Abstract
Introduction
The manuscripts chosen for this year's article on technology and pregnancy demonstrated advances in our understanding of type 1, type 2, and gestational diabetes (GDM) in pregnant women. From over 2800 published articles, we have chosen 10 that represent the diversity of diabetes technology and pregnancy publications. They encompass improvements in our understanding of continuous glucose monitoring (CGM) in pregnancy, early data on the first commercially available closed-loop system used (off-license) during pregnancy, and advancements in our understanding of screening for GDM.
Important data regarding the accuracy of the Dexcom G6 sensor in pregnant women were published (1). They indicated the Dexcom G6 is safe, accurate, and well-tolerated in women with type 1, type 2, and gestational diabetes. Scott et al. demonstrated how functional data analysis can be used to identify distinct glycemic patterns over the day and night using data from the Continuous Glucose Monitoring in Women with Type 1 Diabetes in Pregnancy Trial (CONCEPTT) (2,3). They noted that summary CGM measures may not detect important temporal differences in maternal glycemia highlighted by functional data analysis. For example, women who subsequently delivered a large-for-gestational-age baby spent 14–16 hours/day with significantly higher daytime glucoses, predominantly after meals. A case series of the first commercially available hybrid closed-loop system (the Medtronic 670G) was published (4). The authors highlighted the importance of careful carbohydrate management and strong insulin-to-carbohydrate doses when using hybrid closed-loop during pregnancy.
A large cohort study from Scotland demonstrated that higher HbA1c levels and body mass index (BMI) are important modifiable risk factors for stillbirth in their population of women with type 1 and 2 diabetes (5). The data also suggest that in type 1 and 2 diabetes pregnancies, delivery before 38 weeks may be appropriate in women with babies at birthweight extremes. A multicenter, noninferiority, randomized controlled trial of a lower glucose threshold (36 mg/dl [2.0 mmol/L]) compared to the standard glucose threshold (47 mg/dl [2.6 mmol/L]) in neonates at risk for hypoglycemia was published (6). Authors found no difference in psychomotor development at 18 months in neonates treated using a lower vs standard glucose threshold; although, the authors suggest that a higher target glucose is required for neonates with persistent hypoglycemia or underlying endocrine or metabolic disorders. Unfortunately, no conclusions can be made regarding targets for neonates of mothers with diabetes, as too few women with diabetes were included in the trial.
In the time of COVID-19, we continue to search for simpler methods to predict and screen for GDM in a way that may minimize healthcare resource utilization and reduce exposure for women during testing. In their article, Artzi et al. used machine learning to develop a risk prediction tool for GDM (7). They developed a model using only nine factors that may reduce GDM screening in women with a previous glucose challenge. While this would need to be studied in a more diverse population, it demonstrates the potential of machine learning to help solve common clinical problems. Retnakaran and Shah used Canadian administrative data to determine if pregravid HbA1c may decrease the need for GDM screening (8). They found that there was no threshold for HbA1c that would meaningfully reduce the need for oral glucose tolerance testing without missing a number of cases of GDM. However, pregravid HbA1c was strongly associated with GDM, with a 22% increase in GDM for each 0.1% increase in HbA1c.
New translational research offers insight into the mechanism of improved β-cell function related to breastfeeding, including increased serotonin production and stimulation (9). Waters et al. sought to examine insulin sensitivity and production in the postpartum period in women with GDM (10). They found that insulin sensitivity and production increase rapidly in the immediate postpartum period, with improvements in insulin production persisting over 6–12 weeks. These data suggest that early screening for glucose intolerance postpartum in women with GDM warrants further investigation. Lastly, a pragmatic randomized controlled trial of early vs routine screening for GDM did not demonstrate an improvement in obstetric or neonatal outcomes (11). However, authors found that participants with early screening were more likely to be started on insulin and deliver sooner. In addition to this study, ongoing trials in different populations may help to determine the role, if any, for the early screening and treatment of GDM.
