Glycated albumin and continuous glucose monitoring metrics across pregnancy in women with pre‐gestational diabetes

Abstract Introduction Glycated albumin (GA), a biomarker reflecting short‐term glycaemia, may be useful to assess glycaemic control in pregnancy. We examined the association between GA and continuous glucose monitoring (CGM) metrics across gestation. Methods In this prospective cohort study including 40 women with pre‐gestational diabetes, blood samples for analysis of GA and glycated haemoglobin A1c (HbA1c) were collected at pregnancy week 12, 20, 24, 28, 32 and 36. In the CGM‐group (n = 19), CGM data were collected from first trimester until pregnancy week 36. Receiver operating characteristic (ROC) curves were used to assess the accuracy of GA and HbA1c to detect poor glycaemic control, using CGM metrics as the reference standard. This study was conducted at Stavanger University Hospital, Norway, in 2016–2018. Results Glycaemic control improved across gestation with more time spent in target range, coinciding with decreased glycaemic variability and lower mean GA level. There was statistically significant correlation between GA and most CGM metrics. The area under the ROC curves (AUC) for detecting time in range <70% and time above range >25% for the pregnancy glucose target 63–140 mg/dl (3.5–7.8 mmol/L) were 0.78 and 0.82 for GA, whereas AUCs of 0.60 and 0.72 were found for HbA1c, respectively. Conclusions Higher GA levels were associated with less time spent in target range, more time spent in the above range area and increased glycaemic variability. GA was more accurate than HbA1c to detect time above range >25% and time in range <70%.

Continuous glucose monitoring (CGM) enables users to monitor their glucose level, providing the opportunity to respond to glucose fluctuations as they occur. 5 With randomized controlled trials showing that CGM is associated with improvements in maternal glycaemic control and neonatal outcomes, 6 the use of CGM in antenatal care is increasing. 7 By recent international consensus for CGM monitoring, the pregnancy glucose target range for type 1 diabetes (T1D) was set to 63-140 mg/dl (3.5-7.8 mmol/L). Women should strive to achieve >70% of time within target range. 8 Currently, there are not provided CGM targets for pregnant women with T2D, due to the lack of evidence and limited data. However, access to CGM for all pregnant women with diabetes is still limited.
Glycated albumin (GA), a biomarker reflecting short-term glycaemia (2-4 weeks) has been suggested to supplement glycated haemoglobin A1c (HbA1c) in monitoring glycaemic control. 9 In diabetic pregnancies where strict glycaemic control is important to reduce adverse maternal/foetal outcomes, a marker reflecting recent glycaemic status is preferable. Moreover, GA may be better than HbA1c to detect glucose variability and fluctuations, which have been associated with increased risk of developing large for gestational age (LGA) foetuses. 10 Furthermore, elevated maternal GA levels may predict perinatal complications. 11 Thus, GA may be a useful tool for detecting and monitoring recent glycaemic control in diabetic pregnancies, and in particular, the glucose fluctuations, not provided by HbA1c.
Haemoglobin A1c is recognized as the gold standard of diabetic survey 12 and was included as a diagnostic criterion for diabetes mellitus in 2011. 13 HbA1c reflects mean glycaemia over the preceding 8-12 weeks. 14 There is a linear relationship between average glucose and HbA1c in pregnancy, but the change in HbA1c reflects a smaller difference in mean glucose compared with that found in non-pregnant adults. 15 Moreover, altered erythrocyte turnover and iron deficiency may influence HbA1c, making it less accurate during pregnancy. 16,17 Despite these limitations, HbA1c is used worldwide in clinical practice to monitor glycaemic control during pregnancy.
Recently, a new high-throughput method for GA measurement using liquid chromatography-tandem mass spectrometry (LC-MS/ MS) was developed in our laboratory. 18 Subsequently, the reference interval for GA in healthy pregnant women was established. 19 The primary aim of this study was to explore the association between GA and CGM metrics across gestation in women with pregestational diabetes. Secondly, we investigated the accuracy of GA and HbA1c to detect poor glycaemic control using CGM metrics as the reference standard.

