Urinary and Serum Angiogenic Markers in Women With Preexisting Diabetes During Pregnancy and Their Role in Preeclampsia Prediction

Compared with the general population, patients with preexisting diabetes represent a high-risk cohort of women, having almost four times the risk of preeclampsia during pregnancy (1). Preeclampsia is associated with serious maternal and neonatal morbidity and mortality (2); thus, it would be of great clinical benefit to be able to identify women with preexisting diabetes who are at higher risk of developing this pregnancy-associated condition. The pathophysiology of preeclampsia is thought to be through abnormal cytotrophoblast invasion of maternal spiral arterioles resulting in aberrant placental vasculature and subsequent placental hypoperfusion and ischemia. This leads to an imbalance of angiogenic and antiangiogenic factors, with increasing evidence in women without diabetes that this plays a significant role in the consequent manifestations of preeclampsia (3,4), appearing before clinical signs become apparent (5–7). Thus, angiogenic factors may serve as important prognostic markers to aid risk stratification for high-risk women.

Vascular endothelial growth factor (VEGF) and placental growth factor (PlGF) play a key role in placental angiogenesis in pregnancy, while preeclampsia is associated with elevated serum levels of the soluble receptor for VEGF, also referred to as soluble fms-like tyrosine kinase 1 (sFLT-1), which binds and decreases free levels of VEGF and PlGF resulting in an antiangiogenic state (6–8). Few studies have investigated circulating angiogenic factors in women with preexisting diabetes despite the higher risk of preeclampsia and other vascular complications in this cohort. We previously showed that women with preexisting diabetes and falling insulin requirements have a significantly higher sFLT-1–to–PlGF ratio from 25 weeks onward, which was maintained in the subgroup that developed preeclampsia (9). Two other longitudinal cohort studies examining a population predominately with type 1 diabetes demonstrated that women with diabetes and preeclampsia had increased sFLT-1, decreased PlGF, and an increased sFLT-1–to–PlGF ratio compared with nonpreeclamptic women that preceded the onset of preeclampsia, in keeping with the population without diabetes (10,11). Holmes et al. (12) recently examined the predictive potential of angiogenic factors at 26 weeks in 540 women with type 1 diabetes and showed the ratio of antiangiogenic to proangiogenic factors significantly improved the predictive value of traditional clinic risk factors. However, that study was cross-sectional and did not examine the changes in angiogenic factors prospectively through pregnancy to establish the best time point for these predictive tests. No previously published studies have examined the utility of urinary biomarkers in women with diabetes.

In women with preexisting diabetes, we aim to quantify the urinary PlGF obtained in each trimester to determine whether there is correlation with serum levels. Further, we aim to investigate its prognostic utility with respect to preeclampsia and other adverse pregnancy outcomes compared with serum PlGF and the serum sFLT-1–to–PlGF ratio.

This was a prospective multicenter cohort study conducted between June 2013 and May 2016 at three tertiary referral hospitals (in Western Sydney, Australia) with dedicated diabetes in pregnancy services. Patients were recruited prior to 20 weeks’ gestation, during their first visit to the dedicated diabetes in pregnancy clinic. The study included women over the age of 16 with singleton pregnancies and a diagnosis of preexisting type 1 or type 2 diabetes or a new diagnosis of overt diabetes on a 75-g oral glucose tolerance test before 20 weeks’ gestation (fasting glucose ≥7.0 mmol/L [126 mg/dL], 2-h glucose ≥11.1 mmol/L [200 mg/dL], and/or HbA_1c ≥6.5% [48 mmol/mol]), as per the 2013 World Health Organization criteria (13). For the purpose of analysis, women with overt diabetes in pregnancy were reclassified as having type 1 diabetes if positive for GAD antibodies or type 2 diabetes if negative. This is a substudy of the Falling Insulin Requirements STudy (FIRST) (14), and as this cohort study was designed primarily to investigate the clinical significance of falling insulin requirements in pregnancy, only women requiring insulin treatment during pregnancy were included in the study. Exclusion criteria were pregnancies that did not progress beyond 20 weeks’ gestation or were a multiple gestation or patients not requiring insulin or not complying with regular review for titration. Follow-up time was the length of the pregnancy and the immediate postpartum period. Planned pregnancy was defined as women intentionally attempting to conceive.

