to the age of 55, but they lacked data on IHD incidence. This warranted a further 10 year follow-up, compared with the previous study, now including many more cases of myocardial infarction, and also analysing drinking patterns (White et al., 2002). There is scientific support for increased risk for haemorrhagic stroke with increases alcohol consumption, and for reduced risk with low-to-moderate consumption, but the net effect in these ages is unknown (Ronksley et al., 2011).

The scientific literature on alcohol, IHD and mortality is not consistent. A few studies even report a decreasing IHD risk with increasing alcohol consumption (McElduff and Dobson, 1997; Ariola et al., 2009).

Variation in drinking patterns may explain differences in estimates between studies. In one meta-analysis, binge drinking was associated with a higher IHD risk than other alcohol use, and another found that irregular heavy drinking occasions were associated with a significantly increased IHD risk of 1.45 compared with regular drinking up to 60 g 100% alcohol/day; however, with significant between-study heterogeneity (Bagnardi et al., 2008; Roerecke and Rehm, 2011). An increased risk of IHD with binge drinking might be explained by increased coronary calcification, episodic high blood pressure and adverse changes in the balance of fibrinolytic factors among binge drinkers (Puddey et al., 1999; Pletcher et al., 2005).

The first objective in this study was to assess the association between alcohol use, including heavy drinking and mortality and cardiovascular events in a nationwide large cohort of Swedish men up to the age of 55 years. The second was to assess the population risk for mortality and cardiovascular events to find out whether the protective or harmful effect predominates.


This study includes all 49,411 Swedish males, born 1949–1951 who were conscripted for compulsory military service from 1 July 1969 to 1930 June 1970. The study was approved by the Stockholm Regional Ethical Review Board (Dnr 2007/174-31 and Dnr 2008/1086-31/5).

At conscription, the men were asked to answer two non-anonymous voluntary questionnaires about alcohol use, smoking, and social and psychological factors, which 98% did. Besides the measures of alcohol use, the confounders were selected from these data, based on the scientific literature and earlier studies of this cohort (Andreasson et al., 1988; Romelsjö and Leifman, 1999). The conscripts were informed that the answers would not impact on their military placement. They also underwent examinations by health professionals and psychologists. The data were linked to the cause of death register and to the national inpatient register, maintained by the National Board of Health and Welfare, up to 31 December 2004.


Alcohol (0 = Reference group).The consumption level was calculated from answers to questions about quantity and frequency of beer, wine and spirits consumption converted to 100% alcohol per day based on standard estimates of drink sizes (10–12 g 100% ethanol). We also constructed a crude measure of binge drinking based on these questions, corresponding to intake ever of 4752 and 65 g 100% ethanol for spirits [then the dominating alcohol beverage in Sweden: CAN (2010)], beer and wine respectively in a session (reflecting Swedish ‘measures’). The consumption alternatives were 0.1–10, 20–30, 30–60 and >60 g 100% ethanol, common categories in alcohol epidemiology. The proportion reporting binge drinking ever for the consumption alternative 0.1–10 g was 38.7%, while the percentage was 84.1% for 10–20 g.

Smoking: Reference group 0 = no smoking, 1 = 1–10 cigarettes, 2 ≥ 10 cigarettes/day.

Father's occupation: Reference group = 0 social classes I+II vs. 1 = social class III. Social class I includes proprietors, private entrepreneurs and high salaried employees in private and public sectors, social class II other salaries and small entrepreneurs, and social class III mainly blue-collar workers (Upmark et al., 1999).

Parents divorced: 0 = no, 1 = yes.

Run away from home: 0 = never, 1 = Once or more often.

Truancy: 0 = no, 1 = Once or more often.

Self-assessed health: Reference group 0 = good or very good health, 1 = fair health, 2 = poor or very poor health.

Emotional control: Reference group 0 = fair or higher emotional control vs. 1 = rather low or low emotional control. Emotional control was assessed by psychologists on a 5-point scale representing a standard distribution of 7, 24, 38, 24 and 7% with regularly check of the reliability of the assessments.

Low social maturity: Reference group 0 = fair or higher social maturity vs. 1 = rather low or low social maturity. Assessed by psychologists (Upmark et al. 1999).

Intellectual ability (IQ): Reference group 0 = ≥7 high IQ), 1 (moderate IQ) = 4–6, 2 (low) = 1–3. This represents a general intellectual ability value transformed from results on a 9-point Stanine-scale from tests measuring verbal, logic, inductive and technical abilities (Stenbacka, 2000).

