Increase Font Size Decrease Font Size View as PDF Print
Citation:

Corrao G, Rubbiati L, Bagnardi V, Zambon A, Poikolainen, K. Alcohol and coronary heart disease: A meta-analysis. Addiction. 2000; 95: 1,505-1,523.


PubMed ID: 11070527
Study Design:
Meta-analysis or Systematic Review
Class:
M - Click here for explanation of classification scheme.
POSITIVE: See Research Design and Implementation Criteria Checklist below.
Research Purpose:
  • To summarize estimates of the parameters describing the relationship between alcohol consumption and the risk of coronary heart disease (CHD)
  • To measure the heterogeneity across studies describing the relationship between alcohol consumption and the risk of CHD
  • To evaluate whether the characteristics of the studies and of the individuals may explain a part of the heterogeneity across studies describing the relationship between alcohol consumption and the risk of CHD
  • To evaluate the possibility of publication bias concerning studies describing the relationship between alcohol consumption and the risk of CHD.
Inclusion Criteria:
  • Case control or cohort study
  • Published as a primary article
  • Findings expressed as relative risk, considering three or more levels of alcohol consumption
  • Reported number of cases and non-cases for each exposure level
  • When results of a study were published more than once, the most recent comprehensive article was included.
Exclusion Criteria:

None explicitly stated.

Description of Study Protocol:

Search Procedures

  • Listing in MEDLINE, reference listed in article listed on MEDLINE, reference by other bibliographic databases available at the University of Milan (Current Content from 1996, EMBASE from 1980, CAB Abstracts from 1973, and Core Biomedical Collection from 1993), hand search of general reviews and meta-analyses published on issue.
  • Publication Between 1966 and 1998
  • Keywords listed as disease (coronary heart disease, coronary artery disease, coronary event, coronary death, myocardial infarction, ischemic heart disease and angina pectoris), alcohol consumption (alcohol or ethanol and consumption, intake and drinking) and the association measures (relative risks, risk ratio, odds ratio and rate ratio).

Was Study Quality Assessed?

Study eligibility was assessed by readers blinded to authors’ names and affiliations and to the reported results. These same blinded readers evaluated the quality of studies by scoring quality based on the pre-defined criteria of study design, data collection methods and data analysis. Points were awarded using a standard scale. Maximum scores were given when methods least likely to result in bias were used. Discrepancies between readers were resolved through discussions.

Type of Intervention and Outcomes Investigated, Population Included

  • Relationship between alcohol consumption and the relative risk of CHD
  • Determine if qualitative characteristics of included studies modified effect of alcohol
  • Determine a more reliable function of the relationship between alcohol consumption and CHD risk by including only high-quality studies in analysis
  • Identify sources of heterogeneity of the effects of alcohol intake
  • Determine if publication bias affected validity of the estimates
  • Population included adult males and females.
Data Collection Summary:

What Type of Information Was Abstracted from Articles?

  • Study quality parameters
  • Design
  • Outcomes
  • Diagnostic category
  • Cases and non-cases
  • Country of residence for subjects
  • Age of subjects
  • Gender of subjects
  • Alcohol consumption: Amount and dose
  • Association concerning each level of alcohol consumption and the corresponding confidence interval.

How Was it Combined?

  • Alcohol consumption combined based on grams per day
  • Alcohol consumption given in ranges was combined by assigning dose as the midpoint of the range. For the highest level of consumption, the range was assumed to be close-ended based on the width for all other dose ranges.
  • Association measures and corresponding confidence intervals were translated into natural logarithm relative risks and corresponding variances.

What Analytic Methods Were Used, If Any?

  • Relationship between alcohol consumption and the relative risk of CHD: Non-linear models fitted by pre-pooling the results of all included studies. Models within the fractional polynomial family were used. Specifically, a family of second-degree models was generated by power transformation of the exposure variable. The best-fitting model was chosen to summarize the relation of interest. This step was performed using all studies as well as studies gauged to be of highest quality by blinded readers.
  • Determine if qualitative characteristics of included studies modified effect of alcohol: Meta-regression models were fitted. Qualitative characteristics included in analysis were quality score, three key components of quality score and study design. A significant effect of the interaction term (interaction between alcohol intake and value of covariate describing a qualitative characteristic) implies the covariate modifies the effect of Alcohol intake on the relative risk. This step was performed using all studies as well as studies gauged to be of highest quality by blinded readers. However, the latter included an interaction term representing objects of interest rather than criteria for selection of studies. To perform the latter, the goodness of fit was assessed by the residual deviance and the D statistics were compared for the two models using the likelihood ratio test.
  • Determine if publication bias affected validity of the estimates: Tested by a funnel-plot-based approach where a test of asymmetry of the funnel plot was conducted based on the method proposed by Egger et al.
Description of Actual Data Sample:
  • Number of articles included: 51
  • Number of articles identified: 196
  • Number and type of studies reviewed:
    • 43 cohort studies
    • Eight case control studies
  • Various geographic locations represented
  • Some gender specific results reported (38 for males and 17 for females)
  • Sample size of studies, and characteristics of the study participants.

