With obesity prevalence high (29% of reproductive age women in the US (24)) and rising, incidence of GDM is also increasing, putting infants and their mothers at risk for adverse health outcomes during and after birth (7). Elevated glucose and insulin during pregnancy that characterize GDM can be managed, lowering the risk of large for gestational age infants and infants with high levels of adiposity (25). It is paramount to identify women with GDM and offer treatment. The novel simulation model described in this study is an initial step towards personalized tools for monitoring maternal glucose dynamics over the course of pregnancy that uses biomarkers that are routinely collected during prenatal care. The model output closely resembles plasma insulin and glucose in early and late pregnancy among twenty-eight research participants that underwent OGTTs, and it also resembles steady-state levels under hyperinsulinemic-euglycemic clamp conditions. To our knowledge, this is the first system dynamics model exploring changes in glucose and insulin dynamics known to develop over the course of pregnancy.
Scientists have studied glucose and insulin dynamics for decades in a quest to understand and treat diabetes, which affects about 422 million people worldwide and is estimated to contribute to one in nine deaths among adults ages 20–79 years old (26, 27). Many mathematical models have been developed to move the field closer to a “closed loop” system for managing insulin administration in order to reduce the burden of care on individuals with diabetes (28). Diabetes during pregnancy is a particularly important topic to study for two reasons: 1) if untreated it can increase risks for negative health outcomes for the pregnant patient and offspring, and 2) insulin resistance, a state of decreased insulin sensitivity that is on a continuum whose upper end is called diabetes, naturally develops over the course of pregnancy (29). Studying the development of physiological insulin resistance that occurs during pregnancy and rapidly resolves after delivery gives scientists a unique window into metabolic processes that would not be possible to experimentally manipulate in humans. The use of biomarkers and other data that are routinely collected during prenatal care in conjunction with mathematical modeling is a promising strategy for studying glucose and insulin dynamics in humans in pursuit of developing highly effective and acceptable treatments.
Our system dynamics model contained several key features that help explain its ability to replicate clinical data. There are important delays and functional forms that influence the dynamic relationship between glucose and insulin (30). There is a delay in the dampening effect on plasma insulin levels and hepatic glucose synthesis suppression (22). There is also a delay as insulin moves from plasma to interstitial fluid where it binds with insulin receptors on the surface of peripheral cells and initiates the translocation of GLUT4 to the cell surface (23). Experimental evidence shows that insulin release rate follows a two-phase pattern (31); that non-linear rate is critical for accurately simulating insulin accumulation and action. Two reviews of mathematical models of glucose and insulin dynamics noted similar features in hyperinsulinemic-euglycemic clamp models (28, 32). Challenges of modeling oral glucose intake compared to intravenous clinical tests include factors such as the rate of gastric emptying, extent of glucose absorption, and the effect of hormones such as incretins (28). We addressed these challenges by modeling exogenous glucose intake as a rate that is dependent on the stock of ingested glucose. This created an exponential decline in the rate of glucose uptake over the course of an OGTT, which represented a simplified yet sufficient function for our model.
Our study has strengths and limitations. We used a small sample to calibrate the model, potentially limiting generalizability. Glucose and insulin production, signaling, metabolism, and other biochemical processes are characterized in exceptional detail that was not included in the model presented in this study (33). An important strength is that we used a dynamic modeling approach to study a dynamic biological process. Another strength is our ability to reproduce two glucose tolerance protocols, an OGTT and a hyperinsulinemic-euglycemic clamp, by modifying relevant model parameters (e.g., rate of glucose intake, rate of insulin secretion) within the same model structure. To our knowledge, no other published model can reproduce clinical data under multiple testing protocols (28).
There are a number of opportunities to expand the model presented here to answer important research questions. For example, future models will include fetal glucose and insulin dynamics to test whether those additions might allow us to interrogate the fetal glucose steal hypothesis, a theoretical explanation for why some pregnant women are not diagnosed with GDM yet deliver babies with macrosomia (14). Study results could also inform considerations for different approaches to identify impaired glucose tolerance during pregnancy that might include: 1) greater attention to plasma insulin in addition to the current emphasis on maternal glucose levels, 2) utilizing continuous glucose monitoring to gather information about maternal glucose levels outside of the conditions tested with a routine OGTT, and 3) additional screening at multiple time points during pregnancy for at-risk individuals. Expanding this model to include details about the placenta as the interface between fetal nutrient glucose demands and maternal supply is another potential direction, as is adding lipid handling dynamics since insulin has major effects on lipid biosynthesis and metabolism. The placenta secretes hormones thought to promote insulin resistance as pregnancy progresses (34), so there is an opportunity to explore how placental growth and hormone production influence the development of insulin resistance with advancing gestation. Emerging evidence suggests that obesity and its effects on maternal metabolism may also increase risk of preterm delivery due to placental growth dysregulation (35). We could use a model to elucidate mechanisms underlying this observation and test potential intervention effects aiming to promote healthy fetal growth and prevent extremes on either end of the growth spectrum (i.e., small- or large-for-gestational-age infants). Validating the model with a larger dataset would be beneficial, especially if it afforded the opportunity to study differences in model parameter values by race and ethnicity since evidence suggests potential differences in glucose metabolism between racial groups (22). The model could be further developed for individual-level tailoring and clinical decision-making. A study of a federally qualified health system with significantly better maternal and neonatal outcomes than surrounding clinics (e.g., lower rate of preterm delivery, high rates of breastfeeding) found that participants described personalized care as a critical factor that shaped their positive prenatal care experiences (36). Finally, the model could be adapted to become a teaching tool for health professionals that care for patients experiencing diabetes or impaired glucose tolerance (37).