Elevated Pregnancy Losses at High and Low Extremes of Maternal Glucose in Early Normal and Diabetic Pregnancy: Evidence for a protective adaptation in diabetes

The DIEP was conducted in five academic centers, Cornell University Medical College, Harvard Medical School, Northwestern University Medical School, the University of Pittsburgh Medical School, and the University of Washington School of Medicine. The sponsor of the study and data-coordinating center was the NICHD, Bethesda, Maryland. The study design has been described in detail previously (1,17–19). Briefly, insulin-dependent diabetic women contemplating pregnancy were recruited by public advertising, news releases, and contacting clinics specializing in diabetes between the years of 1980 and 1985. The diagnosis of pregnancy was made during the 1st week of missed menses by plasma human chorionic gonadotropin determination. Diabetic subjects were then admitted to a metabolic ward for monitoring and educated on home glucose monitoring, diary keeping, and insulin adjustment. Fixed goals of glycemic management were not mandated. Nondiabetic control subjects were screened for gestational diabetes at 26 weeks’ gestation (20) and were excluded from the control cohort if they were positive. Glycated protein measurements were performed in 429 control and 389 diabetic pregnancies and fructosamine in 386 control and 358 diabetic pregnancies. The methods of early pregnancy diagnosis, pregnancy dating, and assessment of pregnancy loss have been described in detail previously (1). Early pregnancy loss was defined as <20 weeks’ gestation (1).

Plasma or serum samples were collected weekly from diabetic subjects from gestational weeks 4–12. Nondiabetic subjects had blood samples drawn at biweekly intervals from weeks 4 through 12. Frozen samples were initially shipped to the NICHD and stored at −20° C and then to analytical sites and stored at −80° C.

Glycated serum and plasma proteins were assessed with use of boronate affinity high-performance liquid chromatography methodology (CLC 330 Glycated Hemoglobin Analyzer; Primus, Kansas City, MO) in the laboratories at Sansum Diabetes Research Institute. The normal ranges for glycated serum and plasma protein levels are 16.1–23.0% and 20.1–25.0%, respectively. The interassay coefficient of variation (CV) for glycated protein in serum was 0.87% and in plasma 1.73%.

This assay was performed on thawed and refrozen samples after the glycated protein assay was shipped on dry ice to the reference laboratory. The assay is based on the reduction of nitroblue tetrazolium by the ketoamine bond of glycated protein (15). The analysis was performed using a RoTAG kit (Hoffmann-La Roche, Nutley, NJ) adapted to a COBAS BIO centrifugal analyzer (21) (Roche Biomedical Laboratories, Indianapolis, IN). The normal range is 1.0–2.7 mmol/l with a CV of 3.6% in replicate analyses.

Samples obtained in the first trimester were grouped into three intervals: 4–6, 7–9, and 10–12 weeks of gestation. The mean and maximal values of glycated protein and fructosamine measurements within each of these intervals were standardized to adjust for between-center and plasma versus serum differences by subtracting the value for the corresponding control group and dividing by the SD (standardized value or Z score). The data from individual subjects from all five centers were then combined for analysis. The analysis was carried out using both the mean and maximal measurements in each time interval. However, little difference was seen between the analyses of mean and maximal values; maximal values are therefore presented hereafter.

Descriptive statistics were produced and ordinary and chain logistic regression methods applied to assess the relationship between the protein level and early pregnancy loss. The chain logistic model (22) was used based on the rationale that pregnancy loss in a given time interval might be affected by the values of the glycated protein and fructosamine levels in the preceding interval as well as that interval. The chain logistic model allows a life-table analysis structure between pregnancy loss and covariates as used previously for survival data analysis (22).

Mean standardized maximal glycated protein values were similar between the nondiabetic control pregnancies that resulted in live births and those that resulted in fetal losses (0.0 ± 0.8 and 0.1 ± 1.0, P = 0.64). Mean standardized maximal glycated protein values were also similar in the diabetic pregnancies that resulted in live births and fetal losses (1.7 ± 1.4 and 1.9 ± 1.8, P = 0.53), despite the fact that mean values in the diabetic group were nearly twofold greater than in the control group.