Main Characteristics of the 51 Studies Included in the Meta-analysis

First Author and Year Design Outcome* Diagnostic Category** Subjects***
Klatsky, 1974 Case control  NF  MYI

298/299

Yano, 1977 Cohort  F+NF 

CHD

NFI

ANP 

294/7,705

Blackwelder, 1980 Cohort  CHD 

132/8,006

Dyer, 1980 Cohort  CHD 

89/1,832

Kagen, 1981 Cohort  F+NF  CHD 

287/7,304

Klatsky, 1981 Cohort 

CHD

AMI 

317/8,060

Gordon, 1983 Cohort  F+NF  CHD 

906/4,625

Kittner, 1983 Cohort  F+NF  CHD

643/9,150

Colditz, 1985 Cohort  CHD 

42/1,184

Kaufman, 1985 Case control NF  MYI 

981/2,170

Gordon, 1985 Cohort  F+NF  CHD

159/823

Friedman, 1986 Cohort  CHD

174/2,310

Klatsky, 1986 Cohort  F+NF  CAD

694/5,001

Kono, 1986 Cohort 

CHD

AMI 

113/5,135

Scragg, 1987 Case control  NF  MYI 

456/1,582

Suhonen, 1987 Cohort CHD

140/4,532

Garfinkel, 1988 Cohort  CHD

18,984/581,321

Stampfer, 1988 Cohort

F

NF

F+NF 

NFI

FHD

SHD

400/87,526

Boffetta, 1990 Cohort  CHD

18,771/276,802

Klatsky, 1990 Cohort  CAD 

600/123,840

Miller, 1990 Cohort  F+NF  CHD 

49/1,341

Kono, 1991 Case control  NF  AMI

83/271

Lazarus, 1991 Cohort  IHD 

187/4,070

Rimm, 1991 Cohort 

F

NF

F+NF 

NFI

FHD

CAD 

350/51,529

De Labry, 1992 Cohort  CHD 

74/1,823

Farchi, 1992 Cohort  CHD

104/1,563

Jackson, 1992 Case control 

F

NF 

MYI

COD

526/1,183

Kalandidi, 1992 Case control  NF  CHD

329/570

Klatsky, 1992 Cohort  CHD

940/12,8934

Suh, 1992 Cohort  CHD

190/11,668

Wannamethee, 1992 Cohort  IHD

182/5,778

Bianchi, 1993 Case control  NF  AMI 

298/685

Cullen, 1993 Cohort  CHD

325/2,171

Garg, 1993 Cohort  F+NF  IHD

475/3,718

Hein, 1993 Cohort  F+NF  IHD

86/2,563

Prineas, 1993 Cohort  FMI

115/32,898

Doll, 1994 Cohort  IHD

974/12,321

Goldberg, 1994 Cohort  CHD

132/3,793

Fuchs, 1995 Cohort  CHD

320/85,709

Iso, 1995 Cohort F+NF  CHD

31/2,890

Serdula, 1995 Cohort  IHD

785/8,187

Camargo, 1997 Cohort 

F

NF

F+NF 

CHD

MYI 

ANP

2,243/22,071

Keil, 1997 Cohort  F+NF  CHD 

62/2,084

McElduff, 1997 Case control  F+NF  MCE

8,681/60,65 

Rehm, 1997 Cohort  CHD

434/6,788

Thun, 1997 Cohort  CHD

2,622/489,626

Wannamethee, 1997 Cohort  F+NF  CHD 

490/7,735

Yuan, 1997 Cohort  IHD

104/18,224

Kitamura, 1998 Cohort  F+NF 

CHD

MYI

ANP 

163/8,476

Renaud, 1998 Cohort  CHD  284/34,014 

* Fatal (F); non-fatal (NF); fatal and non-fatal events (F+NF).

** Acute myocardial infarction (AMI); angina pectoris (ANP); coronary artery disease (CAD); coronary heart disease (CHD); coronary death (COD); fatal heart disease (FHD); fatal myocardial infarction (FMI); ischemic heart disease (IHD); major coronary event (MDE); myocardial infarction (MYI); non-fatal infarction (NFI); severe heart disease (SHD).

*** Number of cases/Number of non-cases (total number of subjects in cohort study or number of controls in a case control study.

Summary of Results:

Findings

  • A total of 21 models were tested to identify the best fitting one to describe the relationship between alcohol consumption and the relative risks of coronary heart disease. All 51 included studies were considered in this analysis.
  • All models left a significant residual deviance (tabulated chi square = 311.6, DF=272). The model including the linear and root-squared alcohol terms fit the data best.

