Changes to Gestational Diabetes Mellitus (GDM) Testing and Associations with the GDM Prevalence and Large- and Small-for-Gestational-Age Infants—An Observational Study in an Australian Jurisdiction, 2012–2019

Abstract

Background: Two changes to gestational diabetes mellitus (GDM) testing were implemented in the Australian Capital Territory in 2015 and 2017. Aims: We aimed to determine the associations between testing regimes and the prevalence of GDM and large-for-gestational-age (LGA) and small-for-gestational-age (SGA) infants and to compare the prevalence of LGA and SGA infants between women with and without GDM in each testing period. Methods: A total of 23,790 singleton live births with estimated GDM testing and birth dates between June 2012 and December 2019 were stratified into groups: pre-testing changes (June 2012–December 2014, group 1, n = 8069), revised diagnostic criteria (January 2015–May 2017, group 2, n = 8035) and changed pathology centrifugation protocol (June 2017-December 2019, group 3, n = 7686). Women were allocated to groups based on their estimated GDM testing date and stratified by their GDM status. A chi-square test, pairwise z-tests and logistic regression tested the associations. Results: The GDM prevalence significantly increased from 9.5% (group 1) to 19.4% (group 2) to 26.3% (group 3) (all: p < 0.001). The LGA infant prevalence significantly decreased in non-GDM women following revised diagnostic criteria implementation (11.6% vs. 9.7%, p = 0.001). Compared to group 1, women with GDM in groups 2 and 3 had significantly reduced odds of having LGA infants (aOR = 0.73, 95% CI of 0.56–0.95 and p = 0.021 and aOR = 0.75, 95% CI of 0.59–0.97 and p = 0.029, respectively). Compared to group 1, non-GDM women in groups 2 and 3 had significantly reduced odds of having LGA infants (aOR = 0.83, 95% CI of 0.74–0.92 and p < 0.001 and aOR = 0.88, 95% CI of 0.79–0.99 and p = 0.026, respectively). There were no significant associations for group 3 compared to group 2 nor for SGA infants. Conclusions: While significantly increasing the GDM prevalence, implementing the testing changes was associated with a reduced whole-population LGA infant prevalence without a change in the SGA infant prevalence.

1. Introduction

Globally, gestational diabetes mellitus (GDM) is one of the most common pregnancy-related complications, with an increasing prevalence worldwide, including in Australia [1,2]. GDM is defined as a glucose intolerance that begins, or is first diagnosed, during pregnancy, below the diagnostic threshold for diabetes diagnosed outside of pregnancy [1]. As it exposes the fetus to an excessive nutrient availability, undiagnosed and/or untreated GDM increases the risk of adverse neonatal outcomes, including large-for-gestational-age (LGA) infants [3,4,5].

In 2013 the World Health Organization (WHO) revised its guidelines on the GDM diagnostic criteria (DC) based on the recommendations of the International Association of the Diabetes and Pregnancy Study Group (IADPSG) [1]. The recommendations followed an assessment of evidence from the 2008 Hyperglycaemia and Adverse Pregnancy Outcomes study (HAPO study), which found that adverse neonatal outcomes may occur at lower glycemic levels than had previously been considered [1,6]. The WHO’s revised guidelines have not been universally adopted, and variations continue to exist within and between countries [7].

In 2014, the Australian Diabetes in Pregnancy Society (ADIPS) endorsed the revised criteria, superseding their 1998 version [8]. Consequently, the fasting plasma glucose (FPG) threshold was lowered to ≥5.1 mmol/L (from 5.5 mmol/L), a new 1 h blood glucose threshold was added of ≥10.0 mmol/L and the 2 h threshold was increased to ≥8.5 mmol/L (from 8.0 mmol/L) [2]. Universal one-step testing using a 2 h fasting 75 g oral glucose tolerance test (OGTT) replaced the two-step process of initial screening using either a non-fasting 50 g or 75 g glucose challenge test (GCT) with follow-up testing using a 2 h fasting 75 g OGTT to confirm the diagnosis if positive [2]. The revised criteria were implemented in the Australian Capital Territory (ACT) from January 2015 [9].

An additional source of variation which potentially impacts GDM diagnosis is the pathology testing procedure for the OGTT blood samples [9]. Unless certain pre-analytical procedures are undertaken to inhibit glycolysis once a blood sample is taken, glycolysis prior to analysis may result in a significant loss of glucose and the under-estimation of the glucose concentration [9]. One such procedure is rapid centrifugation (within 10 min) [9]. ACT Pathology changed its blood sample protocol from delayed centrifugation (all samples centrifuged at the end of the 2 h test) to early centrifugation (samples centrifuged within 10 min of collection) in June 2017 [9]. It has been suggested that an increase in GDM diagnoses may result in an increase in small-for-gestational-age (SGA) infants due to women, particularly those with mild GDM, restricting their dietary intake [10]. We therefore also included SGA infants as an outcome in our current study.

Comparisons to other regions or countries may potentially be unreliable due to different diagnostic thresholds and/or pathology centrifugation protocols used in GDM testing. As two changes in GDM testing have been implemented in the ACT with specific commencement dates, this presented us with a unique opportunity to undertake a retrospective comparative analysis of the prevalence of GDM and LGA and SGA infants over the periods in which each change in testing occurred. Our research question was to determine whether GDM testing changes were associated with changes in the prevalence of GDM and LGA and SGA infants. Therefore, our study had three aims: firstly, to determine the associations between the GDM testing regime and prevalence of GDM; secondly, to determine the associations between the testing regime and LGA and SGA infants; and thirdly, to compare the prevalence of LGA and SGA infants in women with and without GDM based on the testing regime.