Pregnancy loss rates at different levels of standardized glycated protein Z scores are presented in Table 1. Diabetic pregnant women with a Z score ≥4.0 above the mean for normal subjects had a pregnancy loss rate of 24% (5 of 21). Those with a Z score <1.0 below the mean had a pregnancy loss rate of 33% (4 of 12). Pregnancy loss rates in the intervening Z score intervals were 8% in the −1.0 to <1.0 interval, 14% in the 1.0 to <2.0 interval, and 11% in the 2.0 to <3 and 3.0 to <4.0 intervals for an average of 11% (39 of 356) in the −1.0 to <4.0 interval. When two extreme Z score intervals are collapsed (i.e., ≥4.0 and <–1.0) and compared with the intermediate Z score intervals (i.e., −1.0 to 4.0), the difference in pregnancy loss rate is significant (OR 3.0 [95% CI 1.3–7.0], P = 0.01 using two-sided Fisher’s exact test). A similar pattern across strata was observed for fructosamine values and pregnancy losses (data not shown).

Normal control subjects had a similar pattern of pregnancy loss rates, higher at the extremes and lower for the middle Z score values but over a narrower range than in the diabetic subjects, ranging from <–1.0 to <3.0 Z score values. Loss rates were 50, 13, 12, and 16%, respectively, in the nondiabetic control subjects at the four intervals of Z score values shown in Table 1. Over the entire range of values, the loss rates were nearly identical in the diabetic and nondiabetic pregnancies at 12 and 13%, respectively, as seen previously (1). Again, a similar pattern was seen for standardized fructosamine values and pregnancy losses (data not shown).

To analyze the data as a continuum, ordinary logistic regression and chain logistic regression analyses were performed. As presented in Table 2, sample sizes are lower in the chain logistic model than in the ordinary logistic model because not all subjects had values at all intervals, causing a break in the chain logistic regression model for those subjects. Regardless, both analyses indicate an excess of early pregnancy loss at the higher and lower extremes of glycemic status in both diabetic and nondiabetic pregnancies. In the ordinary logistic regression model, the quadratic term (x²) testing the J-shaped curve of standardized glycated protein level versus pregnancy losses was statistically significant (P = 0.019 in nondiabetic subjects and P = 0.015 in the diabetic subjects). In the chain logistic model, the value for x² shows a P value of 0.07 in the diabetic subjects and 0.01 in the control subjects. Values for fructosamine were not statistically significant and are not shown.

Logit plots of pregnancy loss versus individual value of glycated protein and fructosamine (i.e., fitted logistic models) are presented in Fig. 1A and B, respectively, obtained from ordinary logistic regression analysis along with the 95% CI. As seen in Fig. 1, diabetic women with a pregnancy loss and a Z score <–1.0 or >4.0 have higher loss rates. For example, for a Z score of −2, the predicted logit is −1.13 and the corresponding predicted probability of pregnancy loss 0.32. For a Z score of 4.5, the predicted logit is −1.36 and the corresponding predicted probability of pregnancy loss 0.26. For a Z score of 0, the predicted probability of pregnancy loss is 0.14. Nondiabetic control pregnant women have a similar pattern but over a narrower range of elevated values. For example, for a Z score of −1.5, the predicted logit is −1.40 and the corresponding predicted probability of pregnancy loss 0.25. For a Z score of 2, the predicted logit is −1.00 and the corresponding predicted probability of pregnancy loss is 0.37. For a Z score of 0, the predicted probability of pregnancy loss is 0.12. The 95% CIs parallel the logit plots and confirm the curvilinearity of the J-shaped plots. The fructosamine values confirm the same trends, though by an entirely separate and analytical method.

The DIEP addressed the following two questions: 1) what is the relationship between glucose levels and malformations? and 2) what is the relationship between these levels and the risk of spontaneous abortion in pregnancies complicated by insulin dependent diabetes (1,18)? The first reports from this study used glycated hemoglobin as a marker of glycemic control (23). The analysis of the relationship of glycated hemoglobin to spontaneous abortion revealed an indistinguishable pregnancy loss rate in the diabetic and control groups (1). However, successive SDs in glycated hemoglobin above the normal range were associated with progressive increases in pregnancy losses. Other investigators have also shown a relationship between glycemia and pregnancy losses in diabetic women (2,3), but none have determined whether the relationship extends to normal pregnancy and whether there is risk for fetal loss if the glycemic levels are too low. The latter possibility is supported by the early observations of Abell et al. (4) who observed increased perinatal losses in pregnant mothers with a flat glucose tolerance test curve, a risk that was exaggerated by the presence of other indexes of fetal distress. This observation implies a role for fetal malnutrition based on impaired delivery of maternal nutrients to the fetus that could extend to the first trimester as well, which could also be aggravated by additional risk factors (24). To our knowledge, no studies have investigated the potential for a deleterious effect of low glycemic status in the first trimester on spontaneous abortion in normal and diabetic pregnancy.