Castorino K, Polsky S, O'Malley G, Levister C, Nelson K, Farfan C, Brackett S, Puhr S, Levy C
Scott EM, Feig DS, Murphy HR, Law GR on behalf of the CONCEPTT Collaborative Group
Polsky S, Akturk HK
Mackin ST, Nelson SM, Wild SH, Colhoun HM, Wood R, Lindsay RS, SDRN Epidemiology Group and Scottish Diabetes Group Pregnancy subgroup
van Kempen AA, Eskes PF, Nuytemans DHGM, van der Lee JH, Dijksman LM, van Veenendaal NR, van der Hulst FJPCM, Moonen RMJ, Zimmermann LJI, van 't Verlaat EP, van Dongen-van Baal M, Semmekrot BA, Stas HG, van Beek RHT, Vlietman JJ, Dijk PH, Termote JUM, de Jonge RCJ, de Mol AC, Huysman MWA, Kok JH, Offringa M, Boluyt N, HypoEXIT Study Group
Artzi NS, Shilo S, Hadar E, Rossman H, Barbash-Hazan S, Ben-Haroush A, Balicer RD, Feldman B, Wiznitzer, Segal E
Retnakaran R, Shah BR
Moon JH, Kim H, Kim H, Park J, Choi W, Choi W, Hong HJ, Ro HJ, Jun S, Choi SH, Banerjee RR, Shong M, Cho NH, Kim SK, German MS, Jang HC, Kim H
Waters TP, Kim SY, Sharma AJ, Schnellinger P, Bobo JK, Woodruff RT, Cubbins LA, Haghiac M, Minium J, Presley L, Wolfe H, Hauguel-de Mouzon S, Adams W, Catalano PM
Harper LM, Jauk V, Longo S, Biggio JR, Szychowski JM, Tita AT
Performance of the Dexcom G6 continuous glucose monitoring system in pregnant women with diabetes
Castorino K1, Polsky S2, O'Malley G3, Levister C3, Nelson K1, Farfan C1, Brackett S2, Puhr S 4, Levy C3
1Sansum Diabetes Research Institute, Clinical Research, CA; 2University of Colorado Denver, Barbara Davis Center for Childhood Diabetes, Denver, CO; 3Icahn School of Medicine at Mount Sinai, New York, NY; 4Dexcom, Inc., San Diego, CA
Background
Data on the performance of real-time continuous glucose monitoring (CGM) systems during pregnancy are limited. The aim of this study was to examine the performance of the Dexcom G6 sensor in pregnant women with type 1 (T1D), type 2 (T2D), and gestational diabetes (GDM) during the second and third trimesters.
Methods
In this prospective observational cohort study, 32 pregnant women with diabetes (20 T1D, 3 T2D, 9 GDM) were enrolled across 3 U.S. study sites. Participants wore two G6 sensors for 10 days on the abdomen or upper buttock (unmasked) and/or on the posterior upper arm (masked). YSI blood glucose values were obtained every 30 minutes for 6 hours in a clinical setting between days 3 and 7 of sensor wear. Mean absolute relative difference (MARD) between CGM-YSI pairs was calculated. Accuracy metrics also included the proportion of CGM glucose values that were within ± 20% of paired YSI reference values >100 mg/dL or ± 20 mg/dL of YSI values ≤100 mg/dL.
Results
A total of 68 G6 sensors were applied to 19 women in their second trimester and 13 women in their third trimester, with 26 (38%) sited on the arm, 25 (37%) on the abdomen, and 17 (25%) on the buttock. Of the 734 CGM-YSI glucose pairs collected, 92.5% of CGM values were within ± 20% of paired YSI reference values >100 mg/dL or ± 20 mg/dL of YSI values ≤100 mg/dL. The overall MARD was 10.3%, with MARD for sensors worn on the abdomen of 11.5%, upper buttock 11.2%, and posterior upper arm 8.7%. There were no device-related adverse events.
Conclusions
The Dexcom G6 CGM system is safe and accurate for use during pregnancies complicated by T1D, T2D, and GDM. Pregnant women tolerated sensor wear well, regardless of sensor site location.
Comment
These data are important as most CGM systems do not include indications for use in pregnancy, and the previous G6 label carried a warning against use in pregnancy. Confirmation that the Dexcom G6 CGM system is safe and accurate is an important step toward improving CGM use during pregnancy, with potential for improved antenatal glucose levels and neonatal health outcomes. Data regarding sensor accuracy at lower glucose levels are limited. However, overall sensor accuracy metrics are comparable to those in nonpregnant adults; although, accuracy is best for sensors placed on the arm during pregnancy. More data on the use of customizable data alerts, share features, and summary reports specific to pregnancy will be of interest.
Continuous glucose monitoring in pregnancy: importance of analyzing temporal profiles to understand clinical outcomes
Scott EM1, Feig DS2, Murphy HR3, Law GR4 on behalf of the CONCEPTT Collaborative Group
1Department of Clinical and Population Science, Leeds Institute of Cardiovascular and Metabolic Medicine, University of Leeds, Leeds, UK; 2Department of Medicine, Sinai Health System, Toronto, Canada; 3Division of Maternal Health, St Thomas' Hospital, King's College London, London, UK; 4School of Health and Social Care, University of Lincoln, Lincoln, UK
Background
With 288 measurements per day, real-time CGM provides a detailed picture of maternal glycemic control throughout pregnancy. However, summary measurements of CGM do not adequately capture daily glycemic excursions and patterns. Using functional data analysis (FDA), this study aimed to identify maternal glycemic patterns associated with CGM use vs standard glucose monitoring, insulin pump use vs multiple daily injections (MDI), and large-for-gestational-age vs not-large-for-gestational-age.
Methods
This study included 200 women who participated in the Continuous Glucose Monitoring in Women with Type 1 Diabetes in Pregnancy Trial (CONCEPTT) (n=100 randomized to CGM and n=100 in the control group). Summary CGM measures and FDA were used to examine CGM profiles at baseline, 24-, and 34-weeks gestation in the following three groups: participants randomized to CGM vs standard glucose monitoring, using insulin pump vs MDI, and with a large-for-gestational-age vs not-large-for-gestational-age infant.
