We carried out a systematic review and meta-analysis of studies that evaluated the association between chocolate consumption and cardiometabolic disorders in adults. Between June 2010 and 5 October 2010 (last date searched) we searched Medline (1950 to present), Embase (1980 to present), the Cochrane Library (1960 to present), Scopus (1996 to present), Scielo (1997 to present), Web of knowledge (1970 to present), AMED (1985 to present), and CINHAL (1981 to present).
We used combinations of text words and thesaurus terms that included cacao[Mesh], cacao[Title/Abstract]*, chocolate, chocolate*[Title/Abstract], cocoa, cocoa*[Title/Abstract], cardiovascular diseases [Mesh], cardiovascular disease, cardiovascular, coronary artery disease [Mesh], atherosclerosis [Mesh], coronary disease [Mesh], coronary heart disease, coronary, myocardial, ischaemic heart disease, ischaemic heart disease*[Title/Abstract], ischemic heart disease, ischemic heart disease*[Title/Abstract], Heart Failure[Mesh], myocardial infarction [Mesh], myocardial ischemia [Mesh], stroke, stroke [Mesh], cerebral stroke, stroke, brain vascular accident, cerebrovascular stroke, cerebrovascular accident, cerebral vascular, cerebral vascular accident, cerebrovascular CVA, metabolic syndrome X [Mesh], metabolic syndrome, metabolic cardiovascular syndrome*[title/abstract], diabetes type 2, diabeti, diabete, diabet*[title/abstract], diabetes mellitus [Mesh], diabetes mellitus, type 2 [Mesh].
Studies were included if they were randomised controlled trials or cohort, case-control, or cross sectional studies; carried out in adults (≥18 years old); studied the effects of levels of chocolate consumption; the outcomes of interest were related to cardiometabolic disorders (cardiovascular disease, myocardial infarction, stroke, ischaemic heart disease, heart failure, diabetes, and metabolic syndrome); and had no language restriction (if necessary, local scientists fluent in the original language helped with translation). We excluded studies including only pregnant participants; letters, abstracts, systematic reviews, meta-analysis, ecological studies, and conference proceedings; and studies carried out in non-humans.
Two independent reviewers working in pairs (AB-L and OHF, JS, LJ, or SW) screened the titles and abstracts of the initially identified studies to determine if they satisfied the selection criteria. Any disagreements were resolved through consensus or consultation with a third reviewer. Full text articles were retrieved for the selected titles. Reference lists of the retrieved articles were searched for additional publications. We also contacted the authors of the retrieved papers directly for any additional and unpublished studies. The retrieved studies were assessed again by two independent authors (AB-L and OHF) to ensure that they satisfied the inclusion criteria.
We designed a data collection form before the implementation of the search strategy. Two independent reviewers used this form to extract the relevant information from the selected studies (AB-L and OHF). The data collection form included questions on qualitative aspects of the studies (such as date of publication, design, geographical origin and setting, funding source, selection criteria, patient samplings, and location of research group), participant characteristics (such number included in the analysis, age range, mean age, sex, ethnicity, recruitment procedures, residential region, socioeconomic status, comorbidities, and drug treatment), characteristics of the exposure or intervention evaluated (such as type, method used to measure), and information on the reported outcomes (such as measure of disease association, type of outcome, outcome assessment method, type of statistical analysis, adjustment variables).
Two independent reviewers (AB-L and OHF) evaluated the quality of the studies included using a modified scoring system that was created on the basis of a recently used system (designed with reference to MOOSE, QUATSO, and STROBE) that allowed a total score of 0–6 points (6 reflecting the highest quality).26
The system allocates one point each for (a
) any justification given for the cohort; (b
) appropriate inclusion and exclusion criteria were used; (c
) diagnosis of cardiometabolic disorders was not solely based on self reporting; (d
) participants’ usual chocolate consumption was assessed with a validated tool; (e
) adjustments were made for age, sex, body mass index, and smoking status; and (f
) any other adjustments were done (such as for physical activity, dietary factors).
We evaluated the differences between low and high chocolate consumption on outcomes such as diabetes, incidence of cardiovascular disease, cardiovascular mortality, coronary heart disease, incidence of stroke, and deaths from stroke. We pooled results using a random effects model, and we did tests for heterogeneity and publication bias. Results were expressed as pooled relative risks with 95% confidence intervals.
For each study, we compared the group with highest chocolate consumption against the group with the lowest consumption. Hazard ratios, relative risks, and odds ratios were assumed to approximate the same measure of relative risk. By pooling the study-specific estimates using a random effects model that included between-study heterogeneity (parallel analyses used fixed effect models), we calculated summary relative risks. We carried out a cumulative meta-analysis by outcome in which the pooled estimate of the association reported was updated each time the results of a new study were included. Possible sources of heterogeneity of relative risks were examined using the Cochran-Mantel-Haenszel test for the null hypothesis of no effect (relative risk=1), and the Mantel-Haenszel common relative risk estimate.27
(which quantifies the percentage of variation attributable to heterogeneity) was reported as a measure of consistency across the studies.
Finally, we assessed publication bias by using a funnel plot and Begg’s test to find out whether there was a bias towards publication of studies with positive results among studies with a smaller sample size.28
To test the robustness of our findings, we repeated the meta-analysis by different outcomes (any cardiovascular disease, diabetes, heart failure, and stroke), sex, and types of chocolate (dark, milk, white).