Body mass index: Reference was body mass index (BMI) 20–24.9, and the other two categories were BMI less than 20, 25–29.9, 30+.

Blood pressure: Diastolic blood pressure >90 mmHg and systolic blood pressure >140 mmHg were analysed.


Information about overall mortality and specific causes of death was obtained from the National Cause of Death register (Statistics Sweden, 2007). Information about hospitalizations with alcohol-related diagnoses (all diagnoses with the word alcohol), and myocardial infarction was obtained from the National Swedish inpatient register that has been operating with virtually full coverage of all hospital episodes in the country since the middle of the 1970s. Myocardial infarction (code 410 ICD-9) makes up ∼90% of all cases of CHD (ICD 9 code 410-414) in Sweden, and is a more precise diagnosis than CHD or IHD. The ICD 9-codes for stroke are 430-438. The diagnosis myocardial infarction has been found to be accurate in ∼85–95% in different assessments of Swedish data (The National Board of Health and Welfare, 2009).

Statistical analyses

Person-time was counted from 1 October until time for death for those three subjects who died July–December 1969, and from 1 January 1970 until death/hospitalization or until 31 December 2004 for all other subjects. The analyses were based on 1,605, 621 person-years. The hazard ratio (HR) for death or hospital admission was first calculated with bivariate Cox proportional hazard regression with 95% confidence interval (CI) for all individuals with information about alcohol use (n = 48,716). The statistically significant variables (almost all) were included in multivariate analyses. The influence of binge drinking, BMI and blood pressure were analysed in separate models. Attributable proportions (APs) were calculated (Rothman et al., 2008).Trend-tests were performed of HR for alcohol-related diagnoses to assess predictive validity of self-reported alcohol consumption.


Mean reported alcohol consumption corresponded to 4.4 l 100% ethanol per year, slightly <10 g 100% ethanol/day.

Table 1 shows that the risk of mortality in bivariate analysis was higher among alcohol consumers and increased with increasing consumption except for tumours. Table 2 shows that HR was significantly increased for non-fatal and total MI at a consumption of 10–60 g 100% ethanol/day, and for fatal MI at 60+ g/day. The mean age both for fatal and non-fatal MI cases was ∼47 years. For a consumption corresponding to 10 g 100% ethanol/day or more the HR was 1.60 (95% CI 1.33–1.91) for total mortality. The association with consumption level was most conspicuous for alcohol-related diagnoses, with P = 0.0001 for mortality and P = 0.0003 for hospitalization in trend-tests (Table 3).

View this table:
Table 1.

The risk of overall death and certain cause-specific deaths, 1970–2004, at increasing consumption levels (g 100% ethanol/day) reported at conscription 1969/1970 among 48,716 conscripts

Alc. cons (g/day)00.1–1010–3030–6060+
Number and proportions (%)3057 (6.3%)30,089 (61.8%)13,457 (27.6%)1500 (3.1%)613 (1.2%)
All deaths (n = 2611)1331411823141103
HR (95% CI)1.001.08 (0.90–1.29)1.42 (1.18–1.70)2.21 (1.74–2.80)4.15 (3.20–5.37)
Stroke (n = 72) n3392532
HR (95% CI)1.001.32 (0.41–4.27)1.91 (0.58–6.34)2.10 (0.42–10.40)3.62 (0.61–21.68)
Tumour (n = 554) n353371512110
HR (95% CI)1.000.99 (069–1.39)0.99 (0.69–1.43)1.26 (0.73–2.16)1.55 (0.77–3.14)
Injury (n = 470) n182551562219
HR (95% CI)1.001.44 (0.89–2.32)1.98 (1.22–3.23)2.52 (1.35–4.70)5.51 (2.89–10.50)
Suicide (= 454) n232641282811
HR (95% CI) Undetermined death (presumed suicide)1.001.17 (0.76–1.79)1.27 (0.82–1.98)2.52 (1.45–4.37)2.52 (1.23–5.16)
(n = 107) n3484295
HR (95% CI)1.001.63 (0.51–5.22)3.20 (0.99–10.30)6.21 (1.68–22.94)8.77 (2.10–36.72)
Alcohol diagnoses (n = 210) n889721823
HR (95% CI)1.001.13 (0.55–2.33)2.07 (0.995–4.29)4.71 (2.05–10.83)15.54 (6.95–34.75)
  • Number and Cox regression bivariate analysis with HR and 95% CI.