Results
Functional data analysis demonstrated that participants using CGM had significantly lower glucose measurements from 8:00 to 12:00 and 16:00 to 19:00 (7 hours/day). Participants using pump therapy had significantly lower glucose levels from 7:30 to 11:30 and 20:00 to 21:30 (5 hours/day) but significantly higher glucose from 3:00 to 6:00, 13:00 to 18:00, and 20:30 to 00:30 (12 hours/day) at 24-weeks gestation. Lastly, FDA showed that participants with a large-for-gestational-age infant had significantly higher glucose measurements for 4.5 hours/day at baseline, 16 hours/day at 24 weeks, and 14 hours/day at 34 weeks predominantly during daytime hours.
Conclusions
Functional data analysis was able to identify glycemic patterns throughout a 24-hour period associated with maternal indicators, use of CGM or insulin pump, and the neonatal outcome large for gestational age.
Comment
This study highlights that CGM summary metrics may miss important differences in glycemic profiles throughout a 24-hour period. Functional data analysis allowed authors to identify substantial differences at varying time frames in maternal glucose in the three groups studied. As CGM in pregnancy continues to gain more widespread use, identifying these patterns using FDA may not only aid in the understanding of the pathophysiology of various glycemic-related complications but also help diabetes clinicians and women with diabetes identify and target patterns to reduce adverse outcomes.
Case series of a hybrid closed-loop system used in pregnancies in clinical practice
Polsky S, Akturk HK
Barbara Davis Center for Childhood Diabetes, University of Colorado, Denver, CO
Background
Data on the performance of the first commercially available hybrid closed-loop system, the Medtronic 670G, during type 1 diabetes (T1D) pregnancy are limited. Polsky et al. present three cases of pregnant women with T1D who used the Medtronic 670G system off-label in routine clinical care.
Methods
The potential risks and benefits of using off-label closed-loop were discussed with each participant, including the glucose target of 120 mg/dl (6.7 mmol/L), which is higher than recommended outside of the postprandial time periods.
Results
Case 1 presented at 4-weeks gestation in her second pregnancy, having had a previous miscarriage at 7-weeks gestation. She had a long duration of T1D (31 years) and conceived via in vitro fertilization with a pre-pregnancy HbA1c of 6.2%. During the first trimester she maintained a mean CGM glucose of 116±44 mg/dL (6.4±2.4 mmol/L) with 13% time below 70 mg/dl (3.9 mmol/l) and 78% time in the nonpregnant target range of 70–180 mg/dl (3.9–10.0 mmol/L). At 18-weeks gestation her mean CGM glucose increased to 134±48 mg/dL (7.4±2.7 mmol/L) and she started using the Medtronic 670G system in auto-mode, with a mean CGM glucose level of 127±38 mg/dL (7.1±2.1 mmol/L) between gestation of 18–28 weeks. She maintained mean glucose levels of 122±34 mg/dL (6.8±1.9 mmol/L) with 90% time in the nonpregnant target range of 70–180 mg/dl (3.9–10.0 mmol/L), with an average total daily insulin dose of 43 units/day (74% bolus, 26% basal) throughout the third trimester. She developed preeclampsia with proteinuria and delivered a large birthweight baby boy (8 lb 14.5 oz, 4040 g) with mild neonatal hypoglycemia and hyper-bilirubinemia by caesarean section at 37 weeks 6 days. She maintained safe glucose levels (mean 126±44 mg/dL [7±2.4 mmol/L], 83% time in the nonpregnant target range with 6% time below range) with a TDD of 32 units (72% bolus, 28% basal) using hybrid closed-loop postpartum.
Case 2 presented at 6-weeks gestation with an unplanned first pregnancy and a booking HbA1c of 9.1%. She had a short duration of T1D (6.5 years). During the first trimester she maintained a mean CGM glucose of 171±62 mg/dL (9.5±3.4 mmol/L), with 57% time in the nonpregnant target range 70–180 mg/dL (3.9–10 mmol/L) and 40% time >180 mg/dL (10 mmol/L). During the second trimester her mean glucose level was 155±53 mg/dL (8.6±2.9 mmol/L), with 69% time in the nonpregnant target range 70–180 mg/dL (3.9–10 mmol/L) and 29% time >180 mg/dL (10 mmol/L). Her glucose levels improved during the third trimester—143±45 mg/dL (7.9±2.5 mmol/L), 76% 70–180 mg/dL (3.9–10 mmol/L), 21% >180 mg/dL (10 mmol/L)—with TDD of 98 units/day (37% basal, 63% bolus). She gained 38.5 kg during pregnancy and delivered a large birthweight baby girl (7 lb 6.5 oz, 3359 g) by emergency caesarean section at 37 weeks.
Case 3 was a young woman (aged 24 years) with long duration of T1D (19 years) complicated by proliferative retinopathy and neuropathy who presented at 8-weeks gestation with an unplanned pregnancy, conceived while taking lisinopril and pregabalin, with a booking HbA1c of 11.7%. She was admitted for initial inpatient glucose management. During the first trimester her use of hybrid closed-loop was suboptimal (38% time in auto-mode), with mean glucose levels of 199±86 mg/dL (11.1±4.8 mmol/L) and 51% >180 mg/dL (10 mmol/L). She had lower glucose levels (161±62 mg/dL [8.9±3.4 mmol/L], 32% >180 mg/dL [10 mmol/L]) despite painful neuropathy during the second trimester and developed severe preeclampsia at 28 weeks, with delivery of a large birthweight (4 lb 7.6 oz, 2030 g) baby girl requiring neonatal care at 31 weeks.