Key Findings

  • Studies of the highest quality (those reporting relative risks adjusted for the main risk indicators, those considering lifetime abstainers as referents, those excluding subjects with pre-existing disease at baseline and those performed with a cohort deign) showed lower protective effect of alcohol.
  • Pooled analysis of all 51 studies found:
    • Protective effect evident up to 90g per day (RR=0.94; 95% CI: 0.90, 1.00)
    • Harmful effect reached at 113g per day (RR=1.08; 95% CI: 1.00, 1.16)
  • Pooled analysis of the 28 highest-quality studies found:
    • Protective effect evident up to 72g per day (RR=0.96; 95% CI: 0.92, 1.00)
    • Harmful effect reached at 89g per day (RR=1.05; 95% CI: 1.00, 1.11)
  • Considering fatal events as outcome, it was noted that:
    • Protective effect evident up to 56g per day (RR=0.96; 95% CI: 0.92, 1.00)
    • Harmful effect reached at 73g per day (RR=1.06; 95% CI: 1.00, 1.12)
  • Considering both fatal and non-fatal events:
    • Protective effect evident up to 114g per day (RR=0.94; 95% CI: 0.89, 1.00)
    • Harmful effect reached at 141g per day (RR=1.10; 95% CI: 1.00, 1.20)
  • Considering females:
    • Protective effect evident up to 31g per day (RR=0.93; 95% CI: 0.87, 1.00)
    • Harmful effect reached at 52g per day (RR=1.12; 95% CI: 1.00, 1.26)
  • Considering males:
    • Protective effect evident up to 87g per day (RR=0.94; 95% CI: 0.88, 1.00),
    • Harmful effect reached at 114g per day (RR=1.09; 95% CI: 1.00, 1.19)
  • Considering the Mediterranean countries: Protective effect evident up to 145g per day (RR=0.76; 95% CI: 0.61, 1.00)
  • Considering all other countries: Protective effect evident up to 80g per day (RR=0.93; 95% CI: 0.87, 1.00)
  • Small studies reporting relative risks greater than one for these intakes were less likely to be published compared with small studies reporting relative risks less than one (low intake: intercept=α=-0.31, 95% CI: -0.58, -0.00; moderate intake: intercept =α=-0.23, 95% CI: -0.46, -0.00).
Author Conclusion:
  • The degree of any protective effect due to moderate doses of alcohol should be reconsidered, since the degree of protection from high-quality studies are smaller than those based on all studies
  • High intake levels of alcohol are related to increased risk of CHD
  • When considering the combination of protective and harmful effects, a J-shaped curve best describes the relationship between alcohol intake and CHD risk
  • Both gender and area in which the study is performed modify the effect
  • These results strongly suggest that the higher alcohol-related susceptibility of women acts towards both protective and harmful effects.
Reviewer Comments:

This study was extremely thorough and demonstrates a strong conclusion as to the knowledge of alcohol consumption effects concerning CHD. Recommendations must take into account where the recommendations are being provided and to whom.


Research Design and Implementation Criteria Checklist: Review Articles
Relevance Questions
  1. Will the answer if true, have a direct bearing on the health of patients?
Yes
  2. Is the outcome or topic something that patients/clients/population groups would care about?
Yes
  3. Is the problem addressed in the review one that is relevant to nutrition or dietetics practice?
Yes
  4. Will the information, if true, require a change in practice?
Yes
 
Validity Questions
  1. Was the question for the review clearly focused and appropriate?
Yes
  2. Was the search strategy used to locate relevant studies comprehensive? Were the databases searched and the search termsused described?
Yes
  3. Were explicit methods used to select studies to include in the review? Were inclusion/exclusion criteria specified and appropriate? Were selection methods unbiased?
Yes
  4. Was there an appraisal of the quality and validity of studies included in the review? Were appraisal methods specified, appropriate, and reproducible?
Yes
  5. Were specific treatments/interventions/exposures described? Were treatments similar enough to be combined?
Yes
  6. Was the outcome of interest clearly indicated? Were other potential harms and benefits considered?
Yes
  7. Were processes for data abstraction, synthesis, and analysis described? Were they applied consistently across studies and groups? Was there appropriate use of qualitative and/or quantitative synthesis? Was variation in findings among studies analyzed? Were heterogeneity issued considered? If data from studies were aggregated for meta-analysis, was the procedure described?
Yes
  8. Are the results clearly presented in narrative and/or quantitative terms? If summary statistics are used, are levels of significance and/or confidence intervals included?
Yes
  9. Are conclusions supported by results with biases and limitations taken into consideration? Are limitations of the review identified and discussed?
Yes
  10. Was bias due to the review’s funding or sponsorship unlikely?
Yes