Risk factor-based screening compared to universal screening for gestational diabetes mellitus in marginalized Burman and Karen populations on the Thailand-Myanmar border: An observational cohort

Background: Gestational diabetes mellitus (GDM) contributes to maternal and neonatal morbidity. As data from marginalized populations remains scarce, this study compares risk-factor-based to universal GDM screening in a low resource setting. Methods: This is a secondary analysis of data from a prospective preterm birth cohort. Pregnant women were enrolled in the first trimester and completed a 75g oral glucose tolerance test (OGTT) at 24-32 weeks' gestation. To define GDM cases, Hyperglycaemia and Adverse Pregnancy Outcomes (HAPO trial) criteria were used. All GDM positive cases were treated. Sensitivity and specificity of risk-factor-based selection for screening (criteria: age ≥30y, obesity (Body mass index (BMI) ≥27.5kg/m 2), previous GDM, 1 st degree relative with diabetes, previous macrosomia (≥4kg), previous stillbirth, or symphysis-fundal height ≥90th percentile) was compared to universal screening using the OGTT as the gold standard. Adverse maternal and neonatal outcomes were compared by GDM status. Results: GDM prevalence was 13.4% (50/374) (95% CI: 10.3-17.2). Three quarters of women had at least one risk factor (n=271 women), with 37/50 OGTT positive cases correctly identified: sensitivity 74.0% (59.7-85.4) and specificity 27.8% (3.0-33.0). Burman women (self-identified) accounted for 29.1% of the cohort population, but 38.0% of GDM cases. Percentiles for birthweight (p=0.004), head circumference (p=0.002), and weight-length ratio (p=0.030) were higher in newborns of GDM positive compared with non-GDM mothers. 21.7% (75/346) of newborns in the cohort were small-for-gestational age (≤10 th percentile). In Burman women, overweight/obese BMI was associated with a significantly increased adjusted odds ratio 5.03 (95% CI: 1.43-17.64) for GDM compared with normal weight, whereas in Karen women, the trend in association was similar but not significant (OR 2.36; 95% CI 0.95-5.89). Conclusions: Risk-factor-based screening missed one in four GDM positive women. Considering the benefits of early detection of GDM and the limited additional cost of universal screening, a two-step screening program was implemented.


Introduction
Gestational diabetes mellitus (GDM) is rising in tandem with obesity globally, including in South-and South-East Asia 1 . The prevalence of GDM in Thailand is estimated between 6.1% and 29.2% 1,2 . In Myanmar, there is insufficient data to provide reliable estimations of the GDM prevalence 1 . Detection of GDM is important as it is associated with neonatal macrosomia, neonatal hypoglycaemia and an increased risk for birth complications, such as shoulder dystocia and the need for caesarean section 3-5 . Furthermore, GDM is associated with an increased risk of preeclampsia, and entails a tenfold risk of developing type II diabetes and doubles the risk of cardiovascular events later in life 6,7 .
In absolute numbers more women are diagnosed with GDM in low-and middle-income countries (LMIC although relative estimates are similar between LMIC and highincome countries (HIC): 13.5% and 13.4%, respectively) 8 . Within HIC, migrant women have a higher risk for GDM and associated adverse birth outcomes, but this is poorly evidenced for migrants in LMIC 9 . In South-East Asia domestic as well as international migration is a dominant feature and access to health care for migrants is problematic 10,11 . While most women receive some form of antenatal care (ANC), screening for GDM is often not available 12,13 . In addition, awareness of GDM is limited, as are adequate protocols and tools to monitor blood glucose, which hinders best-practice management 13, 14 .
Officially Thailand has approximately 2 million migrant workers predominantly from Myanmar, as well as an unknown number of undocumented migrants. Shoklo Malaria Research Unit (SMRU) has provided health care to both the refugee and migrant populations residing along the Thailand-Myanmar border. In the pregnant migrant population attending SMRU antenatal (ANC) clinics, the nutrition transition has been marked by a two-fold increase in first trimester overweight in just over a decade, aggravated by limited awareness of healthy diets and lifestyle 15,16 . These trends in marginalized populations are worrying given the greater risk of cardiometabolic effects occurring at lower BMI in Asians than in white Europeans 17 .
In a meta-analysis, Lee et al. described a GDM prevalence of 11.5% in Asian women and identified the following risk factors: multiparity, previous GDM, or pregnancy-induced hypertension (PIH), a family history of GDM and an increased maternal body mass index (BMI ≥25kg/m 2 ) 18 . An obstetric history of preterm birth, macrosomia, stillbirth, or an infant with congenital anomalies are also recognised GDM risk factors 18 .
GDM diagnosis and management improves maternal and perinatal outcomes, although this is largely evidenced from HIC 13,19 . Both universal and risk-factor-based screening are common practices, with no international consensus about best practice 2,20,21 . In 2011-2012, one of the first surveys conducted in a refugee camp reported a GDM prevalence of 10.1% (95% CI 6.2-14.0%) on the Thailand-Myanmar border with GDM being significantly associated with increased maternal age and parity, and low literacy 20 . Although the proportion of caesarean section and obesity (BMI ≥27.5kg/m 2 ) were higher among women with GDM, this difference was not significant 20 . In the low-resource setting of the refugee camp, the decision at that time was to commence efforts to screen for GDM based on risk factors using the Hyperglycaemia and Adverse Pregnancy Outcomes (HAPO) criteria 22 . SMRU implemented this approach in all its antenatal care clinics on the border in 2018.
The study presented here aimed to evaluate the performance of two screening methods for GDM detection: risk-factor-based identification of pregnant women who were then screened by an OGTT, which was routinely used in antenatal care clinics for migrant women, to universal screening by OGTT. Within this cohort, risk factors for GDM were examined and adverse maternal and neonatal outcomes were evaluated in women with and without GDM.