Conclusions
This case series describes the use of hybrid closed-loop in a broad patient population of pregnant women with HbA1c values ranging from 6.2 to 11.7%. They suggest that careful attention to carbohydrate intake and strong insulin carbohydrate doses are as important with hybrid closed loop as in normal insulin delivery.
Comment
There are three main limitations of using the Medtronic 670G system in auto-mode during T1D pregnancy. Firstly, the target glucose of 120 mg/dL (6.7 mmol/L) is set too high, and secondly, a proportional-integral-derivative algorithm is not ideal for optimal pregnancy glucose levels. Finally, the time in range for CareLink® Clinical Software reports cannot be changed to the recommended pregnancy time in range of 63 to 140 mg/dL (3.5–7.8 mmol/L). Future studies using model predictive control algorithms and tighter glucose targets will be of interest.
Factors associated with stillbirth in women with diabetes
Mackin ST1, Nelson SM2, Wild SH3, Colhoun HM4, Wood R5, Lindsay RS1, SDRN Epidemiology Group and Scottish Diabetes Group Pregnancy subgroup
1Institute of Cardiovascular and Medical Sciences, University of Glasgow, Glasgow, UK; 2School of Medicine, University of Glasgow, Glasgow, UK; 3Usher Institute of Population Health Science and Informatics, University of Edinburgh, Edinburgh, UK; 4Institute of Genetics and Molecular Medicine, University of Edinburgh, Edinburgh, UK; 5ISD Scotland, Edinburgh, UK
Background
To explore the maternal and fetal risk factors associated with stillbirth in a large national cohort of women with diabetes.
Methods
The researchers analyzed a retrospective cohort of 5392 singleton deliveries in women with diabetes (3778 T1D, 1614 T2D) in Scotland from April 1998 to June 2016.
Results
There were 98 stillbirths (61 T1D, 37 T2D), with a higher stillbirth rate in T2D pregnancies (16.1/1000 T1D, 22.9/1000 type 2). Maternal risk factors for stillbirth in both T1D and T2D included higher prepregnancy HbA1c levels (T1D OR 1.03, 95% CI 1.01–1.04, and T2D OR 1.02, 95% CI 1.00–1.04), with third trimester HbA1c also important in T1D (OR 1.06 95% CI 1.04–1.08) and maternal BMI important in T2D (OR 1.07 95% CI 1.01–1.14). The stillbirth risk was higher in small compared to appropriate-for-gestational-age babies (sixfold higher in T1D, threefold higher in T2D). The stillbirth risk was also twofold higher in large birthweight babies (>95th percentile) of women with T2D, with a preponderance of stillbirths in male compared to female fetuses in T2D pregnancies. The highest rates were in deliveries that occurred in week 38 in T1D (7/1000 95% CI 4–13) and in week 39 in T2D (9/1000 95% CI 2–29).
Conclusions
Higher HbA1c levels and higher BMI are key modifiable maternal risk factors for stillbirth in T1D and T2D. Fetal growth restriction and small birthweight are associated with increased risk in T1D, with male sex, small birthweight, and large birthweight important in T2D. Delivery before 38 weeks may be appropriate in women with babies at birthweight extremes.
Comment
Maternal blood glucose is the key modifiable risk factor for adverse perinatal outcome in type 1 and type 2 diabetes, with prepregnancy HbA1c appearing to be more important than second and third trimester values in type 2 diabetes. The etiology of stillbirth in women with diabetes remains unclear, but more vigilant attention to prepregnancy care, maternal obesity, antenatal glucose, and fetal growth profiles are all needed for improved outcomes in women with type 1 and type 2 diabetes.
