- Email: [email protected]
First trimester prediction of gestational diabetes mellitus: A clinical model based on maternal demographic parameters
Accepted Manuscript First Trimester Prediction of Gestational Diabetes Mellitus: A Clinical Model Based on Maternal Demographic Parameters Arianne Natasha Sweeting, Heidi Appelblom, Glynis P. Ross, Jencia Wong, Heikki Kouru, Paul F. Williams, Mikko Sairanen, Jon A. Hyett PII: DOI: Reference:
S0168-8227(16)30525-3 http://dx.doi.org/10.1016/j.diabres.2017.02.036 DIAB 6896
To appear in:
Diabetes Research and Clinical Practice
Received Date: Revised Date: Accepted Date:
5 September 2016 19 February 2017 28 February 2017
Please cite this article as: A.N. Sweeting, H. Appelblom, G.P. Ross, J. Wong, H. Kouru, P.F. Williams, M. Sairanen, J.A. Hyett, First Trimester Prediction of Gestational Diabetes Mellitus: A Clinical Model Based on Maternal Demographic Parameters, Diabetes Research and Clinical Practice (2017), doi: http://dx.doi.org/10.1016/j.diabres. 2017.02.036
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
First Trimester Prediction of Gestational Diabetes Mellitus: A Clinical Model Based on Maternal Demographic Parameters
Arianne Natasha Sweetinga,b, Heidi Appelblomc, Glynis P Rossa,b, Jencia Wonga,b, Heikki Kouruc, Paul F Williamsa,b, Mikko Sairanenc, Jon A Hyettd.e
a
Royal Prince Alfred Hospital, Diabetes Centre, Sydney, Australia
b
Discipline of Medicine, University of Sydney, Sydney, Australia
c
Diagnostics, Perkin Elmer, Turku, Finland
d
Royal Prince Alfred Hospital, Department of High Risk Obstetrics, Sydney,
Australia e
Discipline of Obstetrics, Gynaecology and Neonatology, University of Sydney,
Sydney, Australia
Word Count: 2616 Figures and Tables: 5 (+ 1 supplementary table) Funding sources/Disclosures: The authors have nothing to disclose
Corresponding author: Arianne N Sweeting Royal Prince Alfred Hospital Diabetes Centre Missenden Road, Camperdown NSW AUSTRALIA 2050 Email: [email protected]
1
Abstract Aim Develop a first trimester risk prediction model for GDM based on maternal clinical characteristics in a large metropolitan multiethnic population and compare its performance to that of other recently published GDM prediction models and clinical risk scoring systems. Methods A retrospective case control study of 248 women who developed GDM and 732 controls who did not. Maternal clinical parameters were prospectively obtained at 11-13+6 weeks’ gestation. A predictive multivariate regression model for GDM was developed, evaluated using areas under the receiver-operating characteristic (AUC) curve. The performance of this model was then compared with other published GDM prediction models applied to our cohort and our existing clinical risk scoring system. Results Previous GDM, family history of diabetes, age, south/east Asian ethnicity, parity and body mass index (BMI) were significant predictors for GDM. The AUC of our multivariate regression model was 0.88 (95% Confidence Interval 0.85-0.92). This performed better than other predictive models applied to our cohort (AUCs 0.770.82). Conclusion A multivariate model based on weighted maternal clinical risk factors accurately predicts GDM in early pregnancy and performs better than other proposed multivariate and clinical risk scoring models in a multiethnic cohort.
2
1. Introduction Gestational diabetes mellitus (GDM) is an increasingly common complication of pregnancy associated with significant short- and long-term maternal and neonatal morbidity [1, 2]. GDM is typically diagnosed from 24 weeks’ gestation when interventions that reduce maternal hyperglycaemia have been shown to significantly improve short-term pregnancy outcomes [3, 4].
There is however a compelling argument for detecting GDM earlier in pregnancy, given increasing evidence that the prevailing approach may not mitigate the adverse long-term effects in the offspring of prolonged intrauterine exposure to maternal hyperglycaemia. Specifically, Sovio et al. (2016) recently demonstrated that accelerated fetal growth occurs as early as 20 weeks’ gestation, prior to the diagnosis of GDM [5]. Further, Logan et al. (2016) showed that increased adiposity in the offspring of women with GDM was apparent in early infancy, despite good maternal glycaemic control [6], supporting earlier detection and treatment of GDM. Importantly, there is now strong evidence that GDM and its attendant complications can be prevented by lifestyle interventions commenced in pre-/early pregnancy (<20 weeks’ gestation) in high-risk women [7-10].
