Probabilities

The baseline probabilities for preeclampsia, cesarean delivery, macrosomia, neonatal death, and NICU admission were derived from the literature ( Table 1 ). For the group of patients who would be diagnosed under the 2h OGTT using the new IADPSG guidelines, the baseline probabilities were derived directly from the data for HAPO group 5, representing the patients within the HAPO study whose glycemic values straddle the new IADPSG guidelines. The baseline probability of falling in group 1 was obtained from 2008 data on pregnancy hospital discharges in the United States, while the baseline probability for falling in group 2 was inferred from HAPO study data. As the presence of macrosomia affects the probability of shoulder dystocia, BPI, hypoglycemia, and hyperbilirubinemia, these probabilities were derived from a separate body of work that examined these factors independently in patients with GDM. Probabilities for maternal death were also derived from the literature.

To estimate the effect of GDM treatment on these outcomes for patients in group 1, the baseline probabilities were multiplied by relative risks (RRs) taken both from a recent metaanalysis on the effects of GDM treatment on pregnancy outcomes as well as the RR for treating mild GDM. For patients in group 2, baseline probabilities were multiplied by the RR associated with treatment published in the 2009 RCT on treating mild GDM.

Costs

All costs were in 2012 US dollars and were inflated using the medical component of the consumer price index ( Table 1 ). Of note, the cost of treating GDM was set at a baseline of $1971, which included pharmacotherapy, antenatal visits, ancillary diabetes-related visits, and antepartum fetal surveillance. The cost of GDM diagnosis was accounted for and included the direct cost of test materials and analysis as well as indirect costs of travel and opportunity costs for the patient. The costs accounted for neither antepartum admissions for GDM-related conditions nor broader indirect costs. Because maternal mortality is difficult to quantify with a point estimate as the values vary based on the circumstances of death, a baseline of $100,000 was used and was subject to sensitivity analysis. For BPI, adjusted published costs were $17,322 for permanent injury and $1939 for transient injury. All costs were subject to sensitivity analysis.

Utilities

All utilities were derived from the literature and include utilities from the maternal and neonatal perspective ( Table 1 ). The utility of a vaginal delivery was assumed to be one and the utility of a cesarean delivery was set to a baseline of 0.99 based on previously published research on women’s preferences in mode of delivery. Maternal death by definition was set to a utility of 0. The utility of a neonatal death from the maternal perspective was set at a baseline of 0.92, the published maternal utility of a miscarriage, and was applied over the maternal lifetime. The utility of a neonatal death from the neonatal perspective was by definition 0 and applied over the neonatal lifetime. The neonatal utility of a transient BPI was set to a baseline of 0.99 based on “brachial plexus injury that resolves within 2 months.” For a permanent BPI, a conservative value of 0.6 was used based on a published value for mild to moderate injury, compared with a utility of 0.45 for a severe injury. These utilities were applied over the neonatal lifetime. As there are no published utilities on short-term neonatal consequences such as hypoglycemia, NICU admission, or hyperbilirubinemia, these utilities were set to a conservative baseline value of 1. Utilities were calculated over the course of maternal life expectancy (56.1 years) and neonatal life expectancy (77.2 years) at a discount rate of 3%. All utilities were subject to sensitivity analysis.

Analysis

Baseline analysis compared routine GDM screening with the 1h GCT vs the 2h OGTT to estimate differences in maternal and neonatal outcomes, total costs, and QALYs for each strategy as well as the ICER.

Sensitivity analysis is a statistical tool that helps clarify key variables that most influence outcomes in the model and test the robustness of the model. A threshold value marks the point at which change in a variable would alter the conclusion. We tested the robustness of our model by performing univariate sensitivity analysis on all inputs and also introduced an additional variable in the sensitivity analysis to represent the percent efficacy of treatment, attempting to account for practice variations that can result in a range of outcomes with treatment.

Monte Carlo analysis is a tool that can incorporate uncertainty into the baseline stochastic model by sampling the underlying distributions for each input. One simulation represents a woman undergoing routine GDM screening with either the 1h GCT or the 2h OGTT. Her probabilities and costs are randomly chosen from a prespecified distribution determined from the literature, and this simulation is repeated with a different set of randomly chosen values, with the aggregate representing a theoretical cohort of random women. Based on these simulations, an acceptability curve can be constructed showing the probability of achieving cost effectiveness. To adopt the deterministic model to a stochastic one for the Monte Carlo simulation, costs were modeled using a gamma distribution, and probabilities were modeled using a beta distribution.