Additional evidence of a deleterious effect of glycemic deficiency has been reported. Glycemic deficiency was implicated in teratogenicity and fetal resorption in rodents (25). Fetal growth retardation was associated with hypoglycemia or low-normal glycemic levels in a diabetic population (26). The DIEP study may not have found a relationship between low-level glycated hemoglobin and first trimester loss (1) because of the prolonged time constant for change dictated by the 120-day lifespan of red cells. This time constant is well in excess of the short-term events relating glycemic status to fetal well being in the 8 weeks that can be feasibly monitored in the first trimester. In this light, the measurement of protein glycation is a more serviceable research monitor of glycemic status related to first trimester pregnancy loss and teratogenesis.

Our observations showing a relationship of pregnancy loss to protein glycation are in keeping with other studies showing that glycated proteins are a more sensitive measure of fluctuations of blood glucose in the short term compared with glycated hemoglobin (13,16). Further research is justified to understand adverse fetal and neonatal outcome at glycemic extremes, using protein glycation as an index of intermediate-term glycemic status in both diabetic and nondiabetic pregnancy.

A provocative observation in this study is the finding that early pregnancy loss rates in diabetic pregnancy do not increase until glycated protein Z scores are >4.0, whereas loss rates in nondiabetic pregnancies increase when glycated protein Z scores exceed 2.0. This finding suggests that adaptation may occur to chronic hyperglycemia in the diabetic pregnancy, allowing the embryo or fetus of the diabetic mother to resist the deleterious effect of hyperglycemia. Such adaptations could be anti-inflammatory, osmotic, or involving antioxidant and vasodilatory defense mechanisms that have been observed in in vitro systems under oxidant stress (27–30). For instance, aortic endothelial cells exposed to hydrogen peroxide showed an initial thromboxane and prostacyclin rise, then a decline over a 24-h incubation period (R.H.K. and X.-D.Z., unpublished data). Understanding these mechanisms may provide an approach to the prevention or treatment of some diabetes complications.

This investigation is the first to show increased pregnancy losses at glycemic extremes of both normal and diabetic pregnancies in the largest cohort of normal and diabetic subjects studied to date. Limitations are the smaller numbers of subjects at the extremes and the hypothesizing rather than a priori hypothesis testing nature of this report. The speculation of nutritive deficiency on the one hand and glycemic injury on the other requires further confirmation and investigation.

In conclusion, we have shown that both high and low extremes of glycemia as measured by protein glycation are associated with increased risk for early pregnancy loss. The J-shaped curve relating high and low glycemic status to first trimester pregnancy loss applies to both diabetic and nondiabetic pregnancies. At the high extreme of glycemia, even short-term toxicity of severe hyperglycemia may impact fetal viability. At the low extreme of glycemia, impaired fetal nutrition is implicated. However, the level of hyperglycemia tolerated in the diabetic pregnancy before reaching the threshold of fetal vulnerability is higher than that in nondiabetic pregnancy, implying maternal, placental, or fetal defenses to the stress of hyperglycemia. Important clues to the prevention of diabetes complications may be embedded in this interesting observation and deserve future research.

This study was supported by a grant from the NICHD.

The authors acknowledge the Primus Corporation for assistance with the measurement of glycated protein and the Sansum Research Institute for technical assistance. The authors also thank the clinical centers of the DIEP for their hard work in performing the DIEP study and the persistence of the collaborating investigators.

Participating Centers and Investigators of the Diabetes in Early Pregnancy Study include Cornell University Medical College, Lois Jovanovic, MD; Harvard Medical School, Lewis B. Holmes, MD; University of Pittsburg Medical School, Jerome H. Aarons, MD; Northwestern University School of Medicine, Joe Leigh Simpson, MD, and Boyd Metzger, MD; University of Washington School of Medicine, Robert H. Knopp, MD, Zane Brown, MD, and Margot Van Allen, MD.