Strengths and Limitations
This is the first systematic review evaluating and synthesizing the associations between maternal diet quality and adverse perinatal outcomes. The rigorous search and screening strategies employed, as well as the prospective registration and publication of our protocol limit the possibility of reporting bias. This review has several limitations. The exclusion of non-English publications, of abstracts, and of other grey literature could lead to publication bias. Publication bias could not be formerly assessed due to the limited number of studies reporting each specific outcome (less than 10 for any outcome). Some of the included studies reported diet quality measured with different tools and/or at different timepoints. A single measure had to be selected to conduct the meta-analysis which could influence the results of the analysis. However, this is unlikely as sensitivity analyses conducted using other measures from the same cohorts provided similar estimates. These results are derived from observational studies and therefore no conclusions can be made on causal inference. The interpretation of these results is also limited by the variability in study populations, dietary scores, reported outcomes, and adjustment of confounding factors across studies. The vast majority of cohorts were conducted in high-income countries and therefore these results cannot be applied to middle and low-income countries. Twenty-two different dietary scores were used in the identified studies. The components of each diet quality score vary according to the underlying dietary recommendations, but they included similar adequacy components (vegetables, fruit, fish, nuts and unsaturated fats), and moderation items (added sugars and saturated fats). The robustness of the meta-analysis results in the sensitivity analyses including different diet quality scores measured in the same population is also reassuring as to the comparability of these scores. In the majority of studies, diet was appraised using an FFQ. While the use of FFQs has been associated with recall bias and classification errors, this is likely to bias results towards the null hypothesis. Studies reported variable outcome sets, which limited the number of results available for each outcome. While most studies used validated measurements or medical registration for outcome measurements, others relied on self-reporting. Misclassification related to self-reporting could have diminished the strength of the associations. In studies reporting GDM as an outcome, diet before 24 weeks of pregnancy was measured in six studies35,43,44,47,54,55, but the timing of dietary assessment and recall period were not reported in the last study34. While we cannot exclude reverse causation due to measure of diet after implementation of dietary interventions in the last study, this would be expected to bias results towards the null hypothesis. While the results accounting for the most confounders were used in the meta-analysis, adjusted factors varied greatly between studies, and we cannot exclude the possibility of unaccounted confounding factors.