Indeed, international guidelines now recommend screening in early pregnancy for women deemed to be at high risk of hyperglycaemia in pregnancy [11]. Although the 2 hour 75gram oral glucose tolerance test (OGTT) remains the gold standard for diagnosis of GDM in later pregnancy (i.e. from 24 weeks’ gestation), a comparable test for identifying GDM in early pregnancy is still to be determined [12, 13]. Historically, screening for GDM in early pregnancy has relied on the presence of one
3
or more established maternal clinical risk factors, but this binary approach is limited by its poor sensitivity and specificity [14, 15]. The use of clinical scoring systems based on these risk factors has been proposed to improve the early detection of GDM [16], but more recently it has been argued that screening efficacy can be further improved by combining maternal risk factors in a multivariate logistic regression model [17].
Our objective was to develop a risk prediction model for GDM at 11-13+6 weeks’ gestation using a combination of maternal clinical parameters in a large metropolitan multi-ethnic Australian population characterised by a high prevalence of GDM (~11%) [18, 19]. We also aimed to compare its performance with our existing clinical risk scoring system and other recently published multivariate prediction models for GDM [17, 20, 21].
2. Materials and Methods 2.1 Cohort This was a case control study of 248 women who developed GDM and 732 controls who did not (n=980). All women with a singleton pregnancy attending Royal Prince Alfred Hospital, Sydney at 11-13+6 weeks’ gestation for combined first trimester aneuploidy and pre-eclampsia screening between April 2011 and May 2013 were eligible for inclusion. Controls were randomised based on fetal crown rump length (CRL) (measurement of fetal CRL was undertaken at the time of first trimester ultrasound scan to confirm gestational age) to achieve a case:control ratio of 1:3. Women with pre-existing diabetes, pre-eclampsia, multiple pregnancies, pre-term delivery
(<37
weeks’
gestation),
miscarriage,
stillbirth,
termination,
fetal
4
chromosomal abnormality, missing clinical data and where GDM was diagnosed based on a glucose challenge test alone were excluded from analysis. Ethics approval was granted by the local hospital ethics committee (HREC/11/RPAH/472).
2.2 Maternal clinical parameters Maternal clinical risk factors were prospectively obtained at the time of combined first trimester screening between 11-13+6 weeks’ gestation and recorded in our fetal medicine database with the results of aneuploidy and pre-eclampsia screening. These included age, first trimester body mass index (BMI) (height and weight was formally measured by a midwife and BMI was then calculated in kg/m2), ethnicity and parity (classified as multiparous or nulliparous if a woman had no previous pregnancies ≥24 weeks’ gestation).
Women also completed a risk factor questionnaire at the time of first trimester screening which included family history of diabetes (defined as type 2 diabetes in a first or second degree relative and/or sibling with GDM), medical history including pre-existing hypertension and pre-eclampsia, previous history of GDM or macrosomia (defined as birth weight >4000 grams and/or >90th centile based on local population growth charts [22]), polycystic ovarian syndrome (PCOS), method of conception (spontaneous or assisted conception requiring use of ovulation drugs), medical and smoking history. Clinical risk factors for GDM were subsequently verified by review of individual electronic medical records. Pregnancy outcome data were collated from the electronic medical record and entered into the fetal medicine database as they became available.
5
2.3 Testing for GDM GDM was defined by the Australasian Diabetes in Pregnancy (ADIPS) diagnostic criteria with universal testing for GDM between 24 to 28 weeks’ gestation by either the diagnostic 2 hour 75gram oral glucose tolerance test (OGTT) (GDM diagnosed if fasting blood glucose level (BGL) ≥5.5 mmol/L and/or 1-hour BGL ≥10.5 mmol/L and/or 2-hour BGL ≥8.0 mmol/L); or a screening 50 gram glucose challenge test (GCT) and if positive (1-hour BGL ≥7.8 mmol/L), a subsequent OGTT [23]. Women at high risk for GDM (ie. non-Caucasian ethnicity, age ≥40 years, BMI ≥25 kg/m2, family history of diabetes, previous GDM or macrosomia, PCOS or concurrent medication associated with hyperglycaemia) [23] were advised to undergo early testing for GDM with an OGTT generally soon after the first antenatal appointment; repeated at 18-20 weeks’ and 24-28 weeks’ gestation if still negative. The GDM diagnostic criteria did not change during the study period.
2.4 Statistical analysis Statistical software package S+ v.8.1 (TIBCO Spotfire, Boston, USA) was used for the data analysis. Data are presented as medians, numbers (percentage) or means, as appropriate. Comparison between outcome groups was assessed by Kruskal-Wallis tests for continuous variables and Chi-square for categorical variables. Statistical significance was accepted at p<0.05.