Lower versus traditional treatment threshold for neonatal hypoglycemia
van Kempen AA1, Eskes PF2, Nuytemans DHGM3, van der Lee JH4, Dijksman LM5, van Veenendaal NR1, van der Hulst FJPCM6, Moonen RMJ7, Zimmermann LJI8, van 't Verlaat EP9, van Dongen-van Baal M10, Semmekrot BA11, Stas HG12, van Beek RHT13, Vlietman JJ14, Dijk PH15, Termote JUM16, de Jonge RCJ17, de Mol AC18, Huysman MWA19, Kok JH3, Offringa M20, Boluyt N21, HypoEXIT Study Group
1OLVG, Department of Pediatrics, Amsterdam, The Netherlands; 2Meander Medical Center, Department of Pediatrics, Amersfoort, The Netherlands; 3Academic Medical Center, Emma Children's Hospital, Department of Neonatology, Amsterdam, The Netherlands; 4University of Amsterdam, Pediatric Clinical Research Office, Amsterdam, The Netherlands; 5St. Antonius Hospital, Department of Research and Epidemiology, Nieuwegein, The Netherlands; 6Zaans Medical Center, Department of Pediatrics, Zaandam, The Netherlands; 7Zuyderland Medical Center Heerlen, Department of Pediatrics, Sittard-Geleen, The Netherlands; 8Maastricht University Medical Center, Department of Paediatrics-Neonatology, Schools of Oncology and Developmental Biology (GROW) and NUTRIM, Maastricht, The Netherlands; 9Erasmus MC-Sophia, Department of Neonatology, Rotterdam, The Netherlands; 10St Antonius Hospital, Department of Pediatrics, Nieuwegein, The Netherlands; 11Canisius-Wilhelmina Hospital, Department of Pediatrics, Nijmegen, The Netherlands; 12Maasstad Hospital, Department of Pediatrics, Rotterdam, The Netherlands; 13Amphia Hospital, Department of Pediatrics, Breda, The Netherlands; 14Rijnstate Hospital, Department of Pediatrics, Arnhem, The Netherlands; 15University of Groningen, University Medical Center Groningen, Beatrix Children's Hospital, Department of Neonatology, Groningen, The Netherlands; 16University Medical Center Utrecht, Wilhelmina Children's Hospital, Department of Neonatology, Utrecht, The Netherlands; 17Vrije Universiteit, Department of Neonatology, MB Amsterdam, The Netherlands; 18Albert Schweitzer Hospital, Department of Pediatrics, Dordrecht, The Netherlands; 19St. Franciscus Gasthuis, Department of Pediatrics, Rotterdam, The Netherlands; 20The Hospital for Sick Children, Division of Neonatology/Child Health Evaluative Sciences, University of Toronto, Toronto, Canada; 21National Health Care Institute (ZINL), Diemen, The Netherlands
Background
Asymptomatic mild hypoglycemia (defined as glucose concentration 36–46 mg/dl or 2.0–2.5 mmol/L) is a common metabolic problem affecting up to one in three newborn babies, with potential for brain injury and longer-term cognitive impairment. However, consensus on a safe treatment threshold is lacking.
Methods
HypoEXIT (Hypoglycemia–Expectant Monitoring versus Intensive Treatment) trial was a multicenter, noninferiority randomized controlled trial involving 689 otherwise healthy newborns with risk factors for hypoglycemia (late preterm defined as gestation of 35–37 weeks, small for gestational age, large for gestational age, mother with diabetes) but without initial severe hypoglycemia (≤35 mg/dl or 1.9 mmol/L). The hypothesis was that a lower treatment threshold of 36 mg/dl (2.0 mmol/L) would be noninferior to a standard threshold of 47 mg/dl (2.6 mmol/L) for the primary outcome measure of psychomotor development at 18 months. The lower treatment threshold was considered noninferior if the between-group difference was less than the clinically important difference of 7.5 points (approximately 1 month of developmental delay) in the Bayley Scales of Infant and Toddler Development, third edition (Bayley-III-NL), for which the mean (SD) score is 100 (15) with a range from 50–150. Higher scores reflect more advanced development.
Results
Bayley-III-NL scores were available for 287/348 (82.5%) of the lower threshold intervention group (mean 102.9±0.7 [cognitive] and 104.6±0.7 [motor]) and 295/341 (86.5%) of the standard threshold control group (mean 102.2±0.7 [cognitive] and 104.9±0.7 [motor]), meaning that the inferiority limit was not crossed. The mean glucose values were 57±0.4 mg/dl (3.2±0.02 mmol/L) and 61±0.5 (3.4±0.03), with fewer hypoglycemia episodes (57% vs 47%) and less severe hypoglycemic events (33 vs 18 events; 10% vs 5%) in the standard threshold group. Early breastfeeding rates were higher in the intervention group (28% vs 22%), but there were more breastfed infants in the standard care arm after 3 months (21% vs 29%). Bayley-III-NL scores were not correlated with the number or severity of hypoglycemic episodes.
Conclusion
The lower glucose threshold value to start treatment for asymptomatic moderate hypoglycemia did not lead to worse psychomotor development at 18 months. However, treatment begun at the higher glucose threshold of 47 mg/dl (2.6 mmol/L) prevented infants from having recurrent (≥4 episodes) and severe episodes of hypoglycemia. The authors emphasized the need for a higher target glucose in newborns with persistent hypoglycemia or other endocrine or metabolic disorders. Due to smaller-than-expected numbers, no conclusions can be made for infants of mothers with diabetes.
Comment
The definition of neonatal hypoglycemia remains contentious and more data are needed to better understand the risks and benefits of different treatment thresholds in infants of mothers with diabetes.
Prediction of gestational diabetes based on nationwide electronic health records
Artzi NS1,2, Shilo S1–3, Hadar E4,5, Rossman H1,2, Barbash-Hazan S4, Ben-Haroush A4,5, Balicer RD6,7, Feldman B6, Wiznitzer 4,5, Segal E1,2
1Department of Computer Science and Applied Mathematics, Weizmann Institute of Science, Rehovot, Israel; 2Department of Molecular Cell Biology, Weizmann Institute of Science, Rehovot, Israel; 3Pediatric Diabetes Unit, Ruth Rappaport Children's Hospital, Rambam Healthcare Campus, Haifa, Israel; 4Helen Schneider Hospital for Women, Rabin Medical Center, Petach Tikva, Israel; 5Sackler Faculty of Medicine, Tel Aviv University, Tel Aviv, Israel; 6Clalit Research Institute, Clalit Health Services, Tel Aviv, Israel; 7Department of Public Health, Faculty of Health Sciences, Ben-Gurion University, Beer-Sheva, Israel
Background
Gestational diabetes (GDM) remains a common complication in pregnancy and its incidence continues to rise. The screening for and treatment of GDM remain resource intensive. Risk prediction scores have been evaluated; however, current risk prediction scores lack the sensitivity and specificity to substantially reduce testing and are not frequently used in clinical practice.
