One- Compared to Two-Step Gestational Diabetes Screening and Pregnancy Outcomes: A Systematic Review and Meta-analysis

Abstract

Objective:

To estimate short-term maternal and neonatal outcomes with one- compared to two-step testing for gestational diabetes (GDM).

Data Sources:

A systematic review of randomized controlled trials (RCTs) and observational studies comparing one- and two-step GDM testing strategies prior to September, 2021 was conducted. The primary outcome was large for gestational age (LGA) infants. Secondary outcomes were clinically relevant outcomes for GDM that were selected a priori.

Methods of Study Selection:

Titles, abstracts, and manuscripts were screened, selected, and reviewed by the first two authors. Four RCTs (24,966 patients) and thirteen observational studies (710,677 patients) were analyzed.

Tabulation, Integration, and Results:

Pooled relative risks (RRs) were calculated with a 95% confidence intervals (CI) using random-effects models and plotted graphically with forest plots. Study heterogeneity was evaluated using Cochran’s Q and Higgins I² tests. Quality of studies meeting inclusion criteria was evaluated with the Downs and Black checklist. Publication bias was assessed using asymmetry of funnel plots and Harbord’s test. There was no difference in risk of LGA infants (pooled RR 0.95; 95% CI 0.88–1.04) by testing strategy among RCTs, but patients who underwent one-step testing were more likely to be diagnosed with GDM (pooled RR 2.13; 95% CI 1.61–2.82) and treated with diabetes medications (pooled RR 2.24; 95% CI 1.21–4.15). One-step testing was associated with higher rates of NICU admission (pooled RR 1.12; 95% CI 1.00–1.26) and neonatal hypoglycemia (pooled RR 1.23; 95% CI 1.13–1.34). In analysis of high-quality RCTs and observational studies, one-step testing was associated with a lower risk of LGA infants (pooled RR 0.97; 95% CI 0.95–0.98), but higher rates of GDM diagnosis, treatment, NICU admission, and neonatal hypoglycemia.

Conclusion:

Despite a significant increase in GDM diagnosis and treatment with one-step testing, there is no difference in LGA infants compared to two-step testing among RCTs.

Précis:

There is no difference in large for gestational age infants with a one- compared to two-step testing strategy for gestational diabetes.

Introduction

Significant time, energy, and resources have been invested in determining optimal diagnostic criteria for gestational diabetes mellitus (GDM). Yet, we are no closer to resolving the one-step versus two-step testing dilemma in 2022 than we were in 2008 when the HAPO study was published.¹ A 2017 Cochrane review concluded there was insufficient evidence to recommend one strategy over the other based on available data.² Several studies, including two large randomized controlled trials, published in the interim, advanced our understanding of one- versus two-step testing.^3,4 The objective of this systematic review and meta-analysis was to use the collective power of pooled data to assess the implications of GDM testing strategy on pregnancy outcomes.

Sources

This study followed the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) and the Meta-analysis of Observational Studies (MOOSE) reporting guidelines.^5,6 PROSPERO International Prospective Register of Systematic Reviews registration was completed prior to initiating the review and data extraction. Using the PICOT method to define the study question and eligibility criteria,⁷ studies of pregnant patients, comparing IADPSG one-step to Carpenter-Coustan two-step testing in the late mid-trimester (targeting 24–28 weeks) with clinically relevant outcomes were included. A medical librarian (M.M.D.) created a search strategy for the diagnosis of gestational diabetes using one-step or two-step testing with a combination of standardized terms and keywords, including: gestational diabetes, GDM, pregnancy induced diabetes, one-step, two-step, GDM diagnosis, GDM screening, and oral glucose tolerance test. The search was run on August 29, 2021 without any filters in the databases Ovid Medline 1946-, Embase.com 1947-, Scopus 1960-, Cochrane Central, and Clinicaltrials.gov. Full electronic search strategies are provided in the supplementary material. As all data were de-identified and available in the public domain, Institutional Review Board approval was not necessary.