Ethical approval
The study was approved by the ethics committee of the Faculty of Tropical Medicine, Mahidol University, Bangkok, Thailand (Ethics Reference: TMEC 15-062, initial approval 1

Amendments from Version 1
Version 2 of the manuscript was submitted in response to the reviewer comments on version 1. Comments and suggestions by the reviewers were considered to revise the manuscript or rebuked were appropriate (see response to reviewer comments). Most notably all references to environmental factors (i.e., seasonality of GDM diagnosis) has been removed from the manuscript. This also resulted in the removal of Figure 3a and Figure 3b from version 2 of the manuscript.
The abstract was amended to include all risk factors that were screened for the risk-factor based screening procedure. Sample size justification and a power estimation was added. Some items (e.g., Asian BMI definitions, exact nature of the risk-factor based screening or rational to include symphysis-fundal height (SFH) in the analysis) were clarified. Table 1 was complemented with the addition of number of pregnant women presenting with a SFH ≥90 th centile and figures describing gestational weight gain; how these compare between non-GDM and GDM women was also added.
Centiles for head circumference, length, and weight for length ratio, as published by the Intergrowth 21 st consortium, were added to Table 2, together with the number of neonates admitted to the special care baby unit (SCBU) and number of neonates diagnosed with hypoglycaemia.
As suggested by the reviewer, a paragraph presenting results of oral glucose tolerance test (OGTT) results was added.
A sentence on historical context of GDM screening was added to the discussion and the fact the there is no international consensus on the best screening approach exists, together with the ongoing debate whether screening criteria derived from high-resource settings are applicable to low-resource settings is now discussed in more detail.
Strengths and limitations were expanded to address comments from the reviewers. Lastly, the conclusion was amended to highlight the translational impact of this analysis.

REVISED
December 2015), the Oxford Tropical Research Ethics Committee (Ethics Reference: OxTREC: 33-15, initial approval 16 December 2015) and reviewed by the local Tak Province Community Ethics Advisory Board. The study was conducted in full conformity with the Declaration of Helsinki and followed regulations of the ICH Guidelines for Good Clinical Practice.

Study design
This is a secondary analysis of data from an observational preterm birth cohort study with data collected prospectively between September 2016 and February 2019 in women enrolled in their first trimester of pregnancy (ClinicalTrials.gov Identifier: NCT02797327) with GDM screening occurring from December 2016 to November 2018.
Study setting SMRU was established more than three decades ago and combines research and humanitarian work that serves the migrant population alongside the Thailand-Myanmar border. To be accessible within these communities, which largely depend on below minimum wage jobs, SMRU operates free-of-charge walk-in clinics offering universal antenatal care, as well as 24-hour delivery services, led by trained personnel originating from the local population.
At the same clinics, women may be invited to participate in research. The study was explained to all pregnant women attending SMRU ANC clinics in the first trimester and they were invited to participate if they met the study inclusion criteria and enrolled if consent was forthcoming. Informed consent was obtained in the form of a signature or in the event of an illiterate participant by thumbprint coupled with a confirmatory signature by an impartial literate witness.