Univariate analysis was undertaken to investigate the significance of individual maternal clinical parameters for GDM diagnosed at any time point during pregnancy. A multivariate logistic regression model was then developed for prediction of GDM; fitted with GDM status as the outcome and significant maternal clinical risk factors as
6
predictors (ie. probability of GDM=1/[1+exp(-b)], in which b is calculated as [Intercept
+
(estimate_1*variable_1)
+
(estimate_2*variable_2)
+
...+
(estimate_n*variable_n)]. The selection of variables (stepwise selection) was based on Akaike information criterion (AIC) [24]. The model was evaluated using areas under the receiver-operating characteristic (AUC) curve, and the calculated AUC for our model was compared to our current GDM clinical risk scoring system and to recently published models applied to our cohort [17, 20, 21] (Supplementary Table S1) to determine its relative predictive performance.
3. Results Maternal baseline demographic characteristics and pregnancy outcomes are summarized in Table 1. Compared to controls, women with GDM were older, shorter, predominantly of east or south Asian ethnicity, had a higher first trimester BMI and were more likely to have a family history of diabetes (all p<0.05). The prevalence of macrosomia and PCOS in the GDM cohort was 6.5% and 4.4%, respectively, lower than that reported in the literature [25, 26]; and there were no cases among controls. The majority of the cohort was nulliparous (51.6% and 58.3% for GDM and controls respectively). Multiparous women with GDM were more likely to have had previous GDM compared to controls. Conversely, there were a significantly higher proportion of multiparous controls who did not have previous GDM compared to women with GDM (all p<0.05).
Regarding pregnancy outcomes, women with GDM delivered a median of 4 days earlier and their offspring had a lower mean birthweight (3307±448 grams versus 3492±462 grams for controls, respectively; p<0.05). Almost 50% of women with
7
GDM were induced, compared to a fifth of controls (21.6%); while delivery by caesarean section was also more common among women with GDM at 28.6% versus 20.6%, respectively (all p<0.05).
Univariate logistic regression analysis is summarized in Table 2. Previous GDM (odds ratio (OR) 20.8, [95% Confidence Interval CI 10.0-43.3]) and family history of diabetes (OR 20.9, [95% CI 14.2–30.9]) were associated with the greatest risk of GDM. Multiparous women without a history of previous GDM were 43% less likely to develop GDM (OR 0.57, [95% CI 0.40-0.82]). Maternal age ≥35 years, east and south Asian ethnicity, maternal first trimester weight between 85-100 kg and/or BMI ≥30 kg/m2 were also associated with a significantly increased risk of GDM. Previous history of macrosomia and PCOS were excluded from univariate logistic regression analysis due to their absence in the control population. The logistic regression model (Table 3) showed that previous GDM and family history of diabetes, followed by south and east Asian ethnicity, were the strongest determinants for risk of GDM.
The performance of our GDM risk prediction model (variables included previous GDM, family history of diabetes, south/east Asian ethnicity, parity, maternal age and BMI (kg/m2)) in comparison to 3 recently published clinical risk prediction models and our current GDM clinical risk scoring system is shown in Fig. 1. The AUC of 0.88 [95% CI 0.85-0.92] for our model is higher than either the derived (ie. applied to our cohort) or original AUROCs for the other models. The estimated detection rate for GDM in our cohort was 80.9%, at a false positive rate (FPR) of 20%. The performance of the published clinical risk prediction models when applied to our cohort varied significantly (Table 4): van Leeuwen et al’s (2010) model [21]
8
performed worse when applied to our cohort compared to their original study cohort; Syngelaki et al’s (2014) model [17] performed better and Nanda et al’s (2011) model [20] performed similarly across both cohorts. Our existing RPAH clinical risk scoring model performed well in comparison to the multivariate models in our cohort.
4. Discussion Our study demonstrates that GDM can be accurately predicted in early pregnancy based on simple maternal clinical parameters available at the time of first trimester aneuploidy and pre-eclampsia screening between 11-13+6 weeks’ gestation. Furthermore, our multivariate GDM prediction model based on these weighted maternal risk factors demonstrates superior performance compared to other recently published models in our cohort and our existing clinical risk scoring system for GDM.
Our study also validates the currently recommended criteria for screening high risk women for GDM in early pregnancy [23]. In our large multi-ethnic cohort, the strongest predictors for GDM were previous GDM and family history of diabetes. The contribution of the latter was comparable to the risk associated with previous GDM, but was greater relative to the other studies. This may be accounted for by (1) the lower prevalence of family history of diabetes in our control cohort compared to that seen in other studies [17, 20, 21]; (2) the overall higher prevalence of GDM in our cohort; (3) the inclusion of both first and second degree relatives with diabetes, which may have over-estimated the relative contribution of family history and/or led to measurement (recall) bias; and (4) the contribution of systematic early screening for GDM based on these specific risk factors at our institution. Ethnicity, maternal age
9
and BMI also predicted GDM and were included in our model. There was a notable protective effect of multi-parity with no previous history of GDM.