2. Methods

2.1. Ethics Approval

Ethical approval was obtained from the relevant Health Research Ethics and Governance Office (no.: 2022.LRE.00194).

2.2. Study Design and Population

Our study was an observational retrospective analysis of de-identified data provided by ACT Health’s Birthing Outcomes System (BOS). The BOS contains routinely collected information from all births which occur in ACT public hospitals, with ACT Pathology being the main provider of public hospital pathology services. For our study, the data provided by the BOS were from a single public hospital in the ACT which accepts the pregnancy care transfers of higher-risk pregnancies from other hospitals in the ACT and the surrounding region of New South Wales (NSW).

ACT Pathology implemented the revised DC from January 2015, followed by a change to their pre-analytical blood processing protocols from June 2017 [9]. The date of each change was used to stratify the sample into three groups. To ensure each group represented approximately equivalent periods of time, June 2012 was selected as the commencement date of the study. The period from June 2012 to December 2014 was nominated as group 1, January 2015 to May 2017 as group 2, and June 2017 to December 2019 as group 3. December 2019 was chosen as the cut-off as the COVID-19 pandemic occurred from early 2020, resulting in a change to GDM testing and birthing protocols. Thus, group 1 represented women tested using the pre-revised DC combined with the delayed pathology centrifugation protocol (PCP); group 2 represented women tested using the revised DC combined with the delayed PCP; and group 3 represented women tested using the revised DC combined with the early PCP.

Women were allocated to groups based on the estimated date of GDM testing. It is recommended that women without either pre-existing glucose abnormalities or increased risk factors for hyperglycemia in pregnancy undertake routine testing at 24–28 weeks’ gestation [1,8]. As the BOS records the neonatal date of birth and estimated gestation, but not the date of testing, the testing was estimated to have been undertaken on average at 26 weeks’ gestation. In each group, the women were stratified into two subgroups: those with GDM and those without (non-GDM). Sensitivity analyses were undertaken by using estimated GDM testing dates at 24 weeks’ gestation and 28 weeks’ gestation (see Supplementary Materials Tables S4–S11).

As the maternal characteristics of obesity, ethnicity, parity and age are considered confounders [3], other maternal information included in our study was the body mass index (BMI), region of birth, parity and whether they were aged ≥ 35 years. The BMI is calculated as the weight in kilograms divided by the height in meters squared. The BOS records the BMI only and not the weight or height. The maternal BMI is usually recorded at the first ‘booking in visit’, normally between 12 and 14 weeks of pregnancy [11] (early BMI). When recorded it is rounded to the nearest whole number. Following the method used in a previous study [12], the BMI values were categorized into six groups: underweight (≤18 kg/m²); a healthy weight (19–24 kg/m²); overweight (25–29 kg/m²); obese class I (30–34 kg/m²); obese class II (35–39 kg/m²); and obese class III (≥40 kg/m²).

The consideration of the maternal region of birth followed the method used in a previous study [11]. Women were classified into broad categories: ‘all’ (regardless of their region of birth), ‘Australian-born’ and ‘Asian-born’. The Standard Australian Classification of Countries (SACC), Second Edition, was used to determine which countries to include in the categories [13]. Women who were neither Australian-born nor Asian-born were classified as ‘Other-born’.

Infant information included their gestational age and birthweight. The outcomes of interest were LGA infants, defined as a birthweight greater than the 90th centile for their gestational age, and SGA infants, defined as a birthweight less than the 10th centile for their gestational age [14]. Australian birthweight centiles by gestational age published by Joseph et al. (2020) [15] were used to identify LGA and SGA infants.

The inclusion criteria were singleton births, an estimated date of GDM testing and date of birth within the period from June 2012 to December 2019, an estimated gestation at birth of between 26 and 42 weeks and complete data. The exclusion criteria were multiple births, fetal deaths in utero, an estimated gestation at birth of less than 26 weeks or greater than 42 weeks, Type 1 or Type 2 diabetes and incomplete data.

2.3. Statistical Analysis

The unit of analysis was pregnancy; therefore, women with multiple pregnancies were potentially included more than once [16]. Categorical variables were presented using counts and percentages and the associations assessed using Pearson’s chi-squared independence test. The post hoc pairwise z-test for independent proportions was used to make comparisons between groups. The Bonferroni method was used to adjust the p-values for multiple comparisons. (See Supplementary Materials Tables S1–S3 for detailed between-group comparisons.)

It is considered problematic to compare odds ratios across groups and over time, even using the same predictors, due to the potential variability in the unobserved heterogeneity [17]. Consequently, in this study pre- and post-comparisons between the groups for each outcome of interest were undertaken using the categorical predictor variable ‘Group’ with levels of 1, 2 and 3, which represented groups 1, 2 and 3, respectively. The outcome of interest, LGA or SGA infants, was the dichotomous dependent variable. Firstly, level 1 was used as the reference to compare the odds of the outcome of interest after each GDM testing change (levels 2 and 3). Secondly, level 2 was used as the reference to compare the odds of the outcome of interest for level 3. Multivariate binary logistic regression was then performed using the forced entry method with adjustments for the categorical maternal characteristics of the early BMI group, region of birth, parity group and a maternal age ≥ 35 years, following the method used in a previous study [11].

For each group binary logistic regression was performed to compare the odds of the outcome of interest in women with GDM to women without GDM as the reference. Multivariate binary logistic regression was then performed using the forced entry method with adjustments for the categorical maternal characteristics stated previously.