Sample size
A detailed description of the study protocol and SMRU routine ANC procedures are available elsewhere 23 . Briefly, women were followed fortnightly throughout pregnancy, at delivery, and in the postpartum period. The planned sample size of 400 in the original cohort study was based on estimated preterm birth rates (of approximately 8%) and on the following inclusion criteria: a viable, singleton first trimester pregnancy and an unremarkable medical and obstetric history e.g., no history of caesarean section. For this secondary analysis of the original cohort to determine appropriateness of GDM risk-factor-based screening, additional exclusion criteria were miscarriage prior to GDM screening, maternal death, lost to follow-up, withdrawal of consent (primary cohort), and if OGTT was performed late (gestational age (GA) ≥33 weeks) or not done at all. Women who did not complete follow-up to delivery were replaced as permitted in the original protocol. At an expected GDM rate of 10%, a sample size of 400 is expected to be sufficient to determine population prevalence 20 .

Study variables
Baseline characteristics, regular prenatal symphysis-fundal height (SFH) measurements, blood pressure, weight, and assessment of gestation by ultrasound, as well as birth outcomes, were collected by trained ANC staff and midwives in accordance with the study protocol. GA was estimated by crown rump length measured by first trimester ultrasound 24 . Body-mass index (BMI) definitions followed recommendations for Asian BMI groups: underweight <18.5 kg/m 2 ; normal weight 18.5 to <23 kg/ m 2 ; overweight 23 to <27.5 kg/m 2 ; obese ≥27.5 kg/m 217 .
While the study protocol specified GDM screening with OGTT at 24-26 weeks of gestation, the HAPO study target time for testing was at 28 weeks (24-32 weeks) 22 . Therefore, OGTTs to 32 weeks of gestation were included in this analysis. In women with a history of GDM, an OGTT was performed as early as possible in pregnancy and repeated at 24-26 weeks if previously negative. GDM diagnosis was based on HAPO trial cut-offs: a fasting capillary blood glucose measurement of ≥92mg/dL, ≥180mg/dL one hour or ≥153mg/dL two hours after ingestion of 75g glucose were considered positive 22 .

Risk-factor based screening
In 2018, risk-factor-based screening for GDM commenced at SMRU clinics. The risk factors were based on a survey in Karen and Burmese women in a SMRU refugee clinic screened at 24-28 weeks with a 75-gram OGTT using the HAPO trial cut-offs, where prevalence was 10.1% (95% CI 6.1-14.0). Risk factors in positive cases and review of recommendations from UK and Australia, both of which have populations of South-East Asian women, and Thailand resulted in the final list 20 . The risk factors for GDM screening required at least one positive finding among the following 10 criteria: (i) age ≥30 years, (ii) obesity (BMI ≥27.5kg/m 2 , the WHO definition for Asian populations) 17 , (iii) GDM in a previous pregnancy, (iv) family history (1 st degree relative) of diabetes mellitus (although this is of reduced sensitivity in LMIC as access to diabetes screening is limited), (v) previous macrosomia (≥4kg), (vi) previous stillbirth, (vii) SFH ≥90th percentile, (viii) previous caesarean section regardless of birth-weight, (ix) 2+/3+ glucose on a urine dipstick test, or (x) polycystic ovarian syndrome (PCOS). The following criteria were not included in the analysis: women with a previous caesarean section, as they were excluded from the original study protocol, PCOS, as it was not encountered, and glucosuria, as there was no routine screening, leaving seven criteria.