Study Selection

Randomized controlled trials (RCT) and observational studies published in English before September 2021 comparing the IADPSG one-step and the Carpenter-Coustan two-step methods of GDM screening and diagnosis were included. Both study types were included to understand and balance the findings of “gold standard” RCTs with real-world clinical application and generalizability informed by observational studies. The inclusion of observational studies provided greater statistical power to address the primary outcome at the population level, since a much smaller subgroup of the sample (~18% of patients with one-step and ~6% of patients with two-step testing) would be diagnosed with GDM and drive GDM-related clinical outcomes. All other study designs, including cost-effectiveness analyses, systematic reviews, and meta-analyses, were excluded.

The primary outcome was large for gestational age (LGA) infants (birthweight ≥90^th percentile). Secondary outcomes were selected based on clinical relevance to GDM, a priori, and inclusion in at least two studies meeting criteria so that results could be pooled. Maternal secondary outcomes included the rate of GDM diagnosis by one-step testing with IADPSG or two-step testing with Carpenter-Coustan criteria,⁸ hypertensive disorders of pregnancy (gestational hypertension or preeclampsia),⁹ primary cesarean section, and GDM treatment with oral diabetes medications or insulin. Fetal and neonatal secondary outcomes were macrosomia (birthweight ≥4,000g, as defined by the two largest RCTs included), NICU admission during the first 28 days of life, small for gestational age (SGA) infants (birthweight ≤10^th percentile), preterm birth (delivery <37 weeks gestation), respiratory distress syndrome (RDS) with progressive respiratory failure in conjunction with characteristic chest radiograph findings shortly after birth,¹⁰ hyperbilirubinemia (requiring phototherapy), neonatal hypoglycemia (defined as blood glucose <40 mg/dL in the first 24 hours of life), stillbirth (fetal death after 20 weeks with verified non-live birth), neonatal death (within first 28 days of life), and shoulder dystocia (delivery of the shoulder requiring maneuvers). Outcomes reported in each study are shown in Appendix 1.

Titles and abstracts were screened by two authors (M.B. and D.M.H.). Full-text articles were retrieved based on relevance of the study question and outcomes included. Each article was reviewed separately by two authors (M.B. and D.M.H.) against inclusion and exclusion criteria. Bibliographies of studies meeting criteria were reviewed for additional potentially relevant studies. Two authors (M.B. and D.M.H.) independently abstracted data into standard extraction forms. Discrepancies were resolved by the senior author (E.B.C.).

Study quality was assessed with the Downs and Black checklist, a validated tool that scores studies based on a 27-items evaluating reporting, external validity, bias, internal validity, confounding, and power.¹¹ We made the a priori decision that studies with a score in the top quartile would be deemed high-quality. Two authors (M.B. and D.M.H.) independently completed the checklist. Discrepancies were resolved by the senior author.

Study heterogeneity was evaluated using Cochran’s Q and Higgins I² tests.¹² Heterogeneity was considered significant if P<0.1 or I²>30%. The DerSimonian-Laird random-effects model was used to pool data from individual studies, even if there was no evidence of statistical heterogeneity. This approach resulted in more conservative estimates of effect sizes. Pooled relative risks (RRs) were calculated with a 95% CI and plotted graphically with forest plots. We conducted three versions of meta-analyses including 1) all studies, 2) stratified by design (RCT or observational), and 3) high-quality studies only. We visually inspected funnel plots to assess publication bias for the primary outcome. We tested asymmetry with Harbord’s test for categorical variables. Statistical analyses were performed using Stata, Version 17 (StataCorp LLC) with METAN software.

