Relationship between alcohol consumption and MI
825
................................................................................................................................. European Journal of Public Health, Vol. 22, No. 6, 825–830 The Author 2011. Published by Oxford University Press on behalf of the European Public Health Association. All rights reserved. doi:10.1093/eurpub/ckr013 Advance Access published on 11 March 2011
.................................................................................................................................
Relationship between alcohol consumption and myocardial infarction among ageing men using a marginal structural model Jenni Iloma¨ki1, Anjum Hajat2, Jussi Kauhanen1, Sudhir Kurl1, Jay S. Kaufman3, Tomi-Pekka Tuomainen1, Maarit Jaana Korhonen4 1 2 3 4
Institute of Public Health and Clinical Nutrition, University of Eastern Finland, Kuopio, Finland Department of Epidemiology, University of Michigan, Ann Arbor, MI, USA Department of Epidemiology, Biostatistics and Occupational Health, McGill University, Montreal, Canada Department of Pharmacology, Drug Development and Therapeutics, University of Turku, Turku, Finland
Correspondence: Jenni Iloma¨ki, Institute of Public Health and Clinical Nutrition, University of Eastern Finland, Kuopio, Finland, tel: +358 40 35 53 735, fax: +358 17 16 29 36, e-mail:
[email protected]
Background: Studies on the association between alcohol consumption and myocardial infarction (MI) have typically used baseline data on alcohol consumption and potential confounders. This study aimed at investigating the association between alcohol consumption and MI considering time-varying alcohol consumption and time-varying confounders. Methods: Data were available for 1030 males participating in the Kuopio Ischaemic Heart Disease Risk Factor Study (Finland). Baseline data for the present study were collected in 1991–93. MIs were ascertained from national registries until December 2005. Alcohol consumption was categorized into four groups. Data were analysed using conventional discrete-time hazard and marginal structural models (MSMs). Time-invariant covariates were age, working status, diabetes and cigarette-years. Time-varying covariates in the MSM were prior alcohol consumption, smoking, history of cardiovascular diseases, body mass index, high-density lipoprotein cholesterol, systolic blood pressure, insulin and fibrinogen. Results: An insignificant increase of MI risk among the heaviest alcohol consumers (168 g week1) compared with the reference group (12–83 g week1) was observed when using a conventional model including baseline alcohol consumption and confounders measured prior to baseline [relative risk (RR) = 1.20, 95% confidence interval (95% CI) = 0.68–2.12]. When using a conventional model with time-varying alcohol consumption and adjusting for prior confounders, an increased risk of MI among the heaviest alcohol consumers was revealed (RR = 1.71, 95% CI = 1.03–2.85). There was also a trend towards increased risk among the heaviest consumers using the MSM (RR = 1.59, 95% CI = 0.93–2.72). Conclusion: Our findings suggest that standard methods using only baseline data on alcohol consumption and confounders may lead to biased estimates on the association between alcohol consumption and MI.
.................................................................................................................................