Maternal and Neonatal Outcomes
In resource-limited settings, assessment of the uterus size by SFH measurement as a proxy for fetal size has been suggested as a first level screening tool for fetal growth assessment. SFH measurement is a straightforward and inexpensive method, but its precision is controversial 3 . A previously published bespoke SFH growth curve has been in use for more than 10 years in the pregnant population along the Thailand-Myanmar border 25 ; however, whether increased SFH using this local growth curve is a useful addition to the identification of GDM (macrosomia is a common adverse effect of GDM) has not been assessed.
Serial SFH measurements were included from 16 weeks of gestation on a two-weekly basis and data was examined using both, local population and international centiles 25,26 .
Gestational weight gain was defined as the final maternal weight measured not more than four weeks prior to birth, minus the weight measured at the first antenatal visit. For women with a normal BMI at enrolment (between 18.50 and 24.99kg/m 2 ), Intergrowth-21 st standard percentiles for each weight measurement from ≥26 weeks and ≤40 weeks of gestation were calculated 27 .
Neonatal anthropometry (i.e., birthweight, head circumference, and length) were only considered if measured within 72 hours of birth. If women gave birth at SMRU, the neonate was weighed on a digital SECA 354 scale (precision 5g) with weekly calibration. Percentiles and z-scores for neonatal anthropometry were calculated using standards as published by the Intergrowth-21 st Project 28 . Born too small or large for GA (SGA, LGA) were defined as ≤10 th and ≥90 th percentile, respectively.
Standard management of infants admitted to the special care baby unit included measurement of blood glucose and treatment for neonates with blood glucose below 45 mg/dL

GDM management
If GDM was diagnosed, all women were counselled about lifestyle modification (e.g., diet and exercise) and, due to the unavailability of glucose self-monitoring in the population, the status of GDM control was monitored weekly or every two weeks at the clinic. Monitoring was as follows: women with GDM were asked to attend fasting and blood glucose was checked on arrival; then women ate a typical meal and were retested after one hour (post-prandial) with the desired value of <90 mg/dL (fasting) and <140 mg/dL (after one hour) for satisfactory control. Treatment was provided either directly or if non-pharmacologic interventions led to insufficient glucose control, with metformin as the first choice and glibenclamide as an additional oral agent. Due to the lack of home-based glucose monitoring options and the absence of adequate storage facilities, insulin is rarely prescribed in this population.

Statistical analysis
Data were analysed using Stata, version 17.0 (TX, USA) (Stata, RRID:SCR_012763, https://www.stata.com/). Normally distributed continuous data were presented as means with standard deviation (SD) and non-normally distributed data as medians with interquartile range (IQR). Baseline characteristics as well as birth outcomes were compared between women with and without GDM. For continuous variables, the Student's t-test or Mann-Whitney U test were used, and categorical variables were compared using the Fisher's exact or Chi-square test. Univariate associations were quantified using logistic regression. To evaluate the predictive ability of the risk factors used in the current screening approach to identify women with GDM, all risk factors were combined into one logistic regression model, using GDM as the outcome. The sensitivity and specificity of risk-factor-based screening criteria was calculated using OGTT as the gold standard. An any positive test principle (i.e., if any of the GDM risk factors stated above were positive, an OGTT was performed) was the basis for this assessment. For further in-depth analysis and to identify risks and potential risk groups for GDM in this population, age (30 or older, vs. all others), smoking (yes/no), ethnicity (Karen and Burman), and BMI groups underweight, normal weight (reference group) and overweight/obese were explored using interaction terms and logistic regression modelling.
Overall, 23 women (6.1%) were obese (BMI ≥27.5kg/m 2 ). In the group of women who self-identified as being of Burman descent the GDM prevalence was 17.4% (19/109) compared to 11.7% (29/247) in women of Karen descent and 11.1% (2/18) in women of other ethnicities. Burman women accounted for 29.1% of the cohort population, but 38.0% of GDM cases (Table 1). There were more women with GDM with an SFH ≥90 th centile during pregnancy with gestational week ≥24, 68.0% vs. 52.8%, p=0.044 (Table 2). In particular, from about 224 days (32 weeks) onwards, women with GDM appeared to have larger SFH when compared with women without GDM ( Figure 2).

