Eligibility criteria

We used PICOS framework to formulate eligibility criteria as follows: Population: age ≥ 18 years, healthy or diagnosed with arterial hypertension, and/or type 2 diabetes mellitus, and/or dyslipidemia, and/or overweight/obesity, and/or myocardial infarction/stroke; Intervention: cocoa supplement or chocolate bar (≥ 70% cocoa), cocoa powder, cocoa into snacks; Comparison: placebo or < 70% cocoa white/milk chocolate; Outcomes: measurements of systolic blood pressure (SBP) and diastolic blood pressure (DBP) (mmHg); BMI (kg/m²); body weight (kg); waist or abdominal circumference (cm) as an indicator of central obesity; total cholesterol, HDL cholesterol, LDL cholesterol, triglycerides levels; fasting blood glucose and glycated hemoglobin; Study: RCTs only.

This article presents a comprehensive description of the study population of adults with or without established cardiometabolic risk factors (arterial hypertension, and/or type 2 diabetes mellitus, and/or dyslipidemia, and/or overweight/obesity, and/or myocardial infarction/stroke). The inclusion criteria of our review will include studies with two or more arms; an "intervention" arm involving the consumption of cocoa or ≥ 70% cocoa dark chocolate for at least four weeks and a "comparator" arm of placebo or < 70% cocoa white/milk chocolate.

It should be noted that the study populations, methods and intervention will be labeled according to the authors’ description. When information is not available, authors will be contacted by email and asked to respond within 15 days (three attempts). For subgroup analyses in the meta-analysis, only RCTs reporting polyphenol dose for the intervention group and the control group will be selected. The results of subgroup analyses comparing healthy individuals with those with cardiometabolic risk factors will also be discussed.

Inclusion and exclusion criteria

We will select studies involving adults aged 18 or older and consumption of cocoa (cocoa powder, cocoa into snacks) or ≥ 70% dark chocolate, with no limitation on the amount of cocoa or polyphenols, for at least 4 weeks with detailed assessments of cardiometabolic risk markers, including BP (measured using standard techniques), BMI, waist or abdominal circumference, lipid profile, and glucose metabolism (fasting blood glucose or glycated hemoglobin). Studies combining other dietary interventions with a clearly defined intervention group of cocoa or ≥ 70% dark chocolate intake and a control group will be thoroughly reviewed for inclusion. Studies with participants undergoing medication or diet changes will be eligible if these changes were introduced at least four weeks prior to the start of the intervention or were consistent throughout the study to allow for accurate analysis. Regarding the 4-week intervention, there is no specific guideline or reference, as the outcomes are associated with different timeframes for promoting adaptations. For instance, “polyphenols and weight loss” take 4–6 weeks to show results [28]. Similarly, improvements in “lipid and glycemic profiles with polyphenol supplementation” have been observed in 4–12 weeks, depending on baseline conditions [29,30]. As for changes in blood pressure levels, some studies indicate that cocoa supplementation can reduce blood pressure over a period of 4–8 weeks [23,31]. Therefore, we decided to use 4 weeks as the minimum intervention period.

There will be excluded studies with participants undergoing treatments other than for cardiometabolic conditions; pregnant and post-menopausal women; concomitant use of dietary supplements that are not clearly distinguished from the cocoa intervention; review or protocol studies; animal experimentation studies; and studies of conditions other than those related to cardiometabolic health, such as individuals diagnosed with cancer. Studies involving the same sample published in different journals will be thoroughly reviewed and excluded if duplicate.