Introduction oronary heart disease (CHD) is the leading cause of death in Western
CEurope and in the USA. In Finland, a northern European country of 1
2
5.3 million residents, 23% of all deaths were attributable to CHD in 2008.3 Of the Finnish population aged 45 years, the prevalence of CHD was 16% among men and 14% among women in 2000.4 The relationship between alcohol consumption and CHD has been widely studied.5,6 Most research has suggested a J-shaped dose– response pattern between alcohol consumption and CHD, with the lowest risk among moderate drinkers (25 g day1 or 175 g week1 for men) and the highest risk among men consuming 114 g day1 (798 g week1).5 However, a recent pooled analysis of eight cohort studies revealed that alcohol consumption of 5–89 g day1 (35– 623 g week1) was protective against CHD for men.6 A non-significant reduced risk was also reported for men who consumed >89 g day1 (623 g week1). The analysis showed no evidence of a dose-related increased risk of CHD with higher levels of alcohol consumption. Many previous studies on alcohol consumption and CHD have adjusted for baseline high-density lipoprotein cholesterol (HDL-C), fibrinogen, glycosylated haemoglobin (HbA1c) and/or blood pressure.7– 9 However, since they may be influenced by prior alcohol consumption,10,11 these variables may be considered as intermediate variables on the causal pathway between alcohol consumption and CHD.12 Indeed, these variables may partly explain CHD attributable to alcohol consumption.10,11 Most previous studies have analysed the association between alcohol consumption and CHD using baseline alcohol consumption data,6,13 potentially leading to misclassification. The assumption inherent in these studies is that alcohol consumption does not change over time. However,
several cohort studies have reported age-related increases and decreases in alcohol consumption.14–16 Changes in alcohol consumption over time may confound the relationship between baseline alcohol consumption and cardiovascular disease (CVD).17,18 Changes in smoking pattern and body mass index (BMI) may also occur simultaneously with changes in alcohol consumption, implying that alcohol consumption, smoking pattern and BMI should be considered as time-varying quantities.12 The importance of appropriately including time-varying exposure and confounders in epidemiological models has been recognized. Several authors have described the potential bias introduced when adjusting for variables on the causal pathway.13,19 Marginal structural models (MSMs) are increasingly popular epidemiological methods that incorporate time-varying covariates that are both confounders and intermediaries.12,20 However, to our knowledge, MSMs have not been used to investigate the association between alcohol consumption and myocardial infarction (MI). The objective of this study was to investigate the association between alcohol consumption and MI considering time-varying alcohol consumption and time-varying confounders using an MSM. For comparison, data were also analysed using conventional discrete-time hazards models with and without time-varying alcohol consumption.
Methods Study population and design Data for this study were obtained from the longitudinal Kuopio Ischaemic Heart Disease Risk Factor Study (KIHD).21 The KIHD is an ongoing population-based study which aims to investigate risk factors for CVD.22 The initial examinations were conducted from March 1984 to December 1989 for 2682 men living in Kuopio, eastern Finland.
826
European Journal of Public Health
Participants were recruited in two cohorts; Cohort 1 comprised 1166 men (1498 eligible) aged 54 years whereas Cohort 2 comprised 1516 men (1935 eligible) aged 42, 48, 54 and 60 years. Cohort 1 was invited to take part in initial examination only. Cohort 2 was also examined in 1991–93 (1038 out of 1229 eligible men) and in 1998–2001 (854 out of 1007 eligible men) (figure 1). The study received approval from Ethics Review Board of the University of Kuopio. All study participants provided written informed consent. This study analysed data collected from participants in Cohort 2 only. Data collected in 1991–93 served as the baseline for this study (time t). Data collected at the initial KIHD examinations (time t 4) were used as a source of prior confounders and prior drinking, and data collected in 1998–2001 (t + 7) were used to assess possible changes in alcohol consumption (figure 1). All 1038 participants in the KIHD study at time t were eligible for inclusion in this study. We excluded eight participants with missing alcohol consumption data at time t, resulting in 1030 eligible participants in this study.
Identification of outcomes MI was defined according to the International Classification of Diseases (ICD), ninth revision: 410–414 and tenth revision: I20–I25. Events that occurred from time t until the end of 2005 were used. Data on MI were obtained from the National Hospitalization Registry and from the National Causes of Death Statistics. The National Hospitalization Registry includes the main diagnosis (and up to two secondary diagnoses) on admissions to all hospitals and health centre wards in Finland. The National Causes of Death includes data on all deaths. To validate the register data, all MIs were also extracted from each participant’s medical record by trained research physicians, in accordance with the World Health Organization (WHO) Monitoring of Trends and Determinants in Cardiovascular Disease (MONICA) Project criteria.23 We included only definite MIs in the study.
Measurement of alcohol consumption The same self-administered questionnaire was used to collect alcohol consumption data at all time points. The questionnaire was an adapted
version of the Nordic Alcohol Consumption Inventory that was used in the Scandinavian Drinking Survey.24 It included items on the quantity and frequency of alcohol consumption for wine, beer and hard liquor separately, during the preceding year. From these items, a continuous measure of average alcohol consumption in grams per week (g week1) was calculated. We categorized alcohol consumption for this study into four groups:
