Response to “Studies Assessing Risk of Treatments for Diabetes Mellitus and Adverse Pregnancy Outcomes Should Control for Known Risk Factors”

Dear Editor:

W e thank Edson-Heredia et al. ¹ for their interest in our meta-analysis. ²

We undertook the systematic review and meta-analysis to summarize the available evidence to answer the clinical question on the impact of lispro insulin (LP) use from before pregnancy to delivery in women with type 1 diabetes mellitus (T1DM). We agree that we did not perform either a meta-analysis of individual data or adjustment for known risk factors of large for gestational age (LGA). Nevertheless, it is remarkable that maternal baseline characteristics and hemoglobin A1c (HbA1c) during pregnancy, even when nonsignificantly different between groups, would favor a lower rate of LGA in the LP group (trend to lower age, body mass index, and HbA1c), and nevertheless the contrary was the case.

When describing the articles included in the meta-analysis the authors indicate (1) that significance for LGA was not reached in one of the articles ³ or that when it was reached for LGA this was not the case for macrosomia. ⁴ The fact that individual reports included in a meta-analysis do not reach significance is not relevant; in fact, one of the merits of the meta-analysis tool is to sum up small reports, potentially reaching significance where individual reports do not.

Edson-Heredia et al. ¹ point out that recent literature reviews ^{5 –7} concluded that there were no relevant differences in rate of macrosomia and LGA, but we would like to underscore that all of them are qualitative reviews. The difference is that we performed a systematic review and meta-analysis. Disagreements between narrative and systematic reviews are not unexpected. In their letter, the authors provide a clear example of potential weaknesses of narrative reviews. They selected the publication of Evers and co-workers describing the perinatal outcome in pregnancies of women with T1DM versus the background population ⁸ instead of the corresponding article addressing macrosomia ⁹ where the association with LP was described. This fact would be highly unlikely in a systematic review.

To explain the increased rate of LGA in LP users Edson-Heredia et al. ¹ quote the comment of Evers et al. ⁹ indicating that use of LP could be biased toward women with more difficult glycemic control. We agree on this potential explanation for the results of Evers et al. ⁹ because included women had presented for antenatal care between 1999 and 2000, shortly after the availability of LP. Nevertheless, this should be minimized in later reports where the use of insulin analogs is more frequent, and this is not the case (relative risks are remarkably similar). ²

After reviewing our data, we realized that there was an omission on our report of maternal severe hypoglycemia. While we stated that there was no available information on this outcome, one of the articles ⁴ provided it [3.2% in women using LP in the whole pregnancy vs. 5.4% in those using regular insulin (RI) (relative risk 0.63; 95% confidence interval 0.15, 2.67; difference not significant)]. This information does not modify our conclusion, but we will send the corresponding correction to the journal. In relation to metabolic control and hypoglycemia, Edson-Heredia et al. ¹ state, “Although based on unadjusted analyses, observational studies in pregnant patients also found benefits of LP over RI, with better glycemic control and less hypoglycemia.” In our view, this comment is not appropriate in the context of this meta-analysis. ² First, Lapolla et al. ⁴ observed a lower HbA1c in the first trimester in women treated with LP, and these data are included in the meta-analysis with a final result of a nonsignificant trend. Their data on hypoglycemia were nonsignificant. As to the narrative review of Durnwald and Landon ⁷ published in 2011, their overall conclusion is that “from the entirety of these studies, it can be concluded that insulin lispro does not offer a clear benefit for improved glycemic control or less hypoglycemia.” Finally, Aydin et al. ¹⁰ do not provide segregated information for women with T1DM.

We do not agree on that maintaining normoglycemia is an ultimate goal during pregnancy. The ultimate goal is a safe outcome for the mother and fetus, and this not only requires normoglycemia, but also that the instruments used to achieve it do not have untoward effects of its own.

With the available information, we maintain the conclusion in the abstract: “In relation to women with T1DM treated with RI, those treated with LP display similar baseline characteristics and no differences in metabolic control or perinatal outcome with the exception of a higher rate of large-for-gestational-age newborns.” ² In the full text, the final sentence “Meanwhile, the conclusion should be that there is no evidence of a beneficial effect of LP use during pregnancy” should end “in women with T1DM.” We do not agree with Edson-Heredia et al. ¹ in that our conclusion is premature. Beginning the sentence with “Meanwhile,” it is clear that we did not intend to make a final conclusion.

Finally, we fully agree with the authors that a clear statement on this issue should be the result of a well-designed clinical trial, as already pointed out in the last paragraph of our article.