Search strategy

We conducted a prior search in the MEDLINE database via PubMed to ascertain whether the research question of our review meets the FINER (feasible, novel, and relevant) criteria. A search strategy for RCTs was then developed and will be conducted by four independent reviewers (two pairs) in the databases recommended in the Cochrane Handbook for Systematic Reviews of Interventions [27]: MEDLINE, EMBASE (European literature), Web of Science, and CENTRAL–Cochrane Central Register of Controlled Trials (to allow access to trials that are not indexed in MEDLINE and EMBASE). To broaden our search, we will conduct searches of publications from Latin America in LILACS (Latin American and Caribbean Health Sciences Literature/Virtual Health Library [VHL]) and SciELO (Scientific Electronic Library Online). To minimize publication bias, we will also search the gray literature, including OpenGrey and the Brazilian Coordination for the Improvement of Higher Education Personnel (CAPES) Bank of Theses and Dissertations. For unpublished ongoing studies, we will conduct searches in the following clinical trial registries: ClinicalTrial.gov; and Brazilian Clinical Trials Registry (REBEC). Preprint databases will also be included (preprints.org/, biorxiv.org and medrxiv.org). Table 1 details the search terms and search strategies used for each database. We will gather data through careful review of the articles retrieved and the authors will be contacted by email for additional information as needed. Articles in English, Portuguese and Spanish with no date of publication restrictions will be eligible for inclusion. Upon completion of the review, we will undertake an additional search of all databases and registry platforms to ensure the inclusion of the most recent studies.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Search terms for searches in Medline, EMBASE, Web of Science, CENTRAL–Cochrane Central Register of Controlled Trials, LILACS and SciELO.

https://doi.org/10.1371/journal.pone.0309824.t001

The main search terms will include “cocoa,” “Theobroma cacao,” and “dark chocolate” (Table 1). To increase accuracy and sensitivity of our searches, the search terms for the study design (RCT) will be entered in the databases MEDLINE [32] and EMBASE [33] (Table 1). Four reviewers will independently select studies after an initial screening of titles and abstracts. When there will be no sufficient information in the abstract, the reviewers will read the full text of the article. Any disagreements between the reviewers will be resolved through discussion and, if no consensus can be reached, a fifth reviewer (WG) will be consulted.

Data extraction and management

After completing our searches in each database, all articles retrieved will be exported as “.ris” or “.enib” files and imported into Rayyan reference manager [34]. Duplicate studies will be removed using the deduplication function. Our blinded reviewers will manually check for remaining duplicates by reviewing titles, authors, year of publication and abstracts.

Our reviewers will independently screen eligible articles based on their titles and abstracts. They will use Rayyan application “to include” articles that meet the eligibility criteria. Those that do not meet the inclusion criteria will be marked as “excluded” and categorized by reason for exclusion (i.e., ineligible outcomes or population and non-RCT design). When it is not clear whether a study should be included, it will be marked as “undecided”. In the last step, our reviewers will compare the articles screened for any discrepancies. If there is not enough information in the abstract, we will retrieve and read the full-text article. Disagreements will be resolved by consensus and any disagreements on inclusion criteria will be resolved by a fifth reviewer (WG). Cohen’s kappa statistic will be used to measure the agreement between the reviewers.

The four blinded reviewers will independently conduct data extraction. When studies will be considered relevant and selected, the main data will be extracted and compiled in a pre-structured Excel 2019 database divided into columns: reviewer; author; year; journal; country; inclusion and exclusion criteria; analysis strategy; number of individuals included and excluded; number of individuals analyzed; sex; age; follow-up period; nutritional supplement taken; how it was supplemented; daily dose; and risk of bias. For the main analysis, baseline and post-intervention data will be extracted (mean and dispersion measures) for the following markers: BP, body weight, BMI, waist and abdominal circumference, blood glucose, glycated hemoglobin, total cholesterol, HDL cholesterol, LDL cholesterol and triglycerides. For the extraction of data from eligible studies with results presented in graphs, we will contact the authors by email to obtain these data or use GetDate Graph Digitizer 2.26 to extract the data.

Risk of bias

The risk of bias of RCTs will be assessed using Cochrane risk of bias (RoB) 2 tool included in the Cochrane Handbook [27,35,36]. The assessment is based on a set of five domains of bias rated as low, high, or unclear risk of bias: bias arising from the randomization process; bias due to deviations from intended interventions; bias due to missing outcome data; bias in measurement of the outcome; bias in selection of the reported result. When participants are not blinded, studies will be classified as high risk of bias in the “blinding of participants and personnel” domain. No study will be excluded based on the risk of bias assessment. The risk of bias will be analyzed for the primary outcomes of interest in our review.