Odds ratios were considered significant if the corresponding 95% confidence intervals did not contain the value 1 [18]. Cook’s Distances, the leverage, standardized residuals, DFbetas and the multicollinearity were checked to ensure the assumptions of no undue influence or high correlations between predictors were not violated. No adjustment for multiplicity was made for the subgroup analyses, so the 95% confidence intervals should not be used in place of hypothesis testing [19]. Statistical significance was set at p < 0.05. Microsoft Excel was used for data cleaning and preparation, and SPSS version 29 was used for the analyses.

3. Results

3.1. Maternal Characteristics

There were 26,098 birth records with a birth date between 1 June 2012 and 31 December 2019. After excluding records with an estimated GDM testing date pre-June 2012, missing data, multiple births, fetal deaths in utero, a gestational age at birth of <26 weeks or >42 weeks, and women with Type 1 or Type 2 diabetes, there remained 23,790 live singleton births with a birth date and estimated GDM testing date between 1 June 2012 and 31 December 2019 (Figure 1). Figure 1 shows the total population stratified into the three groups by the date, DC, PCP, testing strategy, length of time in months, and numbers of women.

Table 1. Total study population stratified by maternal and infant characteristics and by group.

	Total	Group 1	Group 2	Group 3	χ2, p-Values	Pairwise z-Test p-Values
					All Groups	G1 vs. G2	G2 vs. G3	G1 vs. G3
	23,790	8069	8035	7686
	n (%)	n (%)	n (%)	n (%)
Maternal characteristics
Early BMI					χ2 = 66.103, p < 0.001
≤18	1402 (5.9)	494 (6.1)	494 (6.1)	414 (5.4)		‡	‡	‡
19–24	11,856 (49.8)	4216 (52.2)	3998 (49.8)	3642 (47.4)		0.005	0.009	<0.001
25–29	5710 (24.0)	1857 (23.0)	1931 (24.0)	1922 (25.0)		‡	‡	0.010
30–34	2560 (10.8)	833 (10.3)	853 (10.6)	874 (11.4)		‡	‡	‡
35–39	1257 (5.3)	393 (4.9)	420 (5.2)	444 (5.8)		‡	‡	0.034
≥40	1005 (4.2)	276 (3.4)	339 (4.2)	390 (5.1)		0.025	0.032	<0.001
Region of birth					χ2 = 80.928, p < 0.001
Australian	15,746 (66.2)	5530 (68.5)	5350 (66.6)	4866 (63.3)		0.025	<0.001	<0.001
Asian	4855 (20.4)	1436 (17.8)	1612 (20.1)	1807 (23.5)		<0.001	<0.001	<0.001
Other	3189 (13.4)	1103 (13.7)	1073 (13.4)	1013 (13.2)		‡	‡	‡
Parity					χ2 = 32.576, p < 0.001
0	10,274 (43.2)	3634 (45.0)	3444 (42.9)	3196 (41.6)		0.016	‡	<0.001
1	8581 (36.1)	2739 (33.9)	2976 (37.0)	2866 (37.3)		<0.001	‡	<0.001
2	3191 (13.4)	1110 (13.8)	1030 (12.8)	1051 (13.7)		‡	‡	‡
3	1071 (4.5)	358 (4.4)	347 (4.3)	366 (4.8)		‡	‡	‡
≥4	673 (2.8)	228 (2.8)	238 (3.0)	207 (2.7)		‡	‡	‡
Age ≥ 35 years					χ2 = 30.864, p < 0.001
No	18,387 (77.3)	6386 (79.1)	6203 (77.2)	5798 (75.4)		0.009	0.028	<0.001
Yes	5403 (22.7)	1683 (20.9)	1832 (22.8)	1888 (24.6)		0.009	0.028	<0.001
GDM					χ2 = 755.277, p < 0.001
No	19,440 (81.7)	7302 (90.5)	6475 (80.6)	5663 (73.7)		<0.001	<0.001	<0.001
Yes	4350 (18.3)	767 (9.5)	1560 (19.4)	2023 (26.3)		<0.001	<0.001	<0.001
Infant characteristics
LGA					χ2 = 12.538, p = 0.002
No	21,205 (89.1)	7116 (88.2)	7223 (89.9)	6866 (89.3)		0.002	‡	‡
Yes	2585 (10.9)	953 (11.8)	812 (10.1)	820 (10.7)		0.002	‡	‡
SGA					p = 0.813
No	21,088 (88.6)	7138 (88.5)	7127 (88.7)	6823 (88.8)		‡	‡	‡
Yes	2702 (11.4)	931 (11.5)	908 (11.3)	863 (11.2)		‡	‡	‡

‡ p ≥ 0.05. LGA: large for gestational age; SGA: small for gestational age; BMI: body mass index; G1 vs. G2: group 1 versus group 2; G2 vs. G3: group 2 versus group 3; G1 vs. G3: group 1 versus group 3.

shows the total study population with the maternal and infant characteristics stratified by the groups. For infant characteristics there was a significant decrease in the group proportion for an LGA outcome from group 1 to group 2 (11.8% vs. 10.1%, p = 0.002).

All the maternal characteristics had a statistically significant association with the groups (all: p < 0.001). For the early BMI, the proportion of women with a BMI of 19–24 significantly decreased across each group and from group 1 to group 3 (52.2% vs. 47.4%, p < 0.001), while the proportion of women with a BMI ≥ 40 significantly increased across each group and from group 1 to group 3 (3.4% vs. 5.1%, p < 0.001). When compared to group 1, there was a significant increase in the proportion of women in group 3 with a BMI of 25–29 (23.0% vs. 25.0%, p = 0.010) or a BMI of 35–39 (4.9% vs. 5.8%, p = 0.034).