OGTT test results
As expected, the absolute blood sugar levels (BSL) levels were higher in the GDM positive group (Table 3). Of the women with GDM, 88.0% (44/50) had only one of the three glucose measurements above the cut-off, 10% (5/50) had two of three glucose measurements above the defined threshold and in only one study participant (1/50, 2.0%) all three measurements were above the defined limits. Screening with fasting and two-hour results, as performed in some institutions to reduce costs, would result in only 66% (33/50) of the GDM cases being detected in this study population (Table 3).
Risk-factor-based screening for GDM There were 37 women in the GDM positive group and 234 women in the GDM negative group who had at least one risk factor, translating into an overall proportion of 72.5% (271/374) ( macrosomia had a positive (wide confidence interval) and non-significant association due to the small number of cases (6.59, 95% CI 0.41-107.1, p=0.185). All other risk factors were not significant at p>0.20.

GDM risk in Burman and Karen ethnic groups
Risk factors for GDM were examined separately for the two main ethnic groups in the population by multivariate analysis (Table 4). After adjustment, overweight or obese Burman women were at a five-fold higher risk of GDM. A different relationship between BMI and GDM was apparent for Karen women where the risks were similarly elevated (non-significant) for both underweight and overweight or obese women (Table 4).

Discussion
The most consequential early GDM definition was published by O´Sullivan and Mahan in 1964 29 . Their criteria were then tried and adapted over decades with the culmination in the HAPO trial 30,31 . Currently there is no consensus on the optimal screening approach with Europe leaning more to risk-factor based screening and USA towards glucose challenge tests; and it is not entirely clear whether criteria derived from high-resource settings are adequate for institutions in low-resource settings 30,32 . Hence, as the main objective of this manuscript was to assess the performance of the risk-factor-based screening used in routine clinical practice and draw conclusions of its fitness, the presented cohort was explored by an any positive approach. This was possible because all women had data on the relevant risk-factors collected, and as they were part of a preterm birth study cohort, all women had an OGTT done. The analysis identified the shortcomings of current clinical practice as almost one in four women with GDM would have been missed based on risk-factor-based selection for screening when compared with universal screening by 75g OGTT.
While the risk-factor-based screening had a sensitivity of 74.0% (95% CI 59.7-85.4), it lacked specificity 27.8% (95% CI 23.0-33.0) and resulted in an inadequate positive predictive value of 13.7% (95% CI 9.8-18.3). Reasons for this underperformance could be related to the limited size of the cohort; due to exclusion of women with a previous caesarean section (potentially due to undiagnosed GDM) from the original cohort; or that risk-factor-based screening is inherently weak for GDM diagnosis in South-East Asian women. The low incidence of reported prior history of GDM or family history of diabetes, most likely results from the limited extent of testing in this population that has limited access to health care 33 .
At least one in seven 'healthy' migrant women presenting to antenatal care in this study cohort had GDM based on the 75g OGTT and thus identifying GDM as a significant health problem in Burman and Karen migrants on the Thailand-Myanmar border. These findings are similar to other migrant populations globally who have to make food choices based on limited expenditure 34 . The BMI-related differences in risk factors observed on regression analysis for GDM in Karen and Burman women may relate to different diets and smoking habits between these ethnic groups. A more detailed dietary analysis based on quantitative 24-hour food recall is currently under evaluation. The similar odds for GDM in underweight and overweight/obese Karen women may be related to the thin-type II diabetic phenotype where individuals are at increased risk at a lower BMI 35 36,37 . Of greatest concern is the propensity for this group of patients with undernutrition to have worse diabetes. Ethnohistorical Burman and Karen are distinct populations with their own pheno-and genotypic peculiarities 38 .
As the slightly different GDM risk-profile is based on a small sample size, these findings must be confirmed in larger cohorts.
In this analysis, there was a positive association between GDM and higher percentiles for infant birthweight, larger head circumference and weight-length ratio composition but no difference was seen in mode of delivery, postpartum haemorrhage, perineal damage or Apgar score by GDM status [39][40][41] . Given that pregnant women with an unremarkable medical and obstetric history were prioritized in the cohort and women with GDM received treatment following the abnormal OGTT result, the low rate of adverse birth outcomes is not unexpected. The high rate of small for gestational age (one in five) newborns has been reported previously and highlights the double burden of nutrition in this population but may also signal a risk for thin-type II diabetes 15,35 . Data from other South-East Asian populations suggest that obese women with GDM have a higher risk of adverse outcome when compared to normal weight pregnant women with GDM 42 . However, considering a significant increase in perinatal morbidity in women with uncontrolled GDM compared to women with adequately treated GDM, different strategies of GDM management for obese and non-obese pregnant women does not seem appropriate at this point 43 .
Early detection of GDM may prevent the need for caesarean section, which limits total expenditure per pregnancy. While the cost for an individual OGTT is small (i.e., approximately 18 THB (0.54 USD) for one glucose test strip, 7.5 THB (0.22 USD) for 75g glucose powder), costs add up if thousands of pregnant women are universally screened each year. Considering the average cost for caesarean section in 2020 for migrant women was 27,695 THB (approximately 824 USD) when referred to the public hospital  44,45 . Since access to adequate diabetes monitoring and pharmacological intervention is severely limited outside of pregnancy in resource-limited settings, there may be added benefit to universal screening in LMIC. The counselling women receive during pregnancy about their GDM may be the first and only information provided on lifestyle modification to prevent the development of type II diabetes later in life 46 . Reducing from three (fasting, one hour, two hours) to two (fasting, two hours) tests to bring down costs is not a useful alternative in this population as nearly nine in 10 were positive at a single timepoint distributed across all three time points. As the majority (68.7%) of GDM positive women in this study used oral hypoglycaemic agents, there is a need for a better understanding of effective lifestyle interventions in this marginalized group 2,16,47 .
The findings on the usefulness of SFH contributes to the ongoing debate on the use of international vs. local centiles. The proportion of pregnant women presenting with a SFH ≥90 th centile using local centiles differs markedly compared to the proportion when using international centiles. Using international standards for SFH, most GDM positive women would not be signalled as women with a problem in this population 26 . This most likely arises from maternal anthropometric differences (e.g., the greater than 10cm difference in maternal height) between the populations participating to the cohorts for centile curve calculation.
From 24 weeks EGA there was a significantly higher proportion of women in the GDM positive group with a SFH above the 90 th centile compared to women without GDM. While this suggests that SFH may have a role, the fact that more than half (52,8%) the women with no GDM had at least one SFH measurement ≥90 th centile renders SFH for GDM as rather unspecific. In addition, the timeframe of detection of increased SFH (32 weeks) is later than when an OGTT identifies GDM.