In addition, our multivariate model performs better than other recently published multivariate models and our existing clinical risk scoring system. The AUC for our model was 0.88, such that screening for GDM identifies 70.2, 80.0 and 89.3% of cases at a false positive rate of 10, 20 and 40%, respectively. Differences in the performance of the models may be accounted for by differences in study cohort characteristics and size; prevalence, diagnostic and screening criteria for GDM; and whether risk factors were treated as continuous rather than categorical variables in the multivariate models [17]. The superior performance of multivariate models in comparison to clinical risk scoring systems has been shown previously [17]. In our cohort, we found that the performance of our clinical risk scoring system was comparable to that seen with the multivariate prediction models. Nevertheless, there are important expected clinical benefits associated with the observed incremental predictive performance of our multivariate model compared to our existing clinical risk scoring system for GDM (AUC 0.88 vs. 0.85), primarily relating to the significant improvement in detection rate. Specifically, the detection rate at a 10% fixed FPR was 70.2% for our multivariate model compared to 60.2% for our clinical risk scoring system; and such an improvement in early pregnancy screening accuracy for GDM would enable better risk stratification, targeting high risk women for preventive interventions or early diagnostic testing. Conversely, implementation of our multivariate model change may streamline subsequent clinical care pathways by better identifying low risk women who do not require further screening or diagnosis for GDM.
10
It is also important to note that the comparable performance of our clinical risk scoring system is likely accounted for by the fact that the threshold for ‘high risk’ for GDM in can be fulfilled by a single risk factor (eg. previous GDM), over-estimating the association of certain risk factors and leading to its better than expected performance. Further, previous PCOS and macrosomia are risk factors for our clinical risk scoring system; and these were absent in the control cohort. In comparison, the clinical risk scoring system published by Teede et al (2011) [16] demonstrated an AUC of only 0.70 [95% CI 0.68-0.73] in their cohort. Unfortunately, we were unable to determine the derived AUROC in our cohort for this clinical risk scoring system due to differences in ethnic classification in our dataset.
There are other potential limitations to our study. Our model’s performance may have been over-estimated, although comparison with other multivariate models within our cohort confirmed its superiority. Nevertheless, validation in a larger cohort utilizing a similar diagnostic strategy is required; this would also enable further evaluation of the predictive utility of previous PCOS and macrosomia which were not included in our final multivariate model. More specifically, while maternal demographic data was collected prospectively, a history of PCOS and macrosomia were not integral to assessing the risk for early onset pre-eclampsia (the original focus of the data set) and thus we must question the quality of this specific data given their absent prevalence among controls. Reassuringly however, only one of the previously published models included these factors (birthweight >90th centile in Nanda et al (2011) [20]), and their exclusion would most likely contribute to an under-estimate rather than over-estimate of our model’s performance.
In addition, measurement bias in self-reporting of
11
family history of diabetes may have over-estimated the accuracy of both our model and existing clinical risk scoring system, however our model’s performance was comparable to that of Syngelaki et al (2014) [17], which also included family history of diabetes as a risk factor.
Strengths of our study include our large multi-ethnic cohort with prospectively collected data obtained at the time of aneuploidy screening between 11-13+6 weeks’ gestation. Importantly, we diagnosed GDM via the 2 hour OGTT in accordance with international guidelines: in early pregnancy for high risk women and as part of universal screening at 24-28 weeks’ gestation [11, 12].
In conclusion, our model provides a highly effective means of screening for GDM in early pregnancy, using simple maternal clinical characteristics at the time of first trimester screening. This model allows the estimation of an individual’s specific a priori risk of GDM, thereby addressing the poor sensitivity and specificity characterizing traditional risk scoring approaches. This method of screening for GDM is not only easily applicable but can potentially be combined with early pregnancy screening for aneuploidy and pre-eclampsia for comprehensive, cost effective pregnancy risk assessment in the first trimester. Future studies should seek to prospectively validate these findings in other cohorts using the revised IADPSG diagnostic criteria for GDM [11] and potentially further improve the performance of the model with the addition of biomarkers implicated in the pathogenesis of GDM [20, 27]. The impact of early preventive and intervention strategies following positive early screening for GDM using these novel prediction models should also be explored.
12
Funding sources This research did not receive any specific grant funding from funding agencies in the public, commercial or not-for-profit sectors. Competing interests None Author contributions A.N.S. researched and compiled data, wrote manuscript. H.A analysed data. G.P.R. and J.W reviewed/edited manuscript. H.K analysed data, reviewed manuscript. P.W compiled data, reviewed/edited manuscript. M.S analysed data, reviewed/edited manuscript. J.H. researched data, contributed to discussion, reviewed/edited manuscript. A.N.S is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
13
References 1.
