Study participants were from population-based cohorts embedded in the Korean Genome Epidemiology Study (KoGES), which is an ongoing longitudinal investigation. Cohort members included Korean men (n = 4,752) and women (n = 5,261), aged 40-69 years and residing in 2 cities, Ansan and Ansung, when they were enrolled between June, 2001 and January, 2003. Detailed information regarding enrollment of cohort members and study procedures is available in a previous report [3]. Briefly, cohort members were recruited based on a two-stage cluster sampling procedure and were enrolled after undergoing a comprehensive health examination and on-site interviews, which were administered by trained personnel. At the health examination, cohort members provided blood specimens for evaluation of biochemical parameters to measure lipid and glucose metabolism as well as for genetic assays. They completed questionnaire-based interviews to collect data on socio-demographics, medical history and health conditions, family history of disease, and lifestyle factors. They have also been followed biennially and have undergone similar health examination and interviews. At each visit, the participants sign an informed consent form, which was approved by the Human Subjects Review Committee either at the Ajou University Medical Center or at the Korea University Ansan Hospital. All procedures and protocols of the study were approved by both institutions.

Among a total of 8,716 individuals whose clinical and genetic data were available at baseline, we excluded 10 persons who reported pregnancy at baseline (n = 1), who did not complete the baseline anthropometric measurement (n = 3), or who did not provide information on alcohol consumption (n = 6). Because the outcomes of this study are new cases of stroke, we excluded 89 persons who reported at baseline a physiciandiagnosis of stroke. Additionally, we further excluded 152 persons who reported at baseline a physician-diagnosis of coronary artery disease, congestive heart failure, or peripheral artery disease to minimize the effects of these diseases. After exclusions, a total of 8,465 participants (47% men and 53% women) were entered for analysis.

The outcomes of this study were incident events or deaths due to stroke during a follow-up period from April, 2003 to December, 2010. Incident events were self-reported by interview-based questionnaires and deaths were ascertained with its cause and date based on vital registration data, which are periodically compiled by the Korea National Statistical Office. We excluded 10 persons who reported a diagnosis of both CAD and stroke during follow-up to distinguish risk factors of stroke from those of CAD. Concordance between self-reported diagnoses of stroke and those ascertained by brain magnetic resonance imaging was 91% in a subgroup of stroke cases.

To examine the ALDH2 polymorphism, we focused on the most significant single-nucleotide polymorphism (SNP), rs2074356, which was reported to be highly associated with alcohol consumption (adjusted P-value < 9.5 × 10^-59) and to be a surrogate marker of ALDH2 Glu504Lys (rs671) in previous genome-wide association (GWA) studies (r² = 0.8) [3,15]. Genotypes of this SNP were observed to be CC, CT, and TT; its minor allele (T) frequency was 0.15 among male drinkers [3]. An earlier report provided information regarding the preparation of genomic samples, genotyping method, and quality control for the GWA data [16].

We collected information on age, sex, family history of stroke, smoking status, snoring during sleep, physical activity, alcohol drinking status, and amount of alcohol consumed from questionnaires used at the initial interview. Participants were asked whether their biological parents and sibling had been diagnosed with stroke. Regarding snoring status, they were asked if they snore; if so, they were further asked how frequently they snore (infrequently, 1 to 3 times per week, 4 to 5 times per week, 6 to 7 times per week). To collect information regarding daily physical activity, participants were asked to report hours spent in a typical day in sleep and 5 categories of activity intensity (sedentary, very light, light, moderate, vigorous), given details on activities corresponding to each category. Specific metabolic equivalent (MET) values assigned to each category were multiplied by the reported hours spent to yield a total metabolic equivalent (MET-hours) score as a quantitative estimate of physical activity. Participants were asked whether they had ever consumed alcoholic beverages in their lifetime, whether there was a time in their life when they regularly consumed at least one drink of any alcoholic beverage in every month, and whether they drank in the past 30 days. They were also requested to complete a table to collect information on the average frequency of drinking and amount of alcoholic beverages consumed during a typical occasion in the past 30 days. Such information was used to calculate daily alcohol consumption (g/day).

At the health examination, height (cm) and body weight (kg) were measured by health professionals trained with a standardized protocol to yield body mass index (BMI, kg/m²). Based on biochemical and clinical evaluations and self-reports on medications, the presence of hypertension (systolic/diastolic pressure ≥ 140/90 mmHg), diabetes mellitus (fasting glucose ≥ 126 mg/dL or 2-hour glucose ≥ 200 mg/dL), or hypercholesterolemia (serum total cholesterol ≥ 240 mg/dL) was determined.