Random spot urine samples along with serum samples were obtained from participants during trimester 1 (14 ± 1 weeks), in trimester 2 (24 ± 1 weeks), and twice in trimester 3 (30 ± 1 weeks and 36 ± 1 weeks), centrifuged, and stored at −80°C until testing. Urinary PlGF concentrations were measured using the Quantikine Human ELISA Kit (R&D Systems, Minneapolis, MN) according to the manufacturer’s instructions. The minimal detectable level for the kit was 7 pg/mL. Samples that resulted in undetectable levels were assigned the value 6 pg/mL for statistical analysis. The same kits were used to quantify serum PlGF and serum sFLT-1 levels. Urine creatinine concentrations were measured in all urinary samples for normalization of urinary PlGF using the urinary parameter creatinine ELISA Kit (R&D Systems). All assays were performed in duplicate, and if duplicate results differed by ≥15%, they were repeated.

The primary outcome measure was preeclampsia. All cases of preeclampsia satisfied the International Society for the Study of Hypertension in Pregnancy 2014 criteria (15), which is new onset of hypertension (≥140 mmHg systolic or ≥90 mmHg diastolic pressure) after 20 weeks’ gestation together with one or more of the following: proteinuria >300 mg/day (or equivalent), the presence of any maternal organ dysfunction (renal insufficiency, liver involvement, neurological or hematological complications), or fetal growth restriction. The secondary outcome was measured as a composite of outcomes that may result from placental insufficiency: preeclampsia, small for gestational age (≤10th percentile for gestational age), stillbirth (>20 weeks), and premature delivery (≤34 weeks).

All data were analyzed using IBM SPSS 20 (IBM, Armonk, NY) or SAS 9.4 (SAS Institute, Cary, NC) statistical software. Biomarkers (urinary PlGF uncorrected and corrected for creatinine, serum PlGF, and serum sFLT-1) were not normally distributed and so were log transformed for analysis, and the results were back transformed and presented as a ratio of means. Results presented are for urinary PlGF, corrected for creatinine concentration, although analysis with uncorrected urinary PlGF yielded similar results. Comparisons of the biomarker levels between those with and without preeclampsia or the secondary composite outcome used a two-sample t test. Relationships between the biomarkers and preeclampsia as an outcome were adjusted for the type of diabetes in logistic regression models. Changes over time were analyzed with a generalized linear model to account for the correlation within a woman, with results presented as retransformed means and mean ratios with 95% CIs for the last time period compared with the others. Correlations were done on raw data and used a Spearman rank correlation. Associations were considered significant with P < 0.05. Receiver operating characteristic (ROC) curves were also calculated to assess the discriminative ability of each of the biomarkers at each time point for preeclampsia. The increased discrimination with the addition of biomarkers to logistic models, including preconception HbA_1c and type of diabetes, was analyzed. The area under the curve (AUC), 95% CI, and P value are reported.

There were 40 women (25.3%) with type 1 diabetes, 94 (59.5%) with type 2 diabetes, and 24 (15.2%) diagnosed with overt diabetes during pregnancy, of whom 1 was reclassified as having type 1 diabetes and 23 as having type 2 diabetes. All 27 patients who developed preeclampsia were late onset (>34 weeks’ gestation). Baseline patient characteristics were similar between women who developed preeclampsia and those who did not (Table 1). Of note, a greater proportion of women with type 1 diabetes had preeclampsia compared with type 2 (26.8 vs. 13.8%, P = 0.054). Additionally, patients who developed preeclampsia had a higher HbA_1c and were less likely to have a planned pregnancy than those without preeclampsia, independent of the type of diabetes. Those who planned their pregnancy were more likely to have an HbA_1c <6.5% (48 mmol/mol, P < 0.001), with 82% of women with unplanned pregnancies having an HbA_1c ≥6.5% (48 mmol/mol) compared with 42% of women with a planned pregnancy. Further analysis of preconception HbA_1c revealed HbA_1c ≥6.5% (48 mmol/mol) was significantly associated with the outcome of preeclampsia (P = 0.01).