Strengths of this study
The strengths of this study include first trimester enrolment and ultrasound dating allowing accurate assessment of neonatal anthropometry based on gestation. The risk of information bias is reduced by the prospective cohort design with minimal missing data. There was also close monitoring throughout pregnancy with a high number of antenatal care visits (median 16,. Furthermore, weight and SFH were measured with calibrated instruments and by well-trained personnel. In addition, this analysis has had a direct local impact resulting in the implementation of universal GDM screening for all women with a two-step approach; with the first step being a glucose challenge test (i.e., 50 g non-fasting oral glucose load, followed by a 1-hour glucose measurement) with 1-hour levels of ≥200 mg/dL being diagnostic of GDM and values 140-199 mg/dL requiring a 2 nd step, namely a complete OGTT. This pragmatic choice to increase the number of women screened and minimize the burden of a full OGTT in all women follows the recommendation of the American College of Obstetricians and Gynecologists.

Potential study limitations
Women with a complicated obstetric or medical history were excluded from the original study. As SMRU does not perform caesarean sections in their clinics, women thought to be at risk of this pregnancy complication were excluded from the original study as they were predicted to not be able to provide a complete set of samples. This was a selection bias for healthier pregnant women, potentially leading to an underestimate of the GDM prevalence in this border population, i.e., the study likely presents the minimum GDM rate in the community of pregnant women. With the selection bias and treatment for all GDM positive women there was a low number of complications; among 37 risk-factor positive cases (vs 17 risk-factor negative cases), there were seven complications overall (preterm (n=1), stillbirths (n=0), caesarean section (n=2), postpartum haemorrhage (n=1), and LGA (n=3)). The study design did not allow exploration of whether those identified in the high-risk group were also the same women who are likely to have complications from GDM.
Due to the relatively small sample, the suggestion of differences in the risk of GDM between the two major ethnic groups requires further verification.

Conclusions
These findings imply that GDM is a problem at the Thailand-Myanmar border with Burman women who are overweight/obese being at the highest risk. GDM determined by risk-factor-based screening performed sub-optimally in this rural, resource-constrained pregnant population. Access to universal screening for GDM can potentially reduce negative impacts for an individual pregnancy but also provide an opportunity to sensitize people in marginalized populations of their potential increased risk for type II diabetes later in life. Considering that additional costs for universal screening appear limited, this is the preferred policy in this population.