For the region of birth, 68.5% were Australian-born in group 1, and this decreased significantly to 66.6% in group 2 (p = 0.025) and to 63.3% in group 3 (p < 0.001). The group proportions for Asian-born women increased significantly in each group from 17.8% in group 1 to 20.1% in group 2 and 23.5% in group 3 (all: p < 0.001). The group proportion for a parity of zero decreased significantly from group 1 to group 2 (45.0% vs. 42.9%, p = 0.016). For a parity of one, there was a significant increase in the group proportion from group 1 to group 2 (33.9% vs. 37.0%, p < 0.001), with no further increase to group 3 (37.3%). For women aged ≥ 35 years, there were significant increases in each group’s proportion and from group 1 to group 3 (20.9% vs. 24.6%, p < 0.001).

The women with GDM are shown in

Table 2. Women with GDM stratified by maternal and infant characteristics and by group.

	Total	Group 1	Group 2	Group 3	χ2, p-Values	Pairwise z-Test p-Values
					All Groups	G1 vs. G2	G2 vs. G3	G1 vs. G3
N	4350	767	1560	2023
	n (%)	n (%)	n (%)	n (%)
Maternal characteristics
Early BMI					χ2 = 19.799, p = 0.031
≤18	114 (2.6)	23 (3.0)	36 (2.3)	55 (2.7)		‡	‡	‡
19–24	1423 (32.7)	287 (37.4)	514 (32.9)	622 (30.7)		‡	‡	0.002
25–29	1214 (27.9)	188 (24.5)	434 (27.8)	592 (29.3)		‡	‡	0.038
30–34	707 (16.3)	115 (15.0)	249 (16.0)	343 (17.0)		‡	‡	‡
35–39	445 (10.2)	89 (11.6)	154 (9.9)	202 (10.0)		‡	‡	‡
≥40	447 (10.3)	65 (8.5)	173 (11.1)	209 (10.3)		‡	‡	‡
Region of birth					p = 0.429
Australian	2317 (53.3)	395 (51.5)	848 (54.4)	1074 (53.1)		‡	‡	‡
Asian	1533 (35.2)	287 (37.4)	524 (33.6)	722 (35.7)		‡	‡	‡
Other	500 (11.5)	85 (11.1)	188 (12.1)	227 (11.2)		‡	‡	‡
Parity					p = 0.413
0	1679 (38.6)	313 (40.8)	593 (38.0)	773 (38.2)		‡	‡	‡
1	1703 (39.1)	274 (35.7)	617 (39.6)	812 (40.1)		‡	‡	‡
2	605 (13.9)	115 (15.0)	215 (13.8)	275 (13.6)		‡	‡	‡
3	215 (4.9)	34 (4.4)	77 (4.9)	104 (5.1)		‡	‡	‡
≥4	148 (3.4)	31 (4.0)	58 (3.7)	59 (2.9)		‡	‡	‡
Age ≥ 35 years					p = 0.875
No	2968 (68.2)	524 (68.3)	1057 (67.8)	1387 (68.6)		‡	‡	‡
Yes	1382 (31.8)	243 (31.7)	503 (32.2)	636 (31.4)		‡	‡	‡
Infant characteristics
LGA					p = 0.169
No	3825 (87.9)	659 (85.9)	1378 (88.3)	1788 (88.4)		‡	‡	‡
Yes	525 (12.1)	108 (14.1)	182 (11.7)	235 (11.6)		‡	‡	‡
SGA					p = 0.159
No	3834 (88.1)	662 (86.3)	1373 (88.0)	1799 (88.9)		‡	‡	‡
Yes	516 (11.9)	105 (13.7)	187 (12.0)	224 (11.1)		‡	‡	‡

‡ p ≥ 0.05. LGA: large for gestational age; SGA: small for gestational age; BMI: body mass index; G1 vs. G2: group 1 versus group 2; G2 vs. G3: group 2 versus group 3; G1 vs. G3: group 1 versus group 3.

, with the maternal and infant characteristics stratified by the groups. The maternal early BMI differed across the groups (p = 0.031). The between-group comparison showed that there was a higher proportion of women with a BMI of 19–24 in group 1 compared to group 3 (37.4% vs. 30.7%, p = 0.002) and a lower proportion of women with a BMI of 25–29 in group 1 compared to group 3 (24.5% vs. 29.3%, p = 0.038). All other maternal characteristic associations across and between the groups did not differ. For the region of birth, approximately 50% were Australian-born and 35% were Asian-born, 80% had a parity of zero or one, and 30% were aged 35 years or older.

The women without GDM are shown in

Table 3. Non-GDM women stratified by maternal and infant characteristics and by group.