As pregnancy progressed, urinary and serum PlGF both increased between 14 and 24 weeks’ gestation, then remained relatively unchanged by 30 weeks’ gestation and slightly decreased at 36 weeks’ gestation. In contrast, the sFLT-1–to–PlGF ratio demonstrated an inverse trend, with a decrease between 14 and 24 weeks’ gestation, and remained relatively unchanged by 30 weeks’ gestation and slightly increased at 36 weeks’ gestation. Serum sFLT-1 remained relatively stable between 14 and 30 weeks’ gestation and then increased at 36 weeks’ gestation (Supplementary Fig. 1). There was correlation between urinary and serum PlGF at all time points, with the exception of 14 weeks’ gestation, which showed a weak but statistically significant correlation (Fig. 1 and Supplementary Table 1). A large percentage of the urinary PlGF samples at week 14 and week 36, 49.3% (n = 70) and 36.9% (n = 50), respectively, were below the minimal detectable level of 7 pg/mL compared with 8.1% (n = 11) and 16.5% (n = 21) of urinary PlGF samples at 24 and 30 weeks’ gestation, respectively.

The incidence of preeclampsia within our study cohort was 17.1%. At all time points, those who developed preeclampsia had lower urinary and serum PlGF levels (P < 0.05) and a higher sFLT-1–to–PlGF ratio (P < 0.05) compared with women who did not develop preeclampsia, with the exception of week 14 urinary PlGF values. These results remained significant after adjusting for baseline differences in the type of diabetes. In contrast, serum sFLT-1 levels were similar in those who did and did not develop preeclampsia, except for levels measured at week 36, where women who developed preeclampsia had a significantly higher serum sFLT-1 level (P = 0.01) (Fig. 2 and Supplementary Table 2).

An ROC curve was generated for urine PlGF, serum PlGF, serum sFLT-1, and the serum sFLT-1–to–PlGF ratio at each time point for the outcome of preeclampsia (Fig. 3). Urinary PlGF levels were found to be a poor predictor of preeclampsia, with the 95% CI for the AUC crossing 0.5 for all time points except week 36. In contrast, serum PlGF as a test for preeclampsia in this cohort reached statistical significance at all time points tested and performed marginally better than the serum sFLT-1–to–PlGF ratio as a predictive test for preeclampsia at all time points. Only trimester 3 (week 30 and 36) serum sFLT-1 levels were predictive of preeclampsia (Fig. 3 and Supplementary Table 3). The addition of these biomarkers to logistic regression models with preconception HbA_1c and type of diabetes increased the AUC, improving predictive value (Supplementary Table 3).

Analysis of the secondary composite outcome (Supplementary Table 4) revealed that serum PlGF levels were significantly lower at all tested gestations in women with a secondary outcome compared with those without. Urinary PlGF levels were only significantly lower during trimester 3 (week 30 and 36) in those who developed a secondary outcome, and similarly, the serum sFLT-1–to–PlGF ratio was only significantly higher at trimester 3 time points for the secondary outcome. Throughout pregnancy, serum sFLT-1 showed no significant difference in levels between those who developed a secondary outcome versus those who did not (Supplementary Fig. 2 and Supplementary Table 5).

Through prospective and longitudinal analyses we have demonstrated that with the exception of very early in pregnancy, there is moderate correlation between urine and serum PlGF levels and that urinary PlGF follows a trend similar to that of serum levels as pregnancy progresses. These results are in keeping with the limited published data available from case-control cross-sectional studies in women without diabetes (16–18). Although we found significantly lower levels of urinary PlGF from 24 weeks onward in women who subsequently developed preeclampsia, a large proportion had urinary PlGF levels below the minimal detectable level in early and late pregnancy. Lower urinary PlGF levels relative to serum levels are likely due to the lower amount of PlGF that is filtered through to urine. Therefore, despite a moderate correlation with serum levels and a significant association with preeclampsia, urinary PlGF was inferior to serum PlGF as a predictive test of preeclampsia. Hence, although urinary PlGF presents an attractive noninvasive biomarker for preeclampsia prediction, it is unlikely to have clinical utility until more sensitive assays become commercially available.