Methods
This large retrospective cohort study in Israel used machine learning to predict the risk of GDM based on electronic health records. The study population was split into a training set and three validations sets. Clinical features were identified. A gradient-boosting machine model built with decision-tree-based learners was used to develop predictions. In addition, a simpler prediction model was developed based on a minimal number of clinical factors. The area under the receiver operating curve (auROC) was used to evaluate model performance. The prediction models developed were compared to a baseline risk score established using a modified National Institute of Health questionnaire to determine GDM risk.
Results
A total of 588,622 pregnancies were included in this study, with 451,402 pregnancies used in the training set and 137,220 used in the three validation sets. In the included pregnancies, 3.9% were diagnosed with GDM. A total of 2355 clinical features available prior to 20-weeks gestation were identified from the electronic health records. The prediction model developed using these features achieved an auROC of 0.854 vs an auROC of 0.682 using the baseline risk score. A simpler prediction model using only nine features achieved an auROC of 0.799. Using this model as a GDM screening tool to avoid testing low-risk individuals and permitting that 20% diagnoses be missed, 79% of women with a glucose challenge in their previous pregnancy could avoid testing.
Conclusions
The risk prediction tools developed outperformed an assessment of classic risk factors in the assessment of GDM risk. An additional simpler model including only nine factors may be able to reduce GDM screening in women with a previous glucose challenge.
Comment
This study demonstrated the promise and power of machine learning to develop risk prediction tools that have the potential to improve clinical practice. The nine-factor prediction model could allow simple risk assessment using a web-based tool or app. This prediction model may be able to reduce testing as well as identify high-risk women who may benefit from earlier intervention. The reduction in testing may be especially relevant during the COVID-19 pandemic, which has seen the introduction of widespread social distancing measures, self-isolation, reductions in public transport, and a strain on healthcare resources. However, given the population studied, the results may not be generalizable, so validation in other populations is still required.
Pregravid HbA(1c) and glucose measurement to rule out future gestational diabetes mellitus and reduce the need for oral glucose tolerance testing in pregnancy
Retnakaran R1–3, Shah BR3–6
1Leadership Sinai Centre for Diabetes, Mount Sinai Hospital, Toronto, Canada; 2Lunenfeld-Tanenbaum Research Institute, Mount Sinai Hospital, Toronto, Canada; 3Division of Endocrinology, University of Toronto, Toronto, Canada; 4Institute for Clinical and Evaluative Sciences, Toronto, Canada; 5Department of Medicine, Sunnybrook Health Sciences Centre, Toronto, Canada; 6Institute for Health Policy Management and Evaluation, University of Toronto, Toronto, Canada
Background
Screening for gestational diabetes (GDM) for all women in pregnancy remains challenging. Many guidelines recommend a two-step screening process that requires healthcare resources as well as the time pregnant individuals sit in the laboratory. Simplifying GDM screening may decrease healthcare resource utilization as well as improve adherence to testing. The aim of this study was to assess if pregravid HbA1c or glucose may rule out GDM prior to pregnancy.
Methods
This was a retrospective cohort study in Ontario, Canada, that took place from 2008 to 2015. Administrative databases were used to identify women without preexisting diabetes who had HbA1c or glucose testing prior to a singleton live-birth pregnancy. Included individuals were randomly assigned to the derivation or validation cohort. Logistic regression was used to assess the association between pregravid HbA1c or glucose and a diagnosis of GDM, adjusting for age, ethnicity, income, and rurality. A negative predictive value (NPV) was calculated to determine if a pregravid HbA1c threshold that would rule out GDM could be identified.
Results
Pregravid HbA1c was a strong predictor of GDM (OR 7.30 [95% CI 6.57, 8.11]). For each 0.1% increase in HbA1c, the odds of GDM rose by 22%. Both fasting glucose and random glucose were associated with GDM (OR 3.11 [95% CI 2.94, 3.30] and OR 1.63 [95% CI 1.58, 1.68] respectively), though the magnitude of association was not as strong. The optimal NPV was 98.2% using an HbA1c cutoff of ≤4.5%. However, this low cutoff would only remove the requirement for an oral glucose tolerance test (OGTT) in 0.3% of individuals in the validation cohort. To avoid an OGTT in 20.5% of individuals using an HbA1c cutoff of ≤5.2%, 8.6% cases of GDM would be missed.
Conclusions
While pregravid HbA1c was an excellent predictor of GDM, there was no threshold identified that would meaningfully reduce the need for an OGTT without missing cases of GDM.
