Alcohol Consumption and Risk of Chronic Obstructive Pulmonary Disease

A Prospective Cohort Study of Men

Joanna Kaluza; Holly R. Harris; Anders Linden; Alicja Wolk


Am J Epidemiol. 2019;188(5):907-916. 

In This Article


Study Population

The Cohort of Swedish Men was established in central Sweden in 1997, when all men born between 1918 and 1952 (45–79 years of age) and residing in Västmanland and Örebro counties were invited to complete a 350-item questionnaire. Among the 48,850 men who returned a completed questionnaire (49% response rate), we excluded those with missing or incorrect national identification numbers (n = 297), those who died prior to baseline (n = 55), those with a previous cancer diagnosis (other than nonmelanoma skin cancer) (n = 2,592), and those who had a COPD diagnosis (International Classification of Diseases, Tenth Revision, code J44) prior to baseline (n = 281). Moreover, we excluded those with implausible values for total energy intake (>3 standard deviations (SDs) from the mean value for log-transformed energy intake; n = 567) and those with missing data on alcohol consumption (n = 804). After these exclusions, the analytical cohort included 44,254 men. As has been previously reported,[9] the original and analytical cohorts well represented the Swedish male population in 1997 in terms of age distribution, educational level, prevalence of overweight/obesity, and smoking status.

The study was approved by the Regional Ethics Committee in Stockholm (protocol 2016/2034-31/1).

Assessment of Alcohol Consumption

Alcohol consumption was assessed in 1997 with a food frequency questionnaire. Participants indicated their average consumption of 96 food items, including 6 types of alcoholic beverages, during the past year. Using 8 prespecified response categories ranging from never to ≥3 times per day, participants indicated how often they had consumed class I beer (<2.25% ethanol by volume), class II beer (2.80%–3.50%), class III beer (>3.50%), wine, strong wine (>18% alcohol), and liquor during the past year, as well as the usual amounts of these beverages consumed on a single occasion (reported in an open-ended question). We calculated weekly alcohol consumption by multiplying the frequency of consumption of each beverage by the amount consumed and expressed consumption in terms of 1 standard drink. The definition of 1 alcoholic drink can vary between countries.[19] We defined 1 standard drink as 12 g of ethanol, corresponding to 660 mL of class I beer, 500 mL of class II beer, 330 mL of class III beer, 150 mL of wine, 80 mL of strong wine, and 40 mL of liquor. In other countries (e.g., the United Kingdom), 1 drink of alcohol is equal to 8 g of ethanol.[19] Partial nonresponse for specific alcoholic beverages (5.6% for beer, 14.0% for wine, and 10.4% for liquor) was assumed to represent never or occasional drinking.

The food frequency questionnaire has been validated in 248 Swedish men aged 40–74 years using fourteen 24-hour recall interviews.[20] The Spearman correlation coefficient for correlation between the two methods used to collect data was 0.81 for alcohol consumption.

Assessment of Other Covariates

Information on educational level, weight, height, physical activity, and cigarette smoking was obtained via the questionnaire.[21] Respondents were asked about their age of smoking initiation, years since stopping smoking (if applicable), and the average number of cigarettes smoked per day at different ages (15–20 years, 21–30 years, 31–40 years, 41–50 years, 51–60 years, and the current year). Pack-years of smoking were estimated by multiplying the number of years of smoking by the reported number of packs of cigarettes smoked per day within each age category. We calculated 2 diet quality scores, recommended food score and nonrecommended food score, which have been described previously.[22]

Case Ascertainment

Using the unique personal identification number assigned to each Swedish resident, we identified incident COPD cases through linkage with the Swedish Patient Register (inpatient and outpatient registers) and the Swedish Cause of Death Register. COPD events were defined using International Classification of Diseases, Tenth Revision, code J44. We defined a COPD case as the first diagnosis of COPD in the Swedish Patient Register (listed at any diagnosis position) or in the Cause of Death Register (listed only in the primary position). It has been previously reported that the proportion of patients diagnosed with COPD in primary care increased from 59% in 1999 to 81% in 2009 (a time period overlapping with our study).[23] Patients were also about 7 years younger when they received a COPD diagnosis in 2009 compared with 1999 (age 66 years vs. age 73 years).[23] These observations may indicate some degree of COPD underdiagnosis during the early years of follow-up of the Cohort of Swedish Men.

Statistical Analysis

Study participants were followed from January 1, 1998, to the date of the first COPD diagnosis, death, or the end of follow-up (December 31, 2014), whichever occurred first. Men were categorized into quintiles of ethanol consumption, as well as into 7 categories of total alcohol consumption: never drinker or occasional drinker (<1.0 standard drink/week), 1.0–3.9, 4.0–6.9, 7.0–14.0, 14.1–20.0, or >20.0 standard drinks/week, and past drinker. We also categorized specific types of alcoholic beverage consumption into 5 categories of beer consumption (<1.0, 1.0–1.9, 2.0–4.0, 4.1–6.0, or >6.0 standard drinks/week) and 4 categories of wine or liquor consumption (<1.0, 1.0–1.9, 2.0–4.0, or >4.0 standard drinks/week).

Cox proportional hazards regression models were used to calculate hazard ratios with 95% confidence intervals. The multivariable hazard ratios were adjusted for age, smoking status and pack-years of smoking, education, physical activity, body mass index (weight (kg)/height (m)2), energy intake, and recommended and nonrecommended food scores. For specific alcoholic beverages, beer, wine, and liquor consumption were mutually adjusted for each other by inclusion in the same multivariable model.

The assumption of proportional hazards was assessed by regressing scaled Schoenfeld residuals against survival time; there was no evidence of departure from the assumption. To test nonlinearity of the associations, we modeled the quantitative exposure of total alcohol consumption and specific alcoholic beverages using restricted cubic splines with 3 knots at fixed percentiles (10th, 50th, and 90th) of the distribution, and coefficients of the second spline transformation equal to zero were tested.[24] Using a likelihood ratio test, the interactions between total alcohol consumption and smoking status, body mass index, and physical activity were examined.

Analyses of data were performed using SAS, version 9.4 (SAS Institute, Inc., Cary, North Carolina) and STATA 13 (StataCorp LLC, College Station, Texas). All reported Pvalues are 2-sided; P values less than or equal to 0.05 were considered statistically significant.