Risk of overall bias in systematic reviews

We will assess the certain of evidence using the Grading of Recommendations Assessment, Development and Evaluation (GRADE) tool [37,38]. This tool assesses confidence in paired effect estimates and classifies treatment in a meta-analysis (high, moderate, low and very low) in the following domains [37,38]: study design, methodological limitations (risk of bias), inconsistency, indirectness of evidence, imprecision, publication bias, magnitude of effect, dose-response gradient, and residual confounders.

Analysis strategy

We will carry out statistical analyses to estimate the effects of cocoa and/or ≥ 70% cocoa dark chocolate consumption on anthropometric measurements, BP, and lipid and glycemic profiles compared to a control group. For outcomes with the same unit, summary effect estimates will be expressed as mean difference (MD) and related 95% confidence interval (95% CI). For outcomes with different units, summary effect estimates will be expressed as standardized mean difference (SMD) and related 95% CI. The SMD expresses the size of the intervention effect in each study relative to the variability in outcome measurements observed in that study. Given that the SMD provides an overall interpretation of the results (absolute unit), we will convert SMD into units of the most common measurement instrument or into proportions (%) so that the data are presented in a language more adequate to the outcome analyzed.

In studies involving humans, inconsistencies across studies are to be expected due to different characteristics of the participants (clinical heterogeneity) or methods (methodological heterogeneity), with different effect sizes between studies. Therefore, to incorporate variability in the meta-analysis and to summarize the data, MDs or SMDs from individual studies will be pooled using a random-effects model. Therefore, we will use a random-effects model to incorporate variability in the meta-analysis. Since the 95% CI from random effects refer to uncertainty in the location of the mean effects across studies, we will consider the calculated values for a 95% prediction interval (95% PI) as they reflect the interval of uncertainty of the effects to be expected in future RCTs [39]. To assess the consistency of cocoa and/or ≥ 70% cocoa dark chocolate effects across studies, the degree of heterogeneity will be tested using the inconsistency test by Higgins (I²) for every pairwise comparison [40,41]. To explore heterogeneity, we will conduct subgroup analyses and meta-regression analyses (≥ 10 studies) for effect modifiers with normal distribution in a quartile-quartile plot (qq-plot) and confirm it using the Shapiro-Wilk test (p > 0.05) [42]. Non-normal data will be normalized before performing meta-regression analysis. To remove discrepant data from the meta-analysis, forest plots will be constructed to display the effect estimates across studies and detect potential outliers based on non-CI overlapping that is due to heterogeneity [41]. Potential effect modifiers (e.g., age, BMI, number of interventions and total polyphenol content) will be analyzed separately. If there will be significant heterogeneity that cannot be explained, we will not perform a meta-analysis, but we will present individual intervention effect estimates from the studies. Considering that selective publication and/or suppression of specific results cause bias and consequently affect the validity of the results [43], if applicable (≥ 10 studies; more than one study with significant statistical data; studies with different sample sizes), we will perform the Egger’s test using a funnel plot to assess potential publication bias in the meta-analysis [44,45]. If publication bias is detected (Egger’s test, p < 0.1), we will use the trim-and-fill method to identify and correct for funnel plot asymmetry by adding information of the bias-corrected data to the original data (prior to trimming and imputation of data by the trim-and-fill method) [46,47]. Alternative analyses to the primary analysis of data including sensitivity analysis will be performed to assess the robustness of our decisions (e.g., imputation method used to impute missing values; inclusion of studies with high risk of bias, data from conference abstracts etc.) [48]. Dispersion measures expressed as confidence intervals (CI) or standard errors (SE) will be converted to standard deviation (SD = EP * √n) before the analysis. For eligible studies that do not report SD of differences, it will be estimated using an imputed correlation coefficient (CC) of 0.5 [49]. CC equation (section 6.5.2.8, Cochrane Handbook): Δ SD = √ SD² baseline + SD² final–(2 * CC * SD baseline * SD final). Two-tailed tests will be used at a significance level of p< 0.05. Data modeling will be performed with RStudio (version 1.3.959) using the R package meta (version 3.6.1) for Windows.