Results

The electronic literature search yielded 2,456 citations with 979 duplicates that were identified and removed, resulting in 1,477 unique citations (Figure 1). Each title was reviewed for relevance, 414 abstracts were reviewed, and 89 abstracts or full-text manuscripts were screened against inclusion and exclusion criteria. Review of bibliographies of selected papers against study criteria did not result in additional studies for consideration. Studies were excluded for the following reasons: trial design paper (n=3), did not include clinical outcomes of interest (n=19), did not compare one- to two-step testing (n=7), only included patients with GDM diagnosis rather than population level screening (n=10), only included patients negative for GDM (n=1), retrospectively reclassified patients originally diagnosed with Carpenter-Coustan to IADPSG for analysis (n=4), stringent exclusion criteria limiting generalizability (Body Mass Index (BMI) exclusions, excluded singletons, preterm births <32 weeks, patients with abnormal fasting values at screening visit) (n=5), used a screening strategy other than IADPSG or Carpenter-Coustan (n=10), conference abstract with duplicate paper (n=10), one-step cohort had more stringent blood glucose targets for medication initiation than two-step (n=1), and inability to obtain data after multiple attempts to contact authors (n=2). One study (Lucovnik et al, 2020) was removed from the macrosomia analysis due to a higher macrosomia cutoff (>4500g) than other included studies.¹³ A sensitivity analysis with inclusion of this study did not change outcomes (data not shown).

Figure 1.

Flowchart showing the methodology of study selection

Seventeen studies including 735,643 patients were analyzed: four RCTs (24,966 patients).^3,4,14,15 and 13 observational studies (710,677 patients)^13,16–27 Study characteristics, inclusion, and exclusion criteria are shown in Table 1. Quality scores on the Downs and Black checklist ranged from 15–28 (median 21, IQR 21–22). Three RCTs and four observational studies were deemed high-quality based on a Downs and Black score in the top quartile (Appendix 2).

Table 1.

Characteristics of studies included in the final meta-analysis

Study	Year	Study Years	Country	Setting	Inclusion	Exclusion
Randomized Controlled Trials
Davis et al1	2021	2015–2019	US	10 clinics in Pittsburgh, PA	Pregnant, no pre-existing diabetes	Miscarriage, multiple gestation, <18yrs of age, history of bariatric surgery, change in health insurance
Hillier et al2	2021	2014–2017	US	Hospital system in Portland, OR and Hawaii	18–45yrs old, 18wks, 0 days to 28wks, 6 days GA at enrollment	Pre-existing Type I or Type II diabetes, diabetes diagnosed <24wks GA, inability to complete screening prior to 30wks GA, multifetal gestation, major congenital anomaly, anticipated preterm birth < 28wks GA, corticosteroids 30 days prior to enrollment, hypertension requiring medication, HIV, liver disease, history of gastric bypass, blood sugar >200 on pre-randomization one-hour screening
Khalifeh et al3	2020	2016	US	Large academic hospital in Philadelphia, PA	24–28wks GA	Pre-gestational diabetes, history of bariatric surgery
Scifres et al4	2015	2012–2013	US	Large academic hospital in Pittsburgh, PA	18–45yrs old, both lab visits between 24–30wks GA	Pre-gestational diabetes, positive first trimester GDM screen, multiple gestation, corticosteroids 30 days prior to enrollment, history of bariatric surgery, use of fertility treatment to conceive, plan to deliver at different hospital, inability to complete glucose testing before 30wks gestation, anticipated preterm delivery for maternal or fetal indications, blood sugar >200
Observational
Aubry et al5	2021	Two-step: 2005–2010 One-step: 2012–2017	Switzerland	100 Swiss obstetrics hospitals	24–28wks GA	Pre-gestational diabetes
Costa et al6	2019	Two-step: 2009–2011 One-step: 2013–2014	Belgium	Large academic center in Brussels, Belgium	24–28wks GA	Pre-gestational diabetes, multiple gestation, no screening data available, patient did not tolerate screening
Duran et al7	2014	Two-step: April 2011-March 2012 One-step: April 2012-March 2013	Spain	Large community hospital in Madrid, Spain	24–28wks GA	Pre-gestational diabetes, diabetes diagnosed before 24wks GA
Feldman et al8	2016	Two-step: 2010–2011 One-step: 2011–2013	US	Large community hospital in Montebello, CA	Managed starting in first trimester	Pre-gestational diabetes, multiple gestation
Fuller et al9	2014	Two-step: 2010–2011 One-step: 2011–2012	US	Tertiary care hospital in Hartford, CT	24–28wks gestation	Pre-gestational diabetes, <18yrs of age, history of gastric bypass
Ghaffari et al10	2020	Two-step: 2012–2015 One-step: 2016–2018	US	Academic hospital in San Francisco, CA	24–28wks GA	Pre-gestational diabetes, multiple gestation
Huhn et al11	2017	Two-step: 2008–2010 One-step: 2010–2013	Switzerland	Academic hospital in Basel, Switzerland	Delivered >22wks GA	Pre-gestational diabetes, multiple gestation, screening after 28wks GA, delivery elsewhere
Hung et al12	2015	Two-step: 2009–2010 One-step: 2012–2013	Taiwan	Large community hospital in Taoyuan City, Taiwan	24–28wks GA, delivered after 24wks GA	Pre-gestational diabetes, multiple gestation, fetal anomalies
Kong et al13	2015	Two-step: 2009 One-step: 2011	Canada	Database with all births in British Columbia, Canada	All subjects delivering from April-September 2009 (CC), and April-September 2011 (IADPSG)	Pre-gestational diabetes
Lee et al14	2019	2011–2013*	US	Academic hospital in Kansas City, MO	24–28wks GA	Pre-existing diabetes, multiple gestation, overt diabetes mellitus (fasting ≥126, HbA1c≥6.5%, random ≥200), Underwent first trimester GDM screening, congenital anomalies
Lucovnik et al15	2020	Two-step: 2004–2010 One-step: 2011–2017	Slovenia	Academic hospital in Ljubjiana, Slovenia	Delivered >22wks GA, birthweight >500g	Pre-gestational diabetes
Palatnik et al16	2017	One-step: 2010–2013 Two-step: 2013–2015	US	Academic hospital in Chicago, IL	>18yrs of age, delivered ≥37wks GA	Pre-gestational diabetes, multiple gestation, HbA1c ≥6.5% at first prenatal visit
Wu et al17	2016	Two-step: 2010 One-step: 2011	Taiwan	Academic hospital in Taipei, Taiwan	All singleton pregnancies delivering in 2011	None reported