View this table:
Table 2.

The risk of overall and fatal, non-fatal and total myocardial infarction 1970–2004 at increasing consumption levels (g 100% ethanol/day) reported at conscription 1969/1970 among 48,716 conscripts

Alc. cons. g/day0 (n = 3057)0.1–10 (= 30,095)10–30 (= 13,457)30–60 (= 1500)60+ (= 600)
Myocardial infarction, fatal (= 135) n4754655
HR (95% CI)1.001.91 (0.70–5.21)2.64 (0.95–7.34)2.63 (0.71–9.80)8.86 (1.84–24.55)
Myocardial infarction non-fatal (= 862) n37506279337
HR (95% CI)1.001.39 (0.999–1.947)1.74 (1.23–2.45)1.88 (1.18–3.01)1.04 (0.47–2.34)
Myocardial infarction, total (= 997) n415813253812
HR (95% CI)1.001.44 (1.05–1.98)1.83 (1.32–2.53)1.96 (1.26–3.05)1.61 (0.85–3.06)
  • Number and Cox regression bivariate analysis with HR and 95% CI.

View this table:
Table 3.

The proportion (%) and numbers of hospitalizations (= 2204) and deaths (= 215) with an alcohol diagnosis during follow-up among 48, 716 conscripts with different levels of alcohol consumption at conscription

Alcohol consumption (g 100% ethanol/day), numbers and proportionsProportion (%) hospitalized with alcohol diagnosisaProportion (%) deceased with alcohol diagnosisa
Alc. cons.NumbersProportions (%)NumbersProportions (%)NumbersProportions (%)
  • aTrend-test P = 0.0003 for hospitalization and 0.0001 for mortality.

Separate analyses of the earlier half of the follow-up period 1970–1987 and the latter period 1988–2004 showed very similar HRs at different levels of alcohol use for different causes of death. There were only eight deaths from MI during 1970–1987; so HRs for 1988–2004 were almost the same as for 1970–2004.

Those reporting binge drinking in the consumption interval 0.1–20 g 100% ethanol/day had a non-significant higher risk to die or to be hospitalized (Table 4), compared with other alcohol consumers. The results were similar in multivariate analyses.

View this table:
Table 4.

The risk of overall and cause-specific mortality and non-fatal and total myocardial infarction 1970–2004 for those with and without binge drinkinga among those with an average consumption of 0.1–20 g 100% ethanol/day and those with a higher average alcohol consumption (= 43, 795)b

0.1–20 g/day, no binge drinking0.1–20 g/day, binge drinking
Total (= 21,67 1,902)1.05 (0.88–1.26)1.25 (1.04–1.49)
Stroke (= 6352)1.22 (0.37–4.04)1.79 (0.55–5.79)
Tumour (= 492,443)0.99 (0.65–1.32)1.04 (0.73–1.49)
Injury (= 388,343)1.34 (0.83–2.19)1.74 (1.08–2.82)
Suicide (= 383, 333)1.11 (0.72–1.71)1.25 (0.81–1.92)
Unclear suicide (= 8772)1.42 (0.43–4.67)2.77 (0.87–8.86)
Alcohol diagnoses (= 138,119)1.16 (0.56–2.13)1.28 (0.62–2.67)
Fatal myocardial infarction (= 118, 104)2.06 (0.74–5.07)2.23 (0.81–6.53)
All myocardial infarction (= 884, 778)1.39 (1.01–1.92)1.71 (1.24–2.35)
  • aBinge drinking corresponds to a consumption of ∼52 g 100% ethanol in a day for beer, to 65 g for wine and to 47 g for spirits.

  • bCox regression analysis with HR and 95% CI in bivariate analyses.