	Total	Group 1	Group 2	Group 3	χ2, p-Values	Pairwise z-Test p-Values
					All Groups	G1 vs. G2	G2 vs. G3	G1 vs. G3
N	19,440	7302	6475	5663
	n (%)	n (%)	n (%)	n (%)
Maternal characteristics
Early BMI					p = 0.507
≤18	1288 (6.6)	471 (6.5)	458 (7.1)	359 (6.3)		‡	‡	‡
19–24	10,433 (53.7)	3929 (53.8)	3484 (53.8)	3020 (53.3)		‡	‡	‡
25–29	4496 (23.1)	1669 (22.9)	1497 (23.1)	1330 (23.5)		‡	‡	‡
30–34	1853 (9.5)	718 (9.8)	604 (9.3)	531 (9.4)		‡	‡	‡
35–39	812 (4.2)	304 (4.2)	266 (4.1)	242 (4.3)		‡	‡	‡
≥40	558 (2.9)	211 (2.9)	166 (2.6)	181 (3.2)		‡	‡	‡
Region of birth					χ2 = 28.047, p <0.001
Australian	13,429 (69.1)	5135 (70.3)	4502 (69.5)	3792 (67.0)		‡	0.007	<0.001
Asian	3322 (17.1)	1149 (15.7)	1088 (16.8)	1085 (19.2)		‡	0.002	<0.001
Other	2689 (13.8)	1018 (13.9)	885 (13.7)	786 (13.9)		‡	‡	‡
Parity					χ2 = 19.5599, p = 0.012
0	8595 (44.2)	3321 (45.5)	2851 (44.0)	2423 (42.8)		‡	‡	0.007
1	6878 (35.4)	2465 (33.8)	2359 (36.4)	2054 (36.3)		0.003	‡	0.009
2	2586 (13.3)	995 (13.6)	815 (12.6)	776 (13.7)		‡	‡	‡
3	856 (4.4)	324 (4.4)	270 (4.2)	262 (4.6)		‡	‡	‡
≥4	525 (2.7)	197 (2.7)	180 (2.8)	148 (2.6)		‡	‡	‡
Age ≥ 35 years					χ2= 11.234, p = 0.004
No	15,419 (79.3)	5862 (80.3)	5146 (79.5)	4411 (77.9)		‡	‡	0.003
Yes	4021 (20.7)	1440 (19.7)	1329 (20.5)	1252 (22.1)		‡	‡	0.003
Infant characteristics
LGA					χ2 = 12.896, p = 0.002
No	17,380 (89.4)	6457 (88.4)	5845 (90.3)	5078 (89.7)		0.001	‡	‡
Yes	2060 (11.1)	845 (11.6)	630 (9.7)	585 (10.3)		0.001	‡	‡
SGA					p = 0.942
No	17,254 (88.8)	6476 (88.7)	5754 (88.9)	5024 (88.7)		‡	‡	‡
Yes	2186 (11.2)	826 (11.3)	721 (11.1)	639 (11.3)		‡	‡	‡

‡ p ≥ 0.05. LGA: large for gestational age; SGA: small for gestational age; BMI: body mass index; G1 vs. G2: group 1 versus group 2; G2 vs. G3: group 2 versus group 3; G1 vs. G3: group 1 versus group 3.

, with the maternal and infant characteristics stratified by the groups. There were across-group differences in the maternal characteristics of the region of birth (p <0.001), parity (p = 0.012) and age ≥ 35 years (p = 0.004). For the region of birth, there was a significantly higher proportion of Australian-born women in group 1 compared to group 3 (70.3% vs. 67.0%, p < 0.001) and in group 2 compared to group 3 (69.5% vs. 67.0%, p = 0.007). There was a significantly lower proportion of Asian-born women in group 1 compared to group 3 (15.7% vs. 19.2%, p < 0.001) and in group 2 compared to group 3 (16.8% vs. 19.2%, p = 0.002). For a parity of zero, there was a significantly higher proportion of women in group 1 compared to group 3 (45.5% vs. 42.8%, p = 0.007). For a parity of one, there was a significantly lower proportion of women in group 1 compared to group 2 (33.8% vs. 36.4%, p = 0.003) and compared to group 3 (33.8% vs. 36.3%, p = 0.009). For an age ≥ 35 years, there was a significantly higher proportion of women not aged ≥ 35 years in group 1 compared to group 3 (80.3% vs. 77.9%, p = 0.003) and a significantly lower proportion of women aged ≥ 35 years in group 1 compared to group 3 (19.7% vs. 22.1%, p = 0.003).

3.2. Were GDM Testing Changes (to DC and PCP) Associated with Prevalence of GDM?

Table 1 shows that the GDM prevalence differed across the groups (p < 0.001). There was a progressively increasing proportion of women diagnosed with GDM from group 1 through to group 3 (9.5% vs. 19.4% vs. 26.3%; all: p < 0.001), such that the change in the DC was associated with more than a doubling in the prevalence (a 104% relative increase from group 1) and that the change in the PCP was associated with a further increased prevalence (a relative increase of 36% from group 2). This resulted in an overall relative increase in GDM from group 1 to group 3 of 177%.

3.3. Were There Changes in the Prevalence of LGA and SGA Infants?

Overall there was a reduction in the proportion of LGA neonates from group 1 to group 2 (11.8% vs. 10.1%, p = 0.002), but the proportion of LGA neonates did not differ between group 1 and 3 (10.7%) or group 2 and 3 (Table 1). The proportion of SGA neonates (11.4% overall) did not differ across the groups (Table 1).

In women with GDM Table 2 shows that there was no significant association between the groups and the LGA or SGA outcomes.

The odds ratios for women with GDM are shown in

Table 4. Bivariate and multivariate analysis of association between GDM status and outcomes of birthing LGA and SGA infants by group. For each, unadjusted odds ratio (OR), adjusted OR and 95% confidence intervals (CIs) are provided. Significance was set at p < 0.05, with significant figures bolded.