Additionally, we demonstrated reduced serum PlGF and an increased serum sFLT-1–to–PlGF ratio in women who developed preeclampsia, which is in keeping with the literature (25). However, serum PlGF performed better as a predictive marker, and the difference in the sFLT-1–to–PlGF ratio between women with and without preeclampsia was largely driven by serum PlGF. This is in contrast to the population without diabetes (19–21) and suggests that preeclampsia predictive models that use the sFLT-1–to–PlGF ratio may not be suitable for use in women with preexisting diabetes. It should be noted that the prognostic utility of the sFLT-1–to–PlGF ratio has been demonstrated to be different in late- and early-onset preeclampsia, with differences being more pronounced in early-onset disease (19,22,23). Consequently, our results may be influenced by the fact that all of the preeclampsia in our cohort was late onset. Our results are in keeping with those of Cohen et al. (11), who demonstrated similar findings in a predominantly type 1 diabetes cohort, with differences in the sFLT-1–to–PlGF ratio at 7–14 weeks, 16–20 weeks, 24–32 weeks, and immediately prior to delivery predominantly influenced by the PlGF difference. Interestingly, compared with normal control subjects, this study also found higher sFLT-1 levels in women with preexisting diabetes who did not develop preeclampsia, while serum PlGF levels were similar (11). Yu et al. (10) also found comparable results among normotensive women with type 1 diabetes. It appears that women with preexisting diabetes have higher levels of sFLT-1, in keeping with their increased vascular risk, and this provides an explanation for the better performance of serum PlGF as a preeclampsia predictive marker in women with preexisting diabetes.

Furthermore, serum PlGF was the only marker with significantly lower levels throughout pregnancy for women who developed our defined secondary composite outcome of factors associated with placental insufficiency—preeclampsia, small for gestational age, stillbirth, and premature delivery. This suggests that in addition to preeclampsia, reduced PlGF levels in pregnancy are associated with other outcomes that result from placental insufficiency and highlights the importance of using serum PlGF rather than the sFLT-1–to–PlGF ratio in women with preexisting diabetes to predict poor outcomes.

Finally, in keeping with the literature, our results have demonstrated that preconception HbA_1c ≥6.5% (48 mmol/mol) is significantly associated with an outcome of preeclampsia and was more likely to occur in women with an unplanned pregnancy (11,12). This reaffirms the importance of prepregnancy planning and optimization of glucose control prior to pregnancy in women with preexisting diabetes. Addition of angiogenic markers to preconception HbA_1c improved the predictive utility. These data have important clinical implications as they could help develop predictive models unique to women with preexisting diabetes by using urinary or serum PlGF, preconception HbA_1c, and other clinical risk factors for earlier identification of a higher-risk cohort for appropriate resource allocation.

We acknowledge that combining women with type 1, type 2, and newly emerged diabetes in pregnancy in our cohort is a limitation of our study. Ideally, all types of diabetes should be studied separately; however, the much larger numbers required for such studies makes this less feasible, and consequently current clinical practice is being guided by studies of women without diabetes. To overcome this, we corrected all of our analyses for type of diabetes, and the results remained robust. Although our cohort was restricted to participants requiring insulin during pregnancy, this is unlikely to affect results as the majority of women with preexisting diabetes require insulin treatment in pregnancy (24). Finally, all our preeclampsia events were late onset, and considering prediction studies in the population without diabetes demonstrate a linear relationship between preeclampsia severity and PlGF levels, prediction may be better in early-onset disease for our population.

In conclusion, proposed predictive models that incorporate the sFLT-1–to–PlGF ratio may not be applicable to women with preexisting diabetes, while serum PlGF is useful in any trimester in preeclampsia prediction with the potential for early implementation of preventative strategies. If more sensitive PlGF kits were made commercially available, a simple and cost-effective urinalysis holds promise for screening, triage, and resource allocation, in particular for resource-poor centers and developing nations.

Funding. The authors would like to acknowledge the Westmead Medical Research Foundation, the Royal Australian and New Zealand College of Obstetricians and Gynaecologists (RANZCOG) Women’s Health Foundation via the Norman Beischer Clinical Research Scholarship, and the Western Sydney Local Health District (WSLHD) Research and Education Network (REN) grant for providing funding for this study.

Duality of Interest. No potential conflicts of interest relevant to this article were reported.

Author Contributions. M.Z. designed the study, performed the ELISA for urinary samples, analyzed the data, and wrote the manuscript. S.P. designed the study and the original cohort study, collected the samples and data, performed the ELISA for serum samples, and reviewed and edited the manuscript. K.Z. performed ELISA on urinary samples and reviewed and edited the manuscript. A.K. analyzed the data, validated the statistical analysis, and reviewed and edited the manuscript. N.W.C. and V.W.L. designed the original cohort study and reviewed and edited the manuscript. V.W.L. and T.I.A. designed the study and reviewed and edited the manuscript. M.Z. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.