HAPO Study Group. Hyperglycaemia and adverse pregnancy outcomes. New Eng J Med. 2008;358:1991-2002.
2.
Barker D. Mothers, babies and diseases in later life. London: BMJ Publishing Group;1994
3.
Crowther CA, Hiller JE, Moss JR, et al. Effect of treatment of gestational diabetes mellitus on pregnancy outcomes. N Engl J Med. 2005;352(24):24772486.
4.
Landon MB, Spong CY, Thom E, et al. A multicenter, randomized trial of treatment for mild gestational diabetes. N Engl J Med. 2009;361(14):13391348.
5.
Sovio U, Murphy H, Smith GCS. Accelerated fetal growth prior to diagnosis of gestational diabetes mellitus: a prospective cohort study of nulliparous women. Diabetes Care, published online ahead of print April 9 2016.
6.
Logan KM, Emsley RJ, Jeffries S, et al. Development of early adiposity in infants of mothers with gestational diabetes mellitus. Diabetes Care, published online ahead of print May 12 2016.
7.
Zhang C, Tobias DK, Chavarro JE, et al. Adherence to healthy lifestyle and risk of gestational diabetes mellitus: prospective cohort study. BMJ. 2014;349:g5450.
8.
Thangaratinam S, Rogozinska E, Jolly K, et al. Effects of interventions in pregnancy on maternal weight and obstetric outcomes: meta-analysis of randomised evidence. BMJ. 2012;344:e2088.
14
9.
Oostdam N, van Poppel MN, Wolters MG, van Mechelen W. Interventions for preventing gestational diabetes mellitus: a systematic review and metaanalysis. J Womens Health (Larchmt). 2011;20(10):1551-1563.
10.
Koivusalo SB, Rono K, Klemetti MM, et al. Gestational Diabetes Mellitus Can Be Prevented by Lifestyle Intervention: The Finnish Gestational Diabetes Prevention Study (RADIEL): A Randomized Controlled Trial. Diabetes Care. 2016;39(1):24-30.
11.
International Association of Diabetes in Pregnancy Study Groups. Metzger BE, Gabbe SE, Persson B, et al. International association of diabetes and pregnancy study groups recommendations on the diagnosis and classification of hyperglycemia in pregnancy. Diabetes Care. 2010;33(3):676-682.
12.
World Health Organisation. Diagnostic criteria and classification of hyperglycaemia first detected in pregnancy: a World Health Organization Guideline. Diabetes Res Clin Pract. 2014;103(3):341-363.
13.
Vandorsten JP, Dodson WC, Espeland MA, et al. NIH consensus development conference: diagnosing gestational diabetes mellitus. NIH Consens State Sci Statements. 2013;29(1):1-31.
14.
Cosson E, Benchimon M, Carbillon L, et al. Universal rather than selective screening for gestational diabetes mellitus may improve fetal outcomes. Diabetes Metab. 2006; 32(2):140-146.
15.
Naylor CD, Sermer M, Chen E, et al. Selective screening for gestational diabetes mellitus. Toronto Trihospital Gestational Diabetes Project Investigators. N Engl J Med. 1997;337(22):1591-1596.
15
16.
Teede HJ, Harrison CL, Teh WT, et al. Gestational diabetes: development of an early risk prediction tool to facilitate opportunities for prevention. Aust N Z J Obstet Gynaecol. 2011;51(6):499-504.
17.
Syngelaki A, Pastides A, Kotecha R, et al. First-Trimester Screening for Gestational Diabetes Mellitus Based on Maternal Characteristics and History. Fetal Diagn Ther. 2015;38(1):14-21.
18.
Australian Institute of Health and Welfare. Diabetes in pregnancy: its impact on Australian women and their babies. Diabetes series no. 14. Cat. No. CVD 52. Canberra: AIHW; 2010.
19.
Flack JR, Ross GP, McElduff A. Recommended changes to diagnostic criteria for gestational diabetes: impact on workload. Aust N Z J Obstet Gynaecol. 2010;50(5): 439-443.
20.
Nanda S, Savviddou M, Syngelaki A, et al. Prediction of gestational diabetes mellitus by maternal factors and biomarkers at 11 to 13 weeks. Prenat Diagn. 2011;31(2): 135-141.
21.
van Leeuwen M, Opmeer BC, Zweers EJ, et al. Estimating the risk of gestational diabetes mellitus: a clinical prediction model based on patient characteristics and medical history. BJOG. 2010;117(1):69-75.
22.
Beeby PJ, Taylor LK. New South Wales population-based birthweight percentile chart. J Paediat Child Heal. 1996;32:512-518.
23.