Reference Group	Group	Unadjusted OR (95% CI)	p-Value	Adjusted OR (95% CI) a	p-Value
(a) LGA
Non-GDM women
1	2	0.82 (0.74, 0.92)	<0.001	0.83 (0.74, 0.92)	<0.001
	3	0.88 (0.79, 0.98)	0.025	0.88 (0.79, 0.99)	0.026
2	3	1.07 (0.95, 1.20)	0.272	1.07 (0.95, 1.20)	0.294
Women with GDM
1	2	0.81 (0.62, 1.04)	0.098	0.73 (0.56, 0.95)	0.021
	3	0.80 (0.63, 1.02)	0.077	0.75 (0.59, 0.97)	0.029
2	3	1.00 (0.81, 1.22)	0.963	1.03 (0.83, 1.27)	0.786
Women with GDM compared to non-GDM women
1 Non-GDM	1 GDM	1.25 (1.01, 1.55)	0.041	1.17 (0.93, 1.47)	0.191
2 Non-GDM	2 GDM	1.23 (1.03, 1.46)	0.023	1.11 (0.91, 1.34)	0.305
3 Non-GDM	3 GDM	1.14 (0.97, 1.34)	0.108	1.06 (0.89, 1.26)	0.542
(b) SGA
Non-GDM women
1	2	0.98 (0.88, 1.09)	0.743	0.98 (0.88, 1.09)	0.686
	3	1.00 (0.89, 1.11)	0.960	1.00 (0.89, 1.11)	0.926
2	3	1.02 (0.91, 1.14)	0.796	1.02 (0.91, 1.14)	0.773
Women with GDM
1	2	0.86 (0.67, 1.11)	0.244	0.92 (0.70, 1.19)	0.513
	3	0.79 (0.61, 1.01)	0.056	0.82 (0.64, 1.06)	0.134
2	3	0.91 (0.74, 1.12)	0.394	0.90 (0.73, 1.11)	0.319
Women with GDM compared to non-GDM women
1 Non-GDM	1 GDM	1.24 (1.00, 1.55)	0.050	1.18 (0.94, 1.48)	0.163
2 Non-GDM	2 GDM	1.09 (0.92, 1.29)	0.340	1.04 (0.87, 1.25)	0.670
3 Non-GDM	3 GDM	0.98 (0.83, 1.15)	0.796	1.02 (0.85, 1.21)	0.849

^a Adjusted for maternal characteristics of early BMI, region of birth, parity and age ≥ 35 years. GDM: gestational diabetes mellitus; non-GDM: non-gestational diabetes mellitus; LGA: large for gestational age; SGA: small for gestational age; OR: odds ratio; CI: confidence interval.

. When compared to group 1, the results showed a significant reduction in the adjusted odds of birthing an LGA infant following the changes in the DC (p = 0.021) and PCP (p = 0.029). There was no difference in the odds of an LGA outcome following the change in the PCP when compared to the change in the DC. There were no significant differences in the odds of an SGA outcome.

In women without GDM Table 3 shows that the proportion of an LGA outcome differed across the groups (p = 0.002). Specifically, the LGA infant prevalence in group 1 was significantly larger than that of group 2 (11.6% vs. 9.7%, p = 0.001). The LGA infant prevalence in group 3 (10.3%) did not differ significantly to that of either group 1 or 2. There was no significant association for an SGA outcome.

The odds ratios for women without GDM are shown in Table 4. For an LGA outcome, and when compared to group 1, women without GDM had significantly reduced odds following the changes in the DC, both unadjusted (p < 0.001) and adjusted (p < 0.001), and the PCP, both unadjusted (p = 0.025) and adjusted (p = 0.026). There was no difference in the odds of an LGA outcome following the change in the PCP when compared to the change in the DC. There were no significant differences in the odds of an SGA outcome.

3.4. Comparison of LGA and SGA Outcomes in Women with GDM to Those of Women Without GDM Following Each GDM Testing Change

Table 4 shows the odds of LGA and SGA outcomes for women with GDM when compared to women without GDM for each group. Women with GDM had significantly higher unadjusted odds of an LGA outcome compared to women without GDM in group 1 (p = 0.041) and in group 2 (p = 0.023), but these became nonsignificant following adjustment. There were no significant differences between the subgroups in group 3. There were no significant differences between the subgroups for any group in the odds of an SGA outcome.

Although nonsignificant, group 1 women with GDM had 17% increased adjusted odds of birthing an LGA infant compared to women without GDM. Following DC changes this was reduced to 11% and following PCP changes this was reduced further to 6%.

Figure 2 shows that the proportion of the total LGA infants birthed by non-GDM women progressively decreased from 88.7% in group 1 to 77.6% in group 2 and to 71.3% in group 3. There was a corresponding increase in the proportion of the total LGA infants birthed by women with GDM from 11.3% to 22.4% to 28.7%. The proportion of the total SGA infants birthed by non-GDM women decreased from 88.7% to 79.4% to 74.0%, and for women with GDM it increased from 11.3% to 20.6% to 26.0%.

4. Discussion

4.1. Summary of Findings

Our study showed that the GDM prevalence increased significantly during each period in which a change to GDM testing was implemented. Non-GDM women were the only subgroup to have a significant decrease in the LGA infant prevalence, and this only occurred during the period in which the revised DC were introduced. However, both subgroups had significantly reduced adjusted odds of birthing an LGA infant after each testing change when compared to group 1. We did not observe a difference in the SGA infant prevalence in either subgroup even though more women were diagnosed with GDM. Our results suggest that the GDM testing changes established a trend towards similar odds of LGA and SGA outcomes in women with GDM compared to those without. Although we did not have access to blood glucose readings, the changes in GDM testing contributed substantially to the changes in the GDM prevalence across the three study periods. However, other changes in the population characteristics over time (e.g., the ethnic mix, BMI and age of the women) cannot be ruled out as having made some additional contributions.