Comment
This study examined an important question: can we use blood testing already performed to reduce the requirement for OGTTs pregnancy? While considered by many to be the “gold standard” for GDM testing, the OGTT is not without its faults, including poor reproducibility, high cost, and the time required of individuals to wait in the laboratory (12,13). Certainly, the use of pregravid HbA1c would be a welcome replacement. However, much like the studies examining HbA1c in pregnancy, this study was unable to find a cutoff HbA1c that could be used to rule out GDM in high numbers without losing specificity. Nonetheless, pregravid HbA1c was identified as a strong predictor of GDM, with a 22% increase in GDM with each 0.1% increase in HbA1c.
Lactation improves pancreatic β cell mass and function through serotonin production
Moon JH1,2, Kim H1,3, Kim H1, Park J1, Choi W1, Choi W1, Hong HJ4, Ro HJ5,6, Jun S5,6, Choi SH2, Banerjee RR7, Shong M4, Cho NH8, Kim SK9, German MS10, Jang HC2, Kim H1
1Graduate School of Medical Science and Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Korea; 2Department of Internal Medicine, Seoul National University College of Medicine, Seoul, Korea; 3Department of Biochemistry, College of Medicine, Chungnam National University, Daejeon, Korea; 4Research Center for Endocrine and Metabolic Diseases, Chungnam National University School of Medicine, Daejeon, Korea; 5Center for Research Equipment, Korea Basic Science Institute, Cheongju, Korea; 6Center for Convergent Research of Emerging Virus Infection, Korea Research Institute of Chemical Technology, Daejeon, Korea; 7Division of Endocrinology, Diabetes and Metabolism, Department of Medicine, University of Alabama School of Medicine, Birmingham, AL; 8Department of Preventive Medicine, Ajou University School of Medicine, Suwon, Korea; 9Department of Developmental Biology and Stanford Diabetes Research Center, Stanford University School of Medicine, Stanford, CA; 10Diabetes Center, Hormone Research Institute and Department of Medicine, University of California San Francisco, San Francisco, CA
Background
The benefits of breastfeeding for the maternal-child dyad are well described. One such benefit is the improvement in maternal glucose metabolism both during lactation and thereafter. This study examined human participants and mice models to better understand β-cell function during lactation.
Methods
The human study was a prospective cohort study including participants with gestational diabetes (GDM) or gestational impaired glucose tolerance. Postpartum assessment included a 75 g OGTT, and insulin sensitivity was evaluated using the Matsuda index. In the mouse study, animals were randomized to lactating or nonlactating groups (n=3 per group). Metabolic profiles were assessed 3 weeks postpartum. After 3 weeks, the mice pups were weaned and the metabolic profiles were assessed again at 6 weeks after delivery. A multiparous study was also performed with a metabolic assessment at 8 to 12 weeks from the last delivery in mice who were and were not lactating.
Results
Of the 174 human participants, 85 and 99 participants did and did not breastfeed. At mean follow-up of 3.6 years postpartum, individuals who did not breastfeed had a deterioration of glucose tolerance, whereas those who breastfeed maintained a similar glucose tolerance compared to their 2-month testing. Insulin sensitivity was higher in the breastfeeding group both at 2 months and at repeat testing. In the animal study, lactated mice had improved glucose tolerance and insulinogenic index at 3 weeks. At 6-week follow-up, glucose tolerance remained improved but the difference in insulin tolerance was no longer apparent. Multiparous lactated mice demonstrated higher glucose tolerance and β-cell function up to 12-weeks postpartum. Mouse models demonstrated that lactation stimulates serotonin production in β-cells via prolactin. Serotonin then stimulates β-cell proliferation through serotonin receptor 2B. Additionally, intracellular serotonin mitigates intracellular oxidative stress by scavenging reactive oxidative species from β-cells.
Conclusions
This study demonstrated improved metabolic profiles both during lactation and post-lactation in humans and mice. Increased β-cell proliferation was mediated by increased serotonin production and serotonin stimulation of β-cells in mouse models.
Comment
This translational research study offers insight into the mechanism of improved β-cell function related to breastfeeding. If the study results can be reproduced, future studies examining modulation of serotonin production and function in the β-cell may offer additional insight into the role of serotonin in β-cells during and after lactation.
Longitudinal changes in glucose metabolism in women with gestational diabetes, from late pregnancy to the postpartum period
Waters TP1, Kim SY2, Sharma AJ2,3, Schnellinger P4, Bobo JK5, Woodruff RT6, Cubbins LA5, Haghiac M4, Minium J4, Presley L4, Wolfe H7, Hauguel-de Mouzon S4, Adams W8, Catalano PM4
1Department of Obstetrics and Gynecology, Loyola University Medical Center, Maywood, IL; 2Division of Reproductive Health, Centers for Disease Control and Prevention, Atlanta, GA; 3U.S. Public Health Service Commissioned Corps, Atlanta, GA; 4Department of Obstetrics and Gynecology, Cleveland, OH; 5Health and Analytics, Battelle Memorial Institute, Seattle, WA; 6Health and Analytics, Battelle Memorial Institute, Durham, NC; 7Department of Obstetrics and Gynecology, University Hospital, Cleveland, OH; 8Department of Public Health Sciences, Loyola University Chicago, Maywood, IL
Background
The purpose of this study was to examine the postpartum changes in insulin sensitivity, insulin production, and disposition index (DI) in women with gestational diabetes (GDM), compared to late pregnancy. If postpartum changes were comparable in the early (1–5 days) and later (6–12 weeks) time points, an earlier oral glucose tolerance test (OGTT) before hospital discharge may be an option for women with GDM.
