Re: Gestational diabetes mellitus and adverse pregnancy outcomes: systematic review and meta-analysis
We followed with interest the study by Ye et al. on the association of gestational diabetes mellitus (GDM) with adverse pregnancy outcomes; its findings could be useful in predicting risk and taking preventive measures to reduce adverse pregnancy outcomes. However, we consider that the study suffers from some methodological issues in the design and assessment, which may distort the true effects.
For the study design, the authors considered that insulin use can indicate poor glycaemic control, and therefore divided the studies into three categories: no insulin use, insulin use (different proportions of patients receiving insulin), and insulin use not reported. We have doubts about such a design: we believe that patients receiving insulin are more likely to have good glycaemic control, rather than poor control. Our argument can be supported by the results of meta-regression in the systematic review (Table S9), which showed that the proportion of patients receiving insulin was negatively associated with the incidence of preterm birth (β = -0.0069), namely, insulin was a protective factor of preterm birth (Here the authors were incorrect in their judgment that the proportion of patients receiving insulin was positively associated with the incidence of preterm birth). This evidence can shake all the conclusions regarding the association between insulin use and adverse pregnancy outcomes. We consider that only direct blood glucose indicators (e.g., fasting blood glucose, and glycated hemoglobin levels) can reflect glycaemic control, rather than using the proportion of patients receiving insulin as a replacement.
We also have some questions about the screening and assessment of this systematic review. First, the review excluded studies that identified GDM by the International Classification of Diseases (ICD) codes, for which the authors did not give a reasonable explanation. We believe that the diagnosis of GDM by the ICD codes is reliable and the exclusion of these studies would lead to selection bias. Second, we are concerned about the reliability of the selection process in Figure 1. Empirically, there should be more than 50% (22,310) duplicate records among the 44,620 records the systematic review screened, because a large proportion of bibliographies indexed in the four databases searched (Web of Science, PubMed, Medline, and Cochrane Database of Systematic Reviews) are identical. However, the authors only identified 1192 duplicates and 23,920 records were excluded as "irrelevant (rather than duplicate)" in the title and abstract screening. Third, although the authors gave definitions of high, medium, and low risk of bias for a single study, these definitions do not cover all scenarios; for example, it is unclear to which level of risk of bias "a study with a score of four for selection, one for comparability, and one for the outcome" can be assigned. Fourth, the authors did not include all subgroup categories in the subgroup analysis. For example, the subgroup analysis stratified by the risk of bias only included 50 studies with a low or medium risk of bias, and 106 studies with a high risk of bias studies were not included, without explanation. We believe that it is more important to evaluate the impact of studies with a high risk of bias on the effects estimates. Omitting the studies with a high risk of bias in the subgroup analysis may seriously underestimate the effect of the risk of bias on heterogeneity.
In summary, we believe that this systematic review and meta-analysis provided comprehensive evidence regarding the association between GDM and adverse pregnancy outcomes, but methodological limitations may affect the reliability of some key findings.
Competing interests: No competing interests