| Blood glucose data
According to recommendations in the Norwegian guideline, the HbA1c level should be <53 mmol/mol (<7%) in the preconception period and <42 mmol/mol (<6%) from second trimester. Throughout pregnancy, treatment goals for glucose are fasting plasma glucose 63-99 mg/dl (3.5-5.5 mmol/L) and <128 mg/dl (<7.1 mmol/L) 2 h postprandial. 20 CGM were offered to women with poor glycaemic control, or additional challenges such as impaired awareness of hypoglycaemia. Otherwise, self-monitoring of blood glucose with frequent daily measurements (7-10 times a day) was advised. In Norway, the use of CGM during pregnancy has markedly increased over the past years. Seventeen women in the study were already users of CGM before pregnancy, whereas four participants were offered CGM during pregnancy.

| CGM system
Among the CGM users, the majority had Dexcom G4 (Dexcom Inc), whereas one had Freestyle Libre (Abbott) and another used the Medtronic CGM system (Medtronic). The Dexcom G4 device, measures subcutaneous interstitial glucose concentration every 10 s and generates a glucose value every 5 min, available for the user real time. Dexcom G4 requires calibration by the user against capillary plasma glucose twice daily. With the Freestyle Libre system, known as a 'flash' glucose monitor, no calibration is required. The interstitial glucose level is measured every 60 s, a glucose value is generated every 15 min, but the results are available only retrospectively when the sensor is scanned with a reading device. The Medtronic CGM system is also a real time system, generating a glucose value every 5 min.

| Glucose data management
At every visit, available data from self-monitored blood glucose and/ or CGM were downloaded from the internet-based Diasend system (Glooko). For the user of Medtronic CGM system, glucose data were downloaded from CareLink (Medtronic). We included CGM data from the 14 days leading up to each blood sampling at pregnancy week 12, 20, 24, 28, 32 and 36. According to recent consensus on CGM use, we required at least 70% coverage (percentage of time CGM is active) for inclusion in the analysis. 8 From CGM data, we calculated mean glucose level and the percentage of time spent in target range (time in range, TIR), time below range (TBR) and time above range (TAR) for the pregnancy glucose target range 63-140 mg/dl (3.5-7.8 mmol/L). 8 We also calculated time below range <54 mg/dl (<3.0 mmol/L), denoted TBR2.
Measures of glycaemic variability included glucose standard deviation (SD) and coefficient of variation (CV). 8

| Obstetric data and outcomes
Information concerning pregnancy outcome was collected from medical records after delivery. Frequencies of small for gestational age and large for gestational age were calculated using the 10th and 90th percentile according to Gjessing et al. 21 In addition, birth weight centiles and percentage birth weight deviations from the median birth weight for gestational age, were calculated. 21

| Ethical considerations/approval
The study was carried out in accordance with the Helsinki Declaration and was approved by the Regional Committees for

| Statistical analyses
Categorical data are shown as percentages. Continuous variables are presented as mean with SD, or median with interquartile ranges (IQR) for skewed distributions. Differences in clinical characteristics between the CGM and non-CGM group were assessed using independent samples t-test (normal distribution) and Mann-Whitney test (skewed distribution) for continuous data, whereas Chi-squared test was performed for categorical data. A p-value < .05 was considered statistically significant.
Mean values of GA and HbA1c at different time points were estimated in mixed linear models with random intercepts and random effects of time points. Comparison of levels between time points was performed with paired samples t-tests.
Correlation coefficients were used to assess relationships between GA, HbA1c and CGM metrics. The correlation coefficients were estimated allowing for the repeated measures design using the approach outlined by Hamlett et al. 22