Table 5 shows that increasing alcohol consumption was associated with increasing total and fatal MI mortality and with decreasing non-fatal and total MI after adjustment for other variables. The cause-specific mortality increased with alcohol consumption, except for tumours. The association was statistically significant for total mortality at consumption over 30 g/day (HR = 1.42, 95% CI 1.10–1.82). HR was significantly reduced for non-fatal MI at consumption over 60 g/day and was 0.62 (95% CI 0.31–1.24) for total MI. The other variables, except fair/poor health, were significantly associated with total mortality, and smoking, blue-collar worker father and low IQ with the MI outcomes. Models also including BMI (model 2) as confounder and separately also blood pressure, which mainly is a mediator, but which also can be seen as confounder to some extent (Roerecke and Rehm, 2011), yielded fewer subjects and cases, but the HRs for alcohol were very similar to the main model (see Table 5), which had very little attrition. HR for MI mortality was 2.17 (95% CI 1.28–3.68) for diastolic blood pressure over 90 mmHg and 2.78 (95% CI 1.62–5.07) for systolic blood pressure over 140 mmHg, and was also significantly elevated for MI. Separate analyses showed that smoking definitively was the most important confounder.

View this table:
Table 5.

The association between alcohol consumption (g 100% alcohol) and total mortality and myocardial infarction mortality and non-fatal and total myocardial infarction among military conscripts (= 43, 583) (model 1) followed from the age of 18–20 years to the age of 52–54 years (1970–2004)

Total Mortality (= 2 279)Myocardial infarction mortality (= 121)Myocardial infarction, non-fatal (= 756)Myocardial infarction total (= 877)
HR95% CIHR95% CIHR95% CIHR95% CI
aAlc 0.1–10 g0.9960.81–1.221.400.43–4.540.880.62–1.260.930.66–1.30
Alc 10–30 g1.080.87–1.331530.46–5.130.830.57–1.200.880.62–1.26
Alc 30–60 g1.220.92–1.601.400.32–6.100.730.43–1.220.780.48–1.27
Alc 60+ g1.891.40–2.563.470.78–15.380.370.15–0.900.620.31–1.24
Smoking 1–10 cigarettes/day1.341.20–1.492.021.19–5.512.001.64–2.442.001.66–2.41
Smoking 11+ cigarettes/day1.551.38–1.733.271.94–5.512.942.40–3.592.982.47–3.60
Father, blue-collar worker1.131.04–1.241.531.04––1.391.241.08–1.42
Divorced parents1.251.11–1.411.180.68–2.040.810.63–1.050.870.69–1.09
Run away from home1.351.14–1.610.660.24–1.861.100.77–1.581.060.76–1.48
Low emotional control1.211.10–1.340.980.64–1.521.120.94–1.331.100.94–1.29
Low social maturity1.321.19–1.460.960.60–1.541.100.91–1.321.090.92–1.29
Moderate IQ1.291.17–1.441.330.83––1.401.191.01–1.41
Low IQ1.561.37–1.771.550.88–2.711.621.31–2.101.601.31–1.96
Fair health1.080.97–1.210.830.30–2.300.900.62–1.310.890.62–1.26
Poor health1.130.94–1.361.370.86–2.190.980.80––1.24
bBMI 25–301.391.17–1.642.801.65–4.742.421.92–3.062.542.05–3.14
BMI >302.541.84–3.514.921.95–12.403.221.92–5.403.662.34–5.74
  • In model 2, BMI is also included: multivariate analysis, HR and 95% CIs. Reference group is abstainers.

  • aModel 1 with adjustment for the other variables in the table.

  • bModel 2 with adjustment also for BMI (= 36,209, 857 deaths, 101 fatal and 630 non-fatal MI cases).

The AP (Rothman et al., 2008) was 18.4% for mortality, 20.4% for non-fatal MI and 12.7% for non-fatal stroke in multivariate analysis. Thus, alcohol may have contributed to 420 of 2279 deaths, to 62 cases of 491 non-fatal stroke and prevented 154 cases of non-fatal MI.


The main finding was that the level of alcohol consumption in late adolescence was associated with increasing mortality and decreasing risk for non-fatal and total MI. This association was statistically significant for total mortality at a consumption of 30 g 100% ethanol/day or more, mainly due to increasing risk over 60 g, and for non-fatal MI at a consumption of 60 g/day or more, in multivariate analysis. Calculation of APs showed that alcohol may have caused 420 deaths and 62 cases of non-fatal stroke, and prevented 154 cases of non-fatal MI. There was somewhat higher risk among those who reported binge drinking (Table 3), consistent with other reports. A closer analysis was hampered by a lack of information on frequency of binge drinking, and as the analyses had to be limited to the drinking range 0.1–20 g ethanol/day (Bagnardi et al., 2008; Roerecke and Rehm, 2011).The increasing risk of mortality overall and in different diagnoses with increasing alcohol consumption is consistent with other studies of usually elderly people and with shorter follow-up periods (Corraro et al., 1999; Rehm et al., 2003). There is some attrition in the multivariate analyses (Table 5) and in the analyses of binge drinking (Table 4), but the results are very similar in bivariate analyses with these reduced samples.