Hyperglycemia increases the risk of having an LGA infant, with the frequency being greater in pregnant women with (untreated) GDM compared to those without [3,4,5]. A prospective cohort study of 2775 women by Langer et al. (2005) demonstrated that untreated GDM significantly increased the odds of an LGA outcome when compared to non-GDM controls (OR of 3.28, 95% CI of 2.53–4.6, p = 0.01) [3]. They also reported little difference in the adverse outcomes between women treated for GDM and those without GDM (OR of 1.06, 95% CI of 0.81–1.38). A randomized control trial of 1000 women with GDM undertaken by Crowther et al. (2005) reported that there were significantly fewer LGA infants born to women in the treatment group compared to those receiving routine care (13% vs. 22%, Adjusted Treatment Effect of 0.62, 95% CI of 0.47–0.81, p < 0.001) [4]. Landon et al. (2009) undertook a randomized trial of 958 women and determined that the treatment of mild GDM reduced the frequency of LGA births compared to controls (14.5% vs. 7.1%, RR of 0.49, 95% CI of 0.32–0.76, p < 0.001) [5]. In a 2014 systematic review and meta-analysis of 10 studies and 3881 women, Poolsup et al. concluded that GDM treatment significantly reduced the risk for LGA births (RR: 0.55; 95% CI: 0.45–0.67; p < 0.00001) [20].

However, there is no clear hyperglycemic threshold at which the risk of an LGA outcome significantly increases [1,6]. The HAPO study demonstrated significant associations between adverse neonatal outcomes, including LGA infants, and levels of maternal glucose within what had previously been considered a non-GDM range [6].

The ADIPS 2014 [8] revised the DC for the FPG threshold to include the range of 5.1–5.4 mmol/L and added a one-hour PG ≥ 10.0 mmol/L, both of which were previously considered not diagnostic of GDM. The revised 2 h threshold excluded the range 8.0–8.4 mmol/L, which was previously diagnostic of GDM but is now considered non-GDM. Following these changes, our study demonstrated a significant increase in the GDM prevalence from 9.5% to 19.4%. A significant reduction in the LGA outcomes from 11.6% to 9.7% was demonstrated in women without GDM. The increase in the GDM prevalence was higher than had been predicted or reported by others elsewhere in Australia with the implementation of the revised criteria. A prospective study conducted by Moses et al. (2011) in NSW compared the results of 1275 glucose tolerance tests using pre- and post-revised criteria [21]. The authors predicted an estimated increase in the prevalence from 9.6% to 13.0%, a relative increase of 35%. Likewise, a retrospective cohort study undertaken by Laafira et al. (2015) in Western Australia compared diagnoses in 3571 women using pre- and post-revised criteria and reported a 20% relative increase in diagnoses [22].

Lower diagnostic thresholds increase the GDM prevalence by including women with milder hyperglycemia, while higher thresholds identify fewer cases but with more severe hyperglycemia [23]. As women with mild hyperglycemia are at risk of having an LGA infant, if left undiagnosed and untreated this cohort may potentially increase the prevalence of LGA infants in women categorized as non-GDM. Ehmann et al. (2019) undertook a retrospective study of 641 singleton births in the ACT between 2011 and 2015 to assess the potential impact of the revised DC [10]. The authors found that a significantly higher proportion of LGA infants were born to women without GDM under the old criteria but who would have been diagnosed under the revised criteria compared with those born to all previously diagnosed women (22% vs. 5.2%, p < 0.0001) [10]. There was a 3% increase in the GDM prevalence, with a 21.8% increase due to women considered not to have GDM under the old criteria but diagnosed based on the new FPG range of 5.1–5.4 mmol/L and/or the 1 h threshold of ≥10.0 mmol/L. This was offset by an 18.9% decrease due to women considered not to have GDM under the revised criteria but diagnosed under the old criteria based on the now excluded 2 h PG range of 8.0–8.4 mmol/L [10].

However, the lower GDM prevalence in group 1 in our study was not only a consequence of the pre-revised DC but also potentially pre-analytical glycolysis due to the delayed PCP used at that time. In addition, in group 2 there potentially were women with undiagnosed mild hyperglycemia who were missed due to the changed 2 h thresholds and/or pre-analytical glycolysis due to the OGTT testing methodology. GDM diagnosis relies on the diagnostic sensitivity and specificity of the OGTT [24]. With the change in the PCP to prevent pre-analytical glycolysis, the results of our study showed a significant increase in the GDM prevalence from 19.4% to 26.3% (a 35.6% relative increase). A prospective cohort study by Potter et al. (2020) on the change in the PCP was undertaken using a similar ACT population and time period to those in our study [9]. The authors reported that the change resulted in a significant increase in the GDM prevalence from 11.6% to 20.6% (p < 0.00001, a 77.6% relative increase). Low-band GDM, that is, women with blood glucose readings within 0.2 mmol/L of the minimum thresholds, comprised 7.1% of diagnoses using early centrifugation. The greatest impact on increased diagnoses was a consequence of low- band FPG concentrations of 5.1–5.2 mmol/L [9]. This range falls within the HAPO study’s glucose category 5 (5.0–5.2 mmol/L), in which the frequency of LGA infants was 16.5% (aOR of 2.73, 95% CI of 2.25–3.31) [6]. Taken together, these results indicate that women with glucose concentrations near diagnostic cut-off points, and those with mild hyperglycemia, are vulnerable to false negatives due to pre-analytical glycolysis. The consequence is a potential underdiagnosis of GDM with missed opportunities to prevent adverse pregnancy outcomes, including LGA infants [25].

In our study there were no significant changes in the LGA outcomes for either subgroup during the period of the PCP change. This differed to the results of a prospective cohort study by Jamieson et al. (2021), who assessed the OGTT’s ability to identify pregnancies at risk of LGA infants in a high-risk population from rural and remote Australia [25]. The authors reported that almost one in five GDM cases missed due to pre-analytical glycolysis had an LGA infant. Following correction for glycolysis, the FPG samples had the largest shift in the LGA frequency, compared to 1 h and 2 h plasma glucose (PG) samples [25]. Using in silico analysis, Mansell et al. (2017) demonstrated that the greatest loss of glucose occurs in fasting samples as they have a longer time delay to processing compared to 1 h and 2 h samples, resulting in their lower diagnostic sensitivity and, consequently, higher false negative outcomes [24].