^*Allowed clinicians to self-select screening method

In analysis limited to the 24,966 patients enrolled in RCTs, the risk of LGA infants was similar between one- and two-step testing (4 studies: pooled rates 8.8% one-step vs. 9.2% two-step, pooled RR 0.95; 95% CI 0.88–1.04) (Table 2 and Figure 2). There was no evidence of significant study heterogeneity (I²=0.0, p=.716; Figure 2), but there was evidence of publication bias (Harbord’s Test p=.084) (Figure 5). Patients undergoing one-step testing were significantly more likely to be diagnosed with GDM (4 studies: pooled rates 16.3% one-step vs. 8.3% two-step, pooled RR 2.13; 95% CI 1.61–2.82, number needed to screen (NNS) 13 patients) (Appendix 4) and receive diabetes medication management (3 studies: pooled rates 7.1% one-step vs 3.8% two-step, pooled RR 2.24; 95% CI 1.21–4.15, NNS 31 patients) (Table 2 and Appendix 5). Infants of patients who underwent one-step testing had higher rates of NICU admission (3 studies; pooled rates 5.1% one-step vs. 4.5% two-step, pooled RR 1.12; 95% CI 1.00–1.26, NNS 167 patients) (Table 2 and Appendix 6) and neonatal hypoglycemia (4 studies; 9.3% one-step vs. 7.6% two-step, pooled RR 1.23; 95% CI 1.13–1.34, NNS 59 patients) (Table 2 and Appendix 7). There were no differences in other maternal or neonatal secondary outcomes among RCTs (Table 2).

Table 2.

Comparison of maternal and neonatal outcomes in patients undergoing one-step versus two-step testing among RCTs

