Screening
After searching all source databases, all the citations will be exported into EndNote (Carlsbad, CA, USA) software. The accumulated citations will be screened electronically using Microsoft Excel (Redmond, WA, USA). The relevant articles will be obtained following removal of duplicates of the records identified through database searching. Two reviewers (JAM and CB) will independently assess titles and abstracts of pertinent articles found according to the inclusion and exclusion criteria. Previously, we developed a study selection sheet containing titles, abstracts, allocated inclusion status (‘e’, excluded; ‘i’, included; and ‘?’, unclear), and reasons for exclusion. Judgment concerning inclusion will be made individually and results compared. Divergences will be discussed by two authors (JAM and CB) to resolve the conflict; if no agreement is reached, a third reviewer (SD) will be consulted to make a final decision. Finally, full texts of all articles preliminarily identified as unclear for inclusion based on title and abstract will be obtained for a second screen.
Data extraction
The final set of studies for inclusion will be obtained after review of full-text articles considered eligible by two reviewers (JAM and CB). Data will be extracted through a standardized data compilation form [Annex 5 in Additional file 1] inspired from the Cochrane Collaboration’s tool and then cross-checked. Author names, sample sizes, and results of studies will be compared to avoid duplicates. This form was pilot-tested on three studies to refine it consequently. The data extraction form provides information on qualitative aspects of studies.
Data will be extracted on: 1) publication source (such as date of publication, design, geographical origin, contact details, and location of research group); 2) eligibility verification (such as study design, population, intervention, outcome, and reasons for exclusion); 3) methods of recruitment (such as duration of enrollment, inclusion and exclusion criteria, diagnostic criteria of preeclampsia, study size, and setting); 4) participant characteristics (such as number included in the analysis, age range, mean age, percentage of subjects per age interval, ethnicity, level of education, socioeconomic status, parity, percentage of women by body mass index (BMI) category, smoking status, and percentage of women affected by gestational diabetes); 5) characteristics of intervention evaluated (such as type, timing of exposure, method and technique of measurement, assessment of chocolate consumption, total women by number of servings of chocolate consumed per week by trimester, and assessment of consumption or median theobromine concentration by trimester and theobromine cord concentration); and 6) information of the reported outcome (such as outcome assessed, number of cases with preeclampsia, total cases by quartiles of theobromine concentrations and categories of chocolate consumption by trimester, measures of disease association (crude and adjusted) by quartiles of theobromine concentration, and categories of chocolate consumption by trimester).
Quality assessment
Two independent reviewers (JAM and CB) will evaluate the risk of bias for eligible studies to ascertain their validity. For this purpose, a standardized form [Annex 6 in Additional file 1] was constructed inspired from the Cochrane Collaboration’s tool for assessing risk of bias, with reference to meta-analysis of observational studies in epidemiology (MOOSE) [16], quality assessment tool for systematic reviews of observational studies (QUATSO) [17], and strengthening the reporting of observational studies in epidemiology (STROBE) [18] quality assessment tools.
The possibility of publication bias will be assessed by evaluating a funnel plot of the studies log(RR) or log(OR) against standard error for asymmetry, which can result from the non-publication of small trials with negative results. Since graphical evaluation can be subjective, an adjusted rank correlation test and a regression asymmetry test will also be conducted as formal statistical tests of publication bias. We acknowledge that other factors, such as differences in study quality or true study heterogeneity, could produce asymmetry in funnel plots.
Summary of study results
The relative risk (RR) of preeclampsia will be the primary measure of treatment effect. If not reported in included studies, the odds ratio (OR) will be considered instead. Crude and adjusted RR and OR will be treated separately; confounding factors will be noted for adjusted measures of association. In the case of data permitting meta-analysis, the authors will be contacted to obtain crude data in order to calculate RR. If databases are not available, the RR or OR will be taken directly from the publications.
Planned methods of analysis
Chocolate intake will be determined from food frequency questionnaires, and categories of consumption will be considered in the analysis. Serum theobromine, a biomarker for cocoa, will also be considered in the analysis by interval of concentration. The specific categories will be determined according to included studies.
Analysis of all studies meeting the inclusion criteria will be carried out using Review Manager (RevMan) software (version 5.2; The Nordic Cochrane Centre, Copenhagen). For each study, n and the proportion of pregnant women with preeclampsia according to the level of chocolate consumption or control group will be entered into the software. For randomized trials, a meta-analysis will be performed if possible by category of intervention (chocolate consumption) and the random effects model will be used. For non-randomized studies, a meta-analysis will be carried out by study design and potential confounding factors. The potential confounding factors to be explored will be age, ethnicity, level of education, parity, BMI, smoking, and gestational diabetes using directed acyclic graphs (DAGs) to determine required adjustments, and subgroup analysis will be used to explore heterogeneity.
Exploring heterogeneity
We hypothesize that the effect size may differ according to the methodological quality of the studies, and we will explore the variability of study results (heterogeneity) before conducting the analysis. Heterogeneity will be explored by subgroup analysis according to confounding factors found. If applicable, the robustness of the results will be tested by different outcomes (if available) and type of chocolate.