Descriptive statistics on baseline characteristics of the study population were calculated. To analyze the association between stroke risk and ALDH2 polymorphic variants and potential risk factors, we conducted Cox proportional hazards regression analysis. The person-years for each subject were calculated from the date when he or she participated in the baseline examination to the date when he or she reported the first stroke events in follow-up examinations or to death date or to December 31, 2010, whichever came first. Persons who died, refused further participation, or were lost to follow-up were censored. The length of follow-up period used for analysis was 9.5 years with a median follow-up period of 7.8 years. Stroke risk was expressed as a hazard ratio (HR) with its 95% confidence interval (CI). In multivariate models, age, BMI, and physical activity were adjusted as a continuous variable and other variables were treated as categorical variables, including current smoking status (nonsmoking, moderate smoking of 20 cigarettes per day or fewer, heavy smoking of 21 cigarettes per day or more), snoring status (no snoring, snoring for 1-3 nights per week, snoring for 4-6 nights per week or every night), and current alcohol drinking status (nondrinking in the past 30 days, light alcohol drinking of 1 drink per day or fewer, moderate alcohol drinking of 2 drinks per day, heavy alcohol drinking of 3 drinks per day or more). However, because the proportion of heavy smokers or heavy drinkers was trivial in women, categories for smokers and drinkers were united. ALDH2 polymorphic variants (mutant type and wild-type), the presence of family history of stroke, and the presence of hypertension, diabetes mellitus, and hypercholesterolemia were fitted as binary variables in the model. Two multivariate Cox models stratified by sex were constructed in this study. The first multivariate model included ALDH2 variants and potential risk factors except current alcohol drinking status. The second model included all variables described above and current alcohol status. Analyses stratified according to alcohol drinking status were also performed. Joint effects of ALDH2 variants and significant risk factors of stroke were analyzed after taking other covariates into account. Proportional hazards assumptions were tested for the second model and no violation was confirmed. All testing was based on a two-sided level of significance. The SAS program (SAS 9.1.3, 2008, SAS Institute, Cary, NC, USA) was used to conduct statistical analyses.

Over median follow-up of 7.8 years, 121 cases (67 men and 54 women) of stroke (99 incident events and 22 deaths due to stroke) were newly identified, with 53035.5 person-years accrued for analysis. A comparison of cohort members who were followed up until 2010 or ascertained as stroke cases with those lost during the follow-up period revealed no significant difference in the distribution of the wild-type (CC) and the mutant type (CT or TT) of the ALDH2 polymorphism (rs2074356) (P = 0.66).

Characteristics of the study population according to ALDH2 polymorphic variants are presented in Table 1. Of these, 27% had the mutant alleles. Persons with the wild-type allele were more likely to suffer from hypertension or diabetes mellitus (P < 0.01), have a greater BMI (P < 0.05), and consume more alcoholic beverages (P < 0.001) compared with those with the mutant alleles.

Table 2 shows the association between stroke risk and ALDH2 variants and potential risk factors among 4,011 men. Men with the wild-type of the ALDH2 polymorphism showed an increased risk of stroke. Compared with the mutant alleles of the ALDH2 polymorphism, the multivariate HR [95% CI] of stroke was 2.02 [1.03-3.99] for those with the wild-type before adjusting for alcohol consumption; values were not significant after adjustment. However, alcohol consumption itself was not significantly associated with stroke risk according to the multivariate model. The lowest risk of stroke was observed in light alcohol drinkers who consumed an average of 1 drink (15 g of alcohol) per day. In the second model that was further adjusted for alcohol consumption, aging, the presence of diabetes mellitus, and habitual snoring were observed to be significant risk factors of stroke. Men with diabetes mellitus had a 1.9-fold higher risk than those without. In addition, men who snored 4-7 nights per week had a 3-fold higher risk than non-snorers (Table 2). After adjusting for alcohol consumption, the previous results, except those for the ALDH2 polymorphism, were not changed (Table 2).

Table 3 shows the association between stroke risk and ALDH2 variants and potential risk factors for 4,454 women. The ALDH2 polymorphism was not associated with the stroke risk, regardless of adjusting for alcohol consumption. In the second multivariate model that was further adjusted for alcohol consumption, aging and habitual snoring were observed to be significant risk factors of stroke. Women who snored 4-7 nights per week had a 2.6-fold higher risk of stroke than non-snorers (Table 3).

Multivariate results stratified by current status of alcohol drinking among men or women are presented in Table 4. Aging was found to be a significant risk factor, regardless of sex and alcohol drinking status. Although the association was not significant, carrying the wild-type of the ALDH2 polymorphism generally increased the risk of stroke in male non-drinkers or drinkers, but generally decreased the risk in women. The presence of diabetes mellitus or snoring significantly increased the risk of stroke in male drinkers. Among women, the presence of snoring tended to increase the risk, but the association was not significant for either non-drinkers or drinkers.

Table 5 shows the joint effects of the ALDH2 polymorphism and other significant risk factors such as aging, presence of diabetes mellitus, and snoring on stroke risk after taking into account other potential risk factors. Among men, the multivariate HR (95% CI) of stroke risk was 6.88 [2.78-17.05] for carriers of the wild-type aged 60 years or older compared with those with the mutant alleles aged 59 years or younger. Carriers of the wild-type allele with diabetes mellitus had a 3.6-fold [1.62-8.03] increased the risk of stroke compared with those with the mutant alleles without diabetes mellitus, while snorers with the wild-type had an 8.5-fold [1.16-61.74] increased risk compared with non-snorers with the mutant alleles among men. However, joining of the ALDH2 polymorphism and snoring status was not significantly associated with stroke risk among women.