Open Peer Review
Why was obesity defined as BMI greater than 27.5? 4.
Better clarification is needed about the analysis of risk factor exploration for GDM and how this is distinct from the risk factor-based screening.

5.
The first sentence of the second paragraph is confusing as written; the prevalence same, but more people living in LMICs so the overall number is higher.

6.
The introduction is a bit long and may not need to be. The connection with malnutrition in the introduction is not entirely clear. The paragraph on SFH seems extraneous. The paragraph on GDM and the environment is interesting but a little off-topic in the manuscript.

7.
Intro sentence that states "adequate diagnosis and mgmt. improves outcomes in GDM"-is this known for LMICs or migrant populations? It might not improve outcomes-this article helps contributes to that literature but is not sufficient to answer the question.

8.
Methods section detailing screening using HAPO criteria-sixth paragraph-seems to reference #24 twice in the same sentence.

9.
Sample size calculations unclear-likely as unclear as written the purpose of the primary cohort.

10.
Exclusion criteria-one is listed as miscarriage. Is this history of miscarriage or pregnancy loss before glucose screening in this population?

11.
Methods-could eliminate some of the details of SFH measurements and reference prior work by this group.

12.
Details of newborn anthropometry measurements-some could be relegated to a supplement to simplify the manuscript. maternal overweight?
Consider adding some historical context of literature from HIC and the switch from RFbased screening to universal.

20.
Consider adding some historical context of the debate around whether to screen for GDM at all-expensive, onerous, does it meaningfully improve outcomes? Have we yet answered this for LMICs?

21.
What proportion of women have at least one risk factor-would it be almost universal? 22.
The paragraph on SFH in conclusions is a bit hard to understand. 23.

If applicable, is the statistical analysis and its interpretation appropriate? Partly
Are all the source data underlying the results available to ensure full reproducibility? Yes

Results
You report non-significant differences in baseline characteristics, which is confusing to the reader. It would be clearer to state there were no significant differences observed in those variables, or alternately that a null association cannot be excluded.

1.
Again, be careful reporting non-significant trends in the difference in prevalence between ethnic groups.
2. Figure 2: The lines seem pretty much the same to me. Is this a significant difference? It would be useful to articulate this in cm difference if it is clinically relevant. 3.
When reporting the birth outcomes, it is important to know gestational age at delivery as the difference in birth weight may be gestation related.

4.
The CI for LGA is very large indicating not many babies were LGA. It would be helpful for the reader to have the absolute number of non-GDM and GDM LGA babies in the text. The fact that so many more GDM babies are SGA should also be highlighted here.

5.
In the risk factors-based screening for GDM section, you start by reporting the sensitivity of the different tests in the OGTT. This should have been pre-empted in the methods and would be better under a subheading about OGTT (relating to Table 3). 6.
I am a bit confused as you report risk factor screening has 74% sensitivity and 27.8% specificity, however, in the results, you then state none of the risk factors was associated with the outcome (GDM). Does this mean that women with GDM had more than one risk factor and this was the difference? What was your definition of a risk factor for the calculation of sensitivity and specificity? If you are using the OGTT values themselves, doesn't that defeat the purpose of risk factor-based screening? 7. Table 4: I worry that you are splitting your sample size and with multiple testing, it is not surprising that eventually one of your tests came up positive. There is no discussion of how you will handle false detection rates in the methods.

8.
The information on seasonality is interesting, although it is unclear how it relates to the study objective. Did you include seasonality as a risk factor? 9.

Discussion
I disagree that the risk factors-based screening was "grossly inadequate", although as per my comments above I am confused as to what the risk factors were that were included in your screening tests. You did not show any great changes in perinatal outcomes, other than a slight increase in the birthweight of babies, which in a setting with such high rates of LGA may not be a bad thing in itself.

1.
The value of screening and treating GDM in non-obese Asian populations has been questioned (see Yue et al., BMC Pregnancy and Childbirth, 2022 1 ). Whilst I appreciate that an everted CS will save a lot of money, the question that arises from this is with such low CS rates (around 5% overall), how much can you extrapolate from data from HAPO derived in 2.
very differently resource settings?