Nankervis A, Moses R, Ross GP, Callaway L, Porter C et al. ADIPS Consensus Guidelines for the Testing and Diagnosis of Hyperglycaemia in Pregnancy in Australia and New Zealand. Updated November 2014. Available from: http://adips.org/downloads/2014ADIPSGDMGuidelinesV18.11.2014_000.pdf.
16
24.
Akaike H. Information theory and an extension of the maximum likelihood principle. Proceedings of the Second International Symposium on Information Theory, ed. Csaki. 1973, Budapest: Akademiai Kiado.
25.
Kousta E, Cela E, Lawrence N, et al. The prevalence of polycystic ovaries in women with a history of gestational diabetes. Clin Endocrinol (Oxf). 2000;53(4):501-507.
26.
Anttila L, Karjala K, Pentilla RA, et al. Polycystic ovaries in women with gestational diabetes. Obstet Gynecol. 1998;92(1):13-16.
27.
Theriault S, Giguere Y, Masse J, et al. Early prediction of gestational diabetes: a practical model combining clinical and biochemical markers. Clin Chem Lab Med. 2016;54(3):509-518.
17
Table 1 Maternal demographics and pregnancy outcomes in the screening population
Variables Maternal age (years) Maternal weight (kg) Maternal height (cm BMI (kg/m2) Ethnicity (%) Caucasian East Asian South Asian Other* Smoker (%) Conception (%) Spontaneous Ovulation induction In vitro fertilization Family history of diabetes** (%) Parity (%) Nulliparous Parous with previous GDM Parous with no previous GDM Previous GDM (%) Polycystic Ovarian Syndrome (%) Previous Macrosomia*** (%) Gestation at delivery (days) Birthweight (grams) Gender (%) Male Female Mode of Delivery (%) Normal**** Caesarean Section Induction
GDM (n=248) 33 (30-36) 64.4 (58.2-75.4) 162.0 (158.0-167.5) 24.5 (22.5-28.3)
Controls (n=732) 32 (29-35) 63.7 (57.4-71.7) 164.0 (159.5-169.0) 23.3 (21.6-26.1)
P value < 0.05 NS < 0.05 < 0.05 < 0.05
108 (43.5%) 91 (36.7%) 47 (19.0%) 2 (0.8%) 4 (1.6%)
530 (72.4%) 134 (18.3%) 54 (7.4%) 14 (1.9%) 21 (2.9%)
235 (94.8%) 0 (0%) 13 (5.2%) 138 (55.6%)
702 (95.9%) 2 (0.3%) 28 (3.8%) 51 (7.0%)
< 0.05
128 (51.6%) 52 (21.0%) 68 (27.4%) 59 (23.8%) 11 (4.4%) 16 (6.5%) 275 (271 - 280) 3307±448
427 (58.3%) 8 (1.1%) 297 (40.6%) 9 (1.2%) 0 (0%) 0 (0%) 279 (273 – 285) 3492±462
NS < 0.05 < 0.05 < 0.05 < 0.05 < 0.05 < 0.05 < 0.05
121 (48.8%) 127 (51.2%)
353 (48.2%) 379 (51.8%)
NS NS < 0.05
177 (71.4%) 71 (28.6%) 120 (48.4%)
581 (79.4%) 151 (20.6%) 158 (21.6%)
NS NS
Values are presented as medians (IQR), numbers (percentages) or means ± SD, as appropriate. *Middle Eastern, African, or combination of south/east Asian and Caucasian. **First or second degree relative with type 2 diabetes and/or sister with GDM. ***Birthweight >4000 grams and/or >90th centile based on local population growth charts [23]. ****Normal delivery: vaginal or assisted (ventouse, forceps). NS: Non-significant.
18
Table 2 Prevalence and logistic regression analysis for the prediction of GDM from maternal clinical characteristics
Variable Age (years) <30 30-34 ≥35 Ethnicity Caucasian East Asian South Asian Other* Weight (kg) <75 75-85 85-100 >100 BMI (kg/m2) <20 20-24.9 25-29.9 30-34.9 >35 Parity Nulliparous Parous (Hx GDM -) Parous (Hx GDM +) Unknown Family history of diabetes** No Yes Unknown
Number of women
Prevalence of GDM (%)
Odds ratio
LCI
UCI
P-value
254 322 404
21.3 24.2 28.7
Reference 1.18 1.49
0.80 1.03
1.76 2.16
0 0.2 0.02
638 225 101 16
16.6 40.4 46.5 12.5
Reference 3.33 4.27 0.70
2.38 2.74 0.16
4.67 6.65 3.12
0 <0.0001 <0.0001 0.03
776 101 66 37
23.6 28.7 34.8 35.1
Reference 1.31 1.73 1.76
0.82 1.02 0.88
2.07 2.95 3.52
0 0.13 0.02 0.06
84 523 255 74 44
20.2 22.6 25.9 41.9 36.4
0.87 Reference 1.20 2.47 1.96
0.49 0.85 1.49 1.03
1.54 1.70 4.10 3.75
0.6 <0.0001 0.31 0.0004 0.04
551 347 64 18
22.7 14.4 85.9 1.0
Reference 0.60 20.80 -
0.40 10.00 -
0.8 43.33 -
0 0.001 <0.0001 -
769 189 22
11.4 73.0 1.0
Reference 20.90 -
14.20 -
30.90 -
0 <0.0001 -
LCI: Lower (95%) Confidence Interval. UCI: Upper (95%) Confidence Interval. SE: Standard error. *Middle Eastern, African, or combination of south/east Asian and Caucasian. **First or second degree relative with type 2 diabetes and/or sister with GDM.