Outcomes	One-step Testing n=12,520 (50.15)	Two-step Testing n= 12,446 (49.85)	RR (95% CI)	I2 (%)	p
Primary Outcome
Large for gestational age infant (>90th percentile)	1,015 (8.8)	1,058 (9.2)	0.95 (0.88–1.04)	0.0	.716
Secondary Outcomes
Maternal Outcomes
Gestational diabetes (GDM) diagnosis	1,910 (16.3)	970 (8.3)	2.13 (1.61–2.82)	26.6	.252
GDM Treatment (insulin or oral medication)	833 (7.1)	446 (3.8)	2.24 (1.21–4.15)	67.9	.044
Hypertensive disorders of pregnancy	1,548 (13.6)	1,536 (13.6)	1.00 (0.94–1.07)	0.0	.391
Primary Cesarean Section	2,826 (24.0)	2,887 (24.7)	0.98 (0.93–1.02)	0.0	-*
Fetal and Neonatal Outcomes
Macrosomia (>4,000g)	1,217 (11.2)	1,225 (11.3)	0.99 (0.92–1.07)	0.0	.695
Small for gestational age infant (<10th percentile)	993 (8.6)	948 (8.3)	1.04 (0.95–1.13)	0.0	.815
Shoulder dystocia	240 (2.1)	224 (2.0)	1.06 (0.89–1.28)	0.0	.655
NICU admission	588 (5.1)	520 (4.5)	1.12 (1.00–1.26)	0.0	.466
Preterm birth (<37wks GA)	728 (6.4)	721 (6.4)	1.00 (0.91–1.11)	0.0	.619
Respiratory distress syndrome	225 (2.0)	227 (2.0)	0.99 (0.82–1.16)	0.0	-*
Neonatal hypoglycemia	1,098 (9.3)	889 (7.6)	1.23 (1.13–1.34)	0.0	.371
Hyperbilirubinemia	506 (4.3)	494 (4.2)	1.18 (0.75–1.85)	45.5	.160
Stillbirth	61 (0.5)	67 (0.6)	0.90 (0.64–1.28)	0.0	.664
Neonatal death	7 (0.1)	12 (0.1)	0.58 (0.23–1.47)	0.0	-*

Bold denotes statistical significance (p<0.05).

^*Only one study included

Figure 2.

Forest plot comparing one-step and two-step testing on LGA diagnosis, RCT

Figure 5.

Funnel plot comparing one-step and two-step testing on LGA diagnosis, RCTs and High-quality Observational studies

X= Khalifeh et al (2020) RCT was not deemed high quality but was added to illustrate all RCTs included in the primary analysis

* Of note, only three out of four Observational studies deemed high-quality reported rates of LGA. Two of the included studies are overlapping at the apex of the graph.

When only high-quality RCTs and observational studies were considered (363,950 patients), one-step testing was associated with a lower risk of LGA infants (pooled RR 0.97; 95% CI 0.95–0.98; Figure 3). There was no evidence of significant study heterogeneity (I²=0.0, p=.733; Figure 3) or publication bias (Harbord’s Test p=.330; Figure 5). Results were otherwise similar to the RCT analysis with higher rates of GDM diagnosis (pooled RR 2.13; 95% CI 1.32–3.42), treatment (pooled RR 1.87; 95% CI 1.40–2.50), NICU admission (pooled RR 1.13; 95% CI 1.05–1.21) and neonatal hypoglycemia (pooled RR 1.25; 95% CI 1.19–1.31) (Table 3). Results stratified by high- and low-quality studies are shown in Appendix 3.

Figure 3.

Forest plot comparing one-step and two-step testing on LGA diagnosis, High-quality studies

Table 3.

Summary of maternal and neonatal outcomes in patients undergoing one-step versus two-step testing by study type and high-quality