In the HAPO study the highest proportion of women with (untreated) GDM was diagnosed at the fasting threshold (8.3%), with an LGA infant prevalence of 19.5% [6]. As noted by Jamieson et al. (2021), in the HAPO study the odds of an LGA outcome increased by 38% with every increase of one standard deviation (1 SD) of 0.4 mmol/L in FPG (compared to 1 SD of 1.7 mmol/L in 1 h PG and 1.3 mmol/L in 2 h PG) [6,25]. Therefore, small differences in FPG values profoundly affect the GDM prevalence [26]. This suggests that false negative results, a consequence of pre-analytical glycolysis in FPG samples, may have the highest potential for missed opportunities to prevent LGA infants and other adverse outcomes.

Our study showed that, despite the GDM testing changes, the majority of LGA and SGA infants were birthed by non-GDM women. Ryan (2011) pointed out that most cases of LGA infants occur in women with normal glycemia, so that glucose is a weak predictor of LGA and instead, maternal obesity is a stronger one [27]. In the HAPO study 78% of all women giving birth to LGA infants were below the revised DC [6,27]; in our study these percentages decreased over the study period but remained above 70%.

A reason proposed by Desoye and Nolan (2016) for why some women who test negative for GDM have LGA babies is that of an exaggerated fetal glucose steal phenomenon, driven by fetal hyperinsulinemia [28]. Pedersen (1961) hypothesized that GDM occurs when maternal hyperglycemia results in fetal hyperinsulinemia [29]. Once established, it is postulated that this phenomenon may siphon glucose from the mother to the fetus [28,30]. Studies undertaken by Weiss et al. (2001) support the premise that this could mask an increase in a mother’s blood glucose levels following an OGTT so that their results appear normal [30].

Although we have considered GDM testing changes and their associations with LGA and SGA outcomes, GDM diagnosis provides an opportunity to mitigate the risk of adverse outcomes by providing treatment. McIntyre et al. (2021) state that the overall pregnancy outcomes are more likely to be related to daily glucose levels rather than those observed during diagnostic testing [7]. We assumed that all women diagnosed with GDM were treated; however, we have no information on treatment targets or adherence to them.

4.2. Strengths and Limitations

A strength of this study is that it was undertaken using routinely collected data from one large regional health system. The two types of changes implemented have each been shown to impact the GDM prevalence. Therefore, another strength of this study is it separately assessed the associations pre- and post-change for each of these two changes.

However, there are a few limitations. Of the women giving birth in the ACT public hospital system, approximately 15% are non-ACT residents [14]. These women may have been from regions in which different GDM DC and/or PCPs were used. Because of the acceptance of high-risk referrals at this single site in the ACT, the percentage of pregnancies complicated by GDM in the BOS database was likely higher than for all the pregnancies in the region. In addition, some women may have had multiple entries listed in the BOS if they had had more than one birth in the ACT health system. As noted in the methods, no adjustment was made for this, although the parity was included as a confounder in the multivariate analysis. The birthweight centiles used to assess infants as LGA were those published by Joseph et al. [15]. However, these may differ to other published centiles. The group allocation was based on our assumption of an estimated GDM testing date at 26 weeks’ gestation. To test this assumption, we undertook sensitivity analyses using estimated testing dates at 24 weeks’ gestation and 28 weeks’ gestation (see Supplementary Materials Tables S4–S11). Both analyses showed no difference to our main result, suggesting our key finding is robust to this assumption. However, some women may have been allocated to a different group if the actual testing date had been known. In addition, from the information provided to us by the BOS, it was not possible to identify or estimate whether women might have had testing outside of the 24–28-week period. It is not known whether any women were tested earlier in their pregnancy, particularly those with risk factors. For the BMI groups, the categories used in this study were rounded as this was the method of data entry in the BOS. Some women may have been included in a category in this study but may have more accurately been classified in a different category if more precise BMI data were available. Ethnic-specific BMI cut-offs were not used; however, the region of birth was considered in the adjusted analysis. While women diagnosed with and treated for GDM were identified from the BOS, it was not known whether individual women maintained targeted levels of glucose control during pregnancy. There were also no records of maternal weight gain during pregnancy. In addition, we did not address controversies surrounding either the one-step versus two-step testing methods or the cut-off points used in diagnostic thresholds.

We suggest that future research consider the potential reasons for our observed trends in the LGA and SGA infant prevalence, such as the maternal dietary intake and other factors.

5. Conclusions

The results of our study suggested that, prior to GDM testing changes, a category of women with mild hyperglycemia were classified as non-GDM. Being at risk of birthing an LGA infant but not treated effectively may have contributed to a higher prevalence of LGA infants. Following both GDM testing changes, and when compared to group 1, the odds of birthing an LGA infant were significantly reduced for both women with and without GDM. These results suggest that the GDM testing changes re-classified this category as having GDM. As the GDM prevalence increased, the LGA infant prevalence did not increase, indicating that more women who might otherwise have been at risk of birthing an LGA infant were able to prevent this outcome by undertaking treatment.

Therefore, the findings of our study suggest that the revised DC and early PCP are effective strategies for diagnosing GDM in pregnant women with previously unidentified hyperglycemia. While each testing change increased the number of women diagnosed with GDM, this was associated with reduced odds of having an LGA infant, when compared to group 1, in both the GDM and non-GDM subgroups.

In summary, this retrospective observational study suggests that implementing these changes may have resulted in a reduction in the LGA infant prevalence while not increasing the SGA infant prevalence.