19
Table 3 Multivariate analysis for the clinical risk prediction model Odds Ratio
LCI
UCI
Estimate
SE
p-value
(Intercept)
-
-
-
-5.9515
1.0700
<0.0001
Previous GDM
17.5049
6.9317
44.2060
2.8624
0.4727
<0.0001
East Asian
4.7117
2.8731
7.7267
1.5500
0.2524
<0.0001
South Asian Family history of diabetes* Parity
5.4921
2.8718
10.5032
1.7033
0.3308
<0.0001
21.4935
13.5238
34.1598
3.0678
0.2364
<0.0001
0.5579
0.3513
0.8859
-0.5836
0.2359
0.007
Maternal age
1.0360
0.9836
1.0912
0.0354
0.0265
0.09
1.0883
1.0409
1.1378
0.08462
0.0227
<0.0001
2
BMI (kg/m )
LCI: Lower (95%) Confidence Interval. UCI: Upper (95%) Confidence Interval. SE: Standard error. *First or second degree relative with type 2 diabetes and/or sibling with GDM.
20
Figure 1 Sensitivity and specificity of the new model compared to other models for GDM prediction within our cohort
Green: van Leeuwen et al. (2010) [21]; Yellow: Nanda et al. (2011) [20]; Brown: Existing clinical risk scoring system; Orange: Syngelaki et al. (2014) [17]; Blue: New multivariate model.
21
Table 4 Comparison of the performance of the GDM prediction models in our cohort compared to the original cohort Original
Detection rate (%) at
AUC
FPR of:
Derived AUC Prediction Model [95% CI] [95% CI]
10%
20%
40%
70.2
80.9
89.3
63.6
77.7
89.5
46.0
59.3
79.4
29.0
41.3
58.5
60.2
73.4
86.2
0.88 New multivariate model [0.85-0.92] Syngelaki et al. (2015)
0.87
0.82
[17]
[0.84 – 0.90]
[0.82–0.83]
0.78
0.79
[0.75 – 0.82]
[0.76–0.82]
van Leeuwen et al.
0.64
0.77
(2010) [21]
[0.61 – 0.69]
[0.69–0.85]
Existing clinical risk
0.85
scoring model
[0.82-0.88]
Nanda et al. (2011) [20]
NA
22
AUC: Area under the receiver operating characteristic curve. 95% CI: Associated 95% Confidence Interval. FPR: False Positive Rate. Derived AUC: Applied to our cohort. Original AUC: Previously published AUC.