Methods
The Matsuda insulin sensitivity index (ISOGTT), insulin response, and DI were calculated following a 75 g OGTT at three time points: T1 late pregnancy (34–37 weeks), T2 early postpartum (1–5 days), and T3 late postpartum (6–12 weeks) in 27 women with GDM based on a 100 g OGTT using the Carpenter–Coustan criteria. Changes in ISOGTT, insulin response, and DI were correlated with changes in maternal adipokines, cytokines, and lipids.
Results
Women with GDM (mean age 31.0 years; BMI 36.2 kg/m2; 63% insulin use; ethnicity: 13 White, 10 Black, 3 Hispanic, 1 Asian) had a significant increase in insulin sensitivity and disposition index and a significant decrease in insulin production in both the early and late postpartum periods. Their fasting and 120-minute OGTT glucose levels were 4.9, 4.3, and 5.4 mmol/L and 9.2, 8.3, and 7.1 mmol/L at T1, T2, and T3, respectively. The changes in insulin sensitivity, production, and disposition index were not correlated with longitudinal changes in maternal bodyweight, lipids, adipokines, or cytokines.
Conclusions
Insulin sensitivity and insulin production increase rapidly in the first few days postpartum, with improvements in insulin production persisting over 6–12 weeks. The immediate postpartum improvement in insulin sensitivity was not sustained, and deteriorated slightly by 6–12 weeks. Early screening for glucose intolerance in the first few days postpartum before maternal hospital discharge is feasible and warrants further evaluation.
Comment
These data provide physiological information confirming significant postpartum improvements in insulin sensitivity and insulin disposition within 48 hours of delivery. Similar changes were observed at 6–12 weeks, suggesting that the postpartum changes in maternal glucose metabolism and insulin action occur immediately after birth and are unrelated to postpartum changes in body weight, lipids, adipokines, or cytokines. Avoiding excess gestational weight gain and postpartum weight retention is a reasonable first step to restoring insulin sensitivity in women with GDM.
Early gestational diabetes screening in obese women: a randomized controlled trial
Harper LM1, Jauk V1, Longo S2, Biggio JR1,2, Szychowski JM1, Tita AT1
1Department of Obstetrics and Gynecology, Center for Women's Reproductive Health, University of Alabama at Birmingham, Birmingham, AL; 2Women's Services Center of Excellence, Ochsner Health System, New Orleans, LA
Background
The benefits of the treatment of gestational diabetes (GDM) at ∼24–28 weeks are well recognized. However, it remains unclear if early screening for and treatment of GDM can decrease the risk of pregnancy complications. This trial aimed to determine if early screening compared to routine screening would improve pregnancy outcomes in women with obesity.
Methods
This pragmatic trial randomized participants with a BMI ≥30 kg/m2 to either early screening (14–20 weeks) or routine screening (24–28 weeks). All women not identified as having diabetes in the early screening cohort were rescreened again at 24–28 weeks. Screening was done using a 50-g glucose challenge followed by a 3-hour 100-g OGTT. The primary outcome was a composite that included adverse pregnancy outcomes associated with GDM. Outcomes were assessed using an intention-to-treat analysis.
Results
A total of 922 participants were included in the intention-to-treat analysis (n=459 in the early screening group and n=463 in the routine screening group). In the early screening group, 387 (84.3%) participants received the assigned intervention and 443 (95.9%) participants completed testing in the routine screening group. The overall incidence of GDM in the study was 13.6%. There was no difference in the composite primary outcome between the early and routine screening groups (261 [56.9%] and 235 [50.8%], respectively; P=0.06). There were no significant differences in the individual outcomes comprising the composite outcomes between the groups. Participants randomized to early screening were more likely to be started on insulin (2.4 vs 0.7%; P=0.03) and delivered earlier (36.7 vs 38.7 weeks; P<0.01).
Conclusions
Early screening for GDM did not improve the pregnancy outcomes associated with GDM in individuals with obesity.
Comment
This pragmatic trial demonstrated no improvement in outcomes of participants who received early screening for GDM. This trial benefits from its “real-world” approach to early screening in pregnant individuals with obesity. It is limited by its inclusion of only two centers in the United States, which may limit its applicability to other populations, as well as its relatively small number of participants with a diagnosis of GDM (n=56). Authors came close to achieving their estimated sample size (58 participants with a diagnosis of GDM), which required an 8% difference in their primary outcome. In addition, the primary composite outcome slightly favored the routine screening group (P=0.06), which may argue against the study being underpowered. The increased insulin use and earlier delivery in the group randomized to early screening may suggest increased intervention in this nonblinded pragmatic trial. Further trials examining early screening for GDM are ongoing and may include different populations allowing for increased generalizability.
Footnotes
Author Disclosure Statement
No competing financial interests exist for JMY. HRM sits on a scientific advisory board for Medtronic and reports speaker's fees from Abbott Diabetes Care, Dexcom, Medtronic, and Novo Nordisk.