A limitation is that we lack information about alcohol consumption and other factors during follow-up. There are few studies on the stability of alcohol consumption over time from the age of 18–20 years. A positive, sometimes significant relationship between adolescent and adult alcohol consumption has found support in longitudinal studies (Fillmore, 1988; Pape and Hammer, 1996). HR for different levels of alcohol use was clearly related to alcohol-related hospitalization (P = 0.0003 in trend-test) and mortality (P = 0.0001) during follow-up and was similar for the earlier and latter half of the long follow-up period. Moreover, Swedish surveys show that self-reported mean annual alcohol consumption among a random sample of Swedish men aged 40–49 years (born 1949–1958) in 1998 corresponded to 4.5 l 100% ethanol, while the figure was 4.4 l for the conscripts in 1969/1970 (Leifman, 2000). Similar data from annual surveys during 2003–2004 showed that mean consumption was 4.7 in men aged 53–55 years, belonging to the same birth cohort as the conscripts (Ramstedt et al., 2009). All this indicates a good and long-standing predictive validity of self- reported alcohol consumption among these men, among whom a response rate of 98% also make them more representative as a birth cohort than usually is the case.

Nevertheless, a consumption change has probably occurred for some conscripts when they became older, maybe due to illness for some. The issue of the influence of ‘sick quitters’ (Shaper et al., 1988; Fillmore et al., 2006) is not an issue at conscription as men with permanent illness or disability were among the only 3–4% of the men who were not called to conscription. If subjects change their alcohol consumption during follow-up for reasons not related to the outcomes (non-differential misclassification as in this study), e.g. if some heavy consumers turn to lower consumption, this would lead to a dilution, a change towards the null condition of increased or decreased HR (Rothman et al., 2008). However, a change in consumption related to outcome can in principle lead either to an increase or a decrease in calculated risk. For instance, if alcohol consumers become abstainers, e.g. due to health problems, this would contribute to an increase of the reduced HR for MI, while a reduction in consumption especially among high consumers would lead to a reduction in the elevated HR for total mortality (differential misclassification), in both cases to a change towards the null condition (Rothman et al., 2008).The size of the impact of possible error cannot be estimated.

The protective effect of high consumption reported at conscription for non-fatal MI may partly be accounted for by change to low-moderate consumption later for some conscripts.

A few other studies report a decreasing risk of IHD incidence with increasing consumption, but our data do not permit a direct comparison with these studies (Ariola et al., 2009). The absolute net effect of consumption changes is not possible to estimate, but it seems likely that they more contributed to lower calculated HRs and APs.

We found different associations between alcohol and fatal compared with non-fatal and total MI, but have seen no other study on this topic up to the age of 55 years.

Some other studies have found that alcohol was associated with a smaller reduction of fatal than of non-fatal IHD or MI (Hvitfeldt et al., 2010; Ronksley et al., 2011) in subjects of higher ages. The impact of risk factors may vary by age, with possibly greater impact of genetic factors, hypertension and stroke at a younger age (Hvitfeldt et al., 2010).

Changes in alcohol use and other risk factors after conscription may be different for those with fatal and non-fatal MI, interaction between alcohol and other risk factors during follow-up might differ, the distribution of unmeasured confounders may vary, and also the genetic setup.

We found that alcohol was a cause of 18% of the deaths, while White et al. (2002, 2004) calculated from their aggregate data that 14% of deaths were attributed to alcohol among men 16–54 years, which is a noteworthy similarity considering the differences in data and study design. Based on their analyses, these authors suggest that men aged up to 34 should be advised to limit their consumption to one unit a day, to two units a day between 35 and 44 years and to three units a day between 45 and 54 years, while the level of alcohol consumption with the lowest risk was considerably lower (White et al., 2002).Our analyses of individual data with adjustment for several variables is a stronger design. Our results do not support recommendations of alcohol consumption from a strict health perspective in men in general up to the age of 55. An informed physician may give other advice to individual subjects, however (Mukamal, 2010).


This work was supported by Research Council of Working Life and Social Research (grants 2009-1611, 2009-0098).

Conflict of interest statement. None declared.


View Abstract