| RE SULTS
In all, 42 women were asked to participate in the study and 41 were included. One participant withdrew during the study period, resulting in a total study population of 40 pregnant women. Among these, 26 (65%), 13 (32.5%) and one (2.5%) had type 1 diabetes, type 2 diabetes and maturity onset diabetes of the young (MODY), respectively.
In total, 17 women were CGM-users before pregnancy. Out of the four women offered CGM during pregnancy, one delivered prematurely a week later. For another woman, the CGM raw data were lost, resulting in 19 women with available CGM-data from first trimester to pregnancy week 36. The majority in the CGM group had T1D, whereas the non-CGM group was more heterogeneous. All insulin-pump users were in the CGM group, and most had Animas vibe pumps (Animas Corporation), while three women had either a Paradigm 715 (Medtronic), Minimed 640G (Medtronic) or an Omnipod (Insulet) pump. In contrast, most women used insulin pens in the non-CGM group. Moreover, women in the CGM group were younger and had longer diabetes duration compared with the non-CGM group. Pre-pregnancy HbA1c level, BMI and weight-gain in pregnancy were comparable between the two groups.
Almost one in five women developed preeclampsia, one third delivered an LGA-newborn and two thirds had a vaginal delivery. The clinical characteristics of the total study population, CGM group and non-CGM group are summarized in Table 1.
The majority (82.5%) completed all six blood samples for analyses of GA and HbA1c, whereas five women (12.5%) missed one blood sample and two women (5%) missed two blood samples. The TA B L E 1 Maternal and neonatal characteristics in the total study population, CGM-group and non-CGM group. After exclusion of six 14-days periods with <70% coverage, 103 14-days periods throughout gestation were available for the analysis of CGM-data (mean coverage 92.6%, SD 4.9). The CGM metrics and laboratory markers of glycaemia varied across gestation (Table 2).
We found correlations between GA and mean glucose, TIR, TAR and glucose SD (Table 3). For HbA1c, correlations were found with mean glucose, TAR, TBR and TBR2 ( Table 3). We observed positive associations between GA and TAR, mean glucose, SD and CV ( Figure 3B,D-F), a negative association with TIR ( Figure 3A) and no association with TBR ( Figure 3C). Corresponding scatterplots showing the association between HbA1c and CGMmetrics are presented in Figure S1.

| DISCUSS ION
In this prospective study of pregnant women with pre-gestational diabetes, overall glycaemic control improved across gestation with  Note: Correlation coefficients for repeated measures design with 95% confidence intervals. CGM metrics were calculated from 103 14-days periods across gestation with >70% coverage. Significant correlations are marked in bold.
to assess glycaemia in pregnant women. 23 The improving glycaemic control throughout pregnancy observed in our study using CGM-metrics as the reference standard, was not at all reflected in lower HbA1c levels, in contrast, GA levels decreased throughout the pregnancy. We found high, statistically significant correlation between GA and glucose SD. Although not statistically significant, the positive correlation between GA and glucose CV and an AUC >0.5 for TBR >4% and TBR2 >1%, are in further support of previous findings indicating that high GA may also detect glycaemic variability, 10 including hypoglycaemic fluctuations.
Others have shown that the GA level also decreases during gestation in women with healthy pregnancies. 24,25 The reasons remain unexplained, but might be due to increased turnover of albumin and/ or increased selective loss of GA through glomerular filtration. 25 Although the GA-values are not directly comparable due to different methods for GA-analysis, the observed decrease in mean GA level in Calculations based on 103 14-days periods with >70% coverage. Data presented as mean with 95% confidence intervals, adjusted predictions. CGM, continuous glucose monitoring our study is more prominent (from 12.1% to 9.3%). In comparison, the mean GA level in healthy pregnant women was 9.5% at pregnancy week 24-28 in our previous study, 19 whereas a mean GA level of 11.3% and 10.3% was found in the CGM and non-CGM group at pregnancy week 24 the present study.
Another population where HbA1c has limitation, haemodialysis patients with diabetes, Divani et al. 26 found higher accuracy for GA than HbA1c to detect TIR <50%. None of the glycaemic markers were able to detect TBR. In the current study, for GA, the AUC of 0.66 for TBR2 >1% was not statistically significant, however suggesting that high GA levels may detect hypoglycaemic excursions.
In contrast, HbA1c detected TBR and TBR2 above thresholds with AUCs of 0.30 and 0.32 (the latter not statistically significant), that is high HbA1c levels indicate reduced risk for these CGM metrics.
Albeit an increase in mean percentage of time spent in target range from 59% in first trimester to 68% in third trimester, most  with larger sample sizes are required to better understand the role of GA in diabetic pregnancies, and for establishing optimal cut-off values for detecting poor glycaemic control.

CO N FLI C T O F I NTE R E S T
All authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.