Outcomes	RCT Pooled RR (95% CI) n=24,966	High-quality Pooled RR (95% CI) n=363,950	Observational Only Pooled RR (95% CI) n=710,677
Primary Outcome
Large for gestational age infant (>90th percentile)	0.95 (0.88–1.04)	0.97 (0.95–0.98)	0.93 (0.90–0.96)
Secondary Outcomes
Maternal Outcomes
Gestational diabetes (GDM) diagnosis	2.13 (1.61–2.82)	2.13 (1.32–3.42)	2.54 (1.92–3.37)
GDM Treatment (insulin or oral medication)	2.24 (1.21–4.15)	1.87 (1.40–2.50)	1.15 (0.70–1.88)
Hypertensive disorders of pregnancy	1.00 (0.94–1.07)	0.92 (0.82–1.04)	0.89 (0.82–0.95)
Primary Cesarean Section	0.98 (0.93–1.02)	1.13 (0.95–1.34)	1.13 (1.04–1.23)
Fetal and Neonatal Outcomes
Macrosomia (>4,000g)	0.99 (0.92–1.07)	0.86 (0.73–1.03)	0.81 (0.75–0.88)
Small for gestational age infant (<10th percentile)	1.04 (0.95–1.13)	1.00 (0.97–1.02)	1.00 (0.99–1.02)
Shoulder dystocia	1.06 (0.89–1.28)	0.96 (0.84–1.09)	1.04 (0.91–1.20)
NICU admission	1.12 (1.00–1.26)	1.13 (1.05–1.21)	1.06 (0.96–1.18)
Preterm birth (<37wks GA)	1.00 (0.91–1.11)	1.00 (0.95–1.05)	1.00 (0.93–1.08)
Neonatal respiratory distress syndrome	0.99 (0.82–1.18)	0.99 (0.82–1.18)	1.13 (1.09–1.16)
Neonatal hypoglycemia	1.23 (1.13–1.34)	1.25 (1.19–1.31)	1.19 (1.02–1.39)
Hyperbilirubinemia	1.18 (0.75–1.85)	1.06 (0.96–1.16)	1.00 (0.78–1.26)
Stillbirth	0.90 (0.64–1.28)	1.02 (0.92–1.12)	1.02 (0.92–1.12)
Neonatal death	0.58 (0.23–1.47)	0.58 (0.23–1.47)	1.08 (0.42–2.79)

Bold denotes statistical significance (p<0.05).

Among 710,677 patients participating in observational studies, those with one-step testing were less likely to deliver LGA infants (pooled RR 0.93; 95% CI 0.90–0.96) (Table 3, Figure 4). A sensitivity analysis comparing pooled adjusted and unadjusted results from observational studies assessing LGA showed similar results (data not shown). Patients receiving one-step testing for GDM were significantly more likely to be diagnosed with GDM, but not treated for it (Table 3). With regard to secondary outcomes, one-step testing was associated with a higher risk of primary cesarean section, neonatal respiratory distress syndrome, and neonatal hypoglycemia, but a lower risk of hypertensive disorders of pregnancy compared to two-step testing (Table 3). There were no differences in other maternal or neonatal outcomes.

Figure 4.

Forest plot comparing one-step and two-step testing on LGA diagnosis, Observational studies

Discussion

This meta-analysis of GDM screening tests showed that there was no difference in risk of LGA infants by one- versus two-step testing strategy in analysis limited to RCTs. However, patients undergoing one-step testing were more than twice as likely to be diagnosed with and treated for GDM, while experiencing higher rates of NICU admission and neonatal hypoglycemia. Analysis limited to high-quality RCTs and observational studies showed similar results, though those undergoing one-step testing had a modest decrease in LGA risk. Study quality varied greatly among observational studies, but pooled results showed a significant decrease in LGA and hypertensive disorders of pregnancy with one-step testing. These findings were offset by a significant increase in primary cesarean section, neonatal respiratory distress syndrome, and neonatal hypoglycemia.

Our findings add critical insight to the literature because the two most recent meta-analyses on this topic, published in 2018 and 2019,^28,29 were limited by lack of high-quality RCTs. The 2017 Cochrane review analyzed several aspects of GDM diagnosis, including one- vs. two-step screening,² but only one included study assessed results by screening strategy.³⁰ A 2019 systematic review and meta-analysis²⁸ included three randomized trials, one of which is included here.³¹ The two trials that we excluded were a cost-analysis and a study that did not include patients with GDM in their analysis of outcomes.