23
Highlights: • Recent studies show that lifestyle intervention commenced in early pregnancy can prevent the subsequent onset of gestational diabetes mellitus (GDM) • The best screening test for GDM in early pregnancy is however unknown • We report that a multivariate GDM prediction model based on maternal clinical risk factors can accurately predict GDM in early pregnancy • We also compare our model’s performance to that of other recently published GDM prediction models, demonstrating the superior performance of our model • We hope that our findings will improve early pregnancy screening programmes for GDM, facilitating early intervention and prevention strategies
24
S0168-8227(16)30525-3 http://dx.doi.org/10.1016/j.diabres.2017.02.036 DIAB 6896
To appear in:
Diabetes Research and Clinical Practice
Received Date: Revised Date: Accepted Date:
5 September 2016 19 February 2017 28 February 2017
Please cite this article as: A.N. Sweeting, H. Appelblom, G.P. Ross, J. Wong, H. Kouru, P.F. Williams, M. Sairanen, J.A. Hyett, First Trimester Prediction of Gestational Diabetes Mellitus: A Clinical Model Based on Maternal Demographic Parameters, Diabetes Research and Clinical Practice (2017), doi: http://dx.doi.org/10.1016/j.diabres. 2017.02.036
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
First Trimester Prediction of Gestational Diabetes Mellitus: A Clinical Model Based on Maternal Demographic Parameters
Arianne Natasha Sweetinga,b, Heidi Appelblomc, Glynis P Rossa,b, Jencia Wonga,b, Heikki Kouruc, Paul F Williamsa,b, Mikko Sairanenc, Jon A Hyettd.e
a
Royal Prince Alfred Hospital, Diabetes Centre, Sydney, Australia
b
Discipline of Medicine, University of Sydney, Sydney, Australia
c
Diagnostics, Perkin Elmer, Turku, Finland
d
Royal Prince Alfred Hospital, Department of High Risk Obstetrics, Sydney,
Australia e
Discipline of Obstetrics, Gynaecology and Neonatology, University of Sydney,
Sydney, Australia
Word Count: 2616 Figures and Tables: 5 (+ 1 supplementary table) Funding sources/Disclosures: The authors have nothing to disclose
Corresponding author: Arianne N Sweeting Royal Prince Alfred Hospital Diabetes Centre Missenden Road, Camperdown NSW AUSTRALIA 2050 Email: [email protected]
1
Abstract Aim Develop a first trimester risk prediction model for GDM based on maternal clinical characteristics in a large metropolitan multiethnic population and compare its performance to that of other recently published GDM prediction models and clinical risk scoring systems. Methods A retrospective case control study of 248 women who developed GDM and 732 controls who did not. Maternal clinical parameters were prospectively obtained at 11-13+6 weeks’ gestation. A predictive multivariate regression model for GDM was developed, evaluated using areas under the receiver-operating characteristic (AUC) curve. The performance of this model was then compared with other published GDM prediction models applied to our cohort and our existing clinical risk scoring system. Results Previous GDM, family history of diabetes, age, south/east Asian ethnicity, parity and body mass index (BMI) were significant predictors for GDM. The AUC of our multivariate regression model was 0.88 (95% Confidence Interval 0.85-0.92). This performed better than other predictive models applied to our cohort (AUCs 0.770.82). Conclusion A multivariate model based on weighted maternal clinical risk factors accurately predicts GDM in early pregnancy and performs better than other proposed multivariate and clinical risk scoring models in a multiethnic cohort.
2
1. Introduction Gestational diabetes mellitus (GDM) is an increasingly common complication of pregnancy associated with significant short- and long-term maternal and neonatal morbidity [1, 2]. GDM is typically diagnosed from 24 weeks’ gestation when interventions that reduce maternal hyperglycaemia have been shown to significantly improve short-term pregnancy outcomes [3, 4].
There is however a compelling argument for detecting GDM earlier in pregnancy, given increasing evidence that the prevailing approach may not mitigate the adverse long-term effects in the offspring of prolonged intrauterine exposure to maternal hyperglycaemia. Specifically, Sovio et al. (2016) recently demonstrated that accelerated fetal growth occurs as early as 20 weeks’ gestation, prior to the diagnosis of GDM [5]. Further, Logan et al. (2016) showed that increased adiposity in the offspring of women with GDM was apparent in early infancy, despite good maternal glycaemic control [6], supporting earlier detection and treatment of GDM. Importantly, there is now strong evidence that GDM and its attendant complications can be prevented by lifestyle interventions commenced in pre-/early pregnancy (<20 weeks’ gestation) in high-risk women [7-10].
Indeed, international guidelines now recommend screening in early pregnancy for women deemed to be at high risk of hyperglycaemia in pregnancy [11]. Although the 2 hour 75gram oral glucose tolerance test (OGTT) remains the gold standard for diagnosis of GDM in later pregnancy (i.e. from 24 weeks’ gestation), a comparable test for identifying GDM in early pregnancy is still to be determined [12, 13]. Historically, screening for GDM in early pregnancy has relied on the presence of one
3
or more established maternal clinical risk factors, but this binary approach is limited by its poor sensitivity and specificity [14, 15]. The use of clinical scoring systems based on these risk factors has been proposed to improve the early detection of GDM [16], but more recently it has been argued that screening efficacy can be further improved by combining maternal risk factors in a multivariate logistic regression model [17].
Our objective was to develop a risk prediction model for GDM at 11-13+6 weeks’ gestation using a combination of maternal clinical parameters in a large metropolitan multi-ethnic Australian population characterised by a high prevalence of GDM (~11%) [18, 19]. We also aimed to compare its performance with our existing clinical risk scoring system and other recently published multivariate prediction models for GDM [17, 20, 21].
2. Materials and Methods 2.1 Cohort This was a case control study of 248 women who developed GDM and 732 controls who did not (n=980). All women with a singleton pregnancy attending Royal Prince Alfred Hospital, Sydney at 11-13+6 weeks’ gestation for combined first trimester aneuploidy and pre-eclampsia screening between April 2