This meta-analysis builds upon prior work in two ways. First, the addition of three well-executed trials, published in the interim, provide further insight to determine the impact of testing strategy on short-term maternal and neonatal outcomes.^3,14,32 Second, pooling high-quality studies provided adequate power to detect differences in diabetes-related pregnancy outcomes that are difficult to detect with a population-based screening strategy when only 6–18% of the population receives a GDM diagnosis. A recent commentary by Coustan et al argued that the largest RCT published to date, with more than 20,000 patients, was not adequately powered to detect differences in pregnancy outcomes for this reason.³³ We included a secondary analysis of high-quality RCTs and observational studies of GDM testing strategy, pooling nearly 400,000 patients, to address this issue of power and minimize the risk of a type 2 error. While there were many more differences in maternal and neonatal outcomes among observational studies overall, these findings should be taken with caution in the context of significant study heterogeneity and varying levels of quality among these studies. Our analysis of high-quality studies is well-powered to conclude that the most likely differences in pregnancy outcomes by testing strategy are higher rates of NICU admission and neonatal hypoglycemia with one-step testing and, potentially, a modest decrease in LGA that is largely driven by observational studies.

Expert editorials on the recently published RCTs concede that both favor two-step testing in the absence of demonstrable neonatal or parental benefit with significantly higher rates of diagnosis and treatment with one-step testing.^33,34 Our large, pooled, internationally representative sample provides further credence to this argument. We hypothesize that adverse outcomes associated with one-step testing correspond to relatively lower-risk patients being given a diagnosis of GDM, such that the “treatment” may no longer justify the “cure.” A GDM diagnosis likely changes the way that health care professionals engage with birthing people and babies through seemingly risk-reducing interventions that may paradoxically worsen outcomes. For example, pediatricians only screen for neonatal hypoglycemia when GDM or other risk factors are present, leading to greater detection and treatment with one-step testing, as seen in all of our analyses. This may also explain the higher rates of NICU admission seen in RCTs and high-quality study analyses.

There are many strengths to this study. The recent publication of two well-executed RCTs allowed us to pool high-quality data to help obstetricians decide which universal screening method to use. The large pooled sample size provided greater power to detect smaller differences than would have been possible by limiting the sample to the nearly 25,000 patients in RCTs. This approach also allowed us to leverage the strengths of each study design—the gold standard RCT designs which may have limited generalizability compared to more pragmatic observational studies that more closely resemble real-world settings. The Downs and Black checklist assessed the quality of both RCTs and observational studies based on our pre-determined high-quality threshold.

Despite its strengths, this study should be considered in the setting of the following limitations. We limited our two-step eligibility criteria to studies utilizing Carpenter-Coustan criteria, both because it is most commonly used clinically and to limit heterogeneity among studies. While preterm birth rates <37 weeks were similar between groups, pooled data were not available to compare rates of early preterm (<34 or <28 weeks) or early term (37–38 6/7 weeks) birth, which could have helped contextualize differences in hypertensive disorders of pregnancy and RDS for observational studies. There were significant differences between observational studies and RCTs beyond their study design. All four RCTs were conducted in the United States at large tertiary care centers with a high-risk population compared to the general population. In contrast, more than half of the observational studies were conducted internationally at both academic and community hospitals. Given the major differences in clinical characteristics of each population, the prevalence of some secondary outcomes were significantly different between observational studies and RCTs. For example, the prevalence of neonatal hypoglycemia was over seven times higher in the RCT population and hypertensive disorders of pregnancy was over three times more likely. There was evidence of publication bias for the primary outcome with RCTs, likely due to the influence of the large Hillier trial; however, publication bias was not observed in the high-quality analysis.

Finally, whether the two-fold increase in GDM diagnosis with one-step testing results in improvement in long-term cardiometabolic outcomes remains a critical question that is not answered in the primary literature nor this meta-analysis. Pregnancy care should be framed as an important component of life course care. Thus, while this meta-analysis suggests that the diagnostic burden of one-step testing is not justified by differences in pregnancy outcomes, there may still be significant long-term benefits to the parent and offspring if this information is used to guide aggressive risk-reduction strategies. Long-term studies are necessary to answer this question of great public health import.

Our results show that patients undergoing one- and two-step testing had equal rates of LGA infants, despite a greater likelihood of GDM diagnosis and treatment with one-step testing. Our findings favor two-step testing to minimize the increased burden of GDM diagnosis resulting from one-step testing. However, understanding the long-term implications of such a strategy across the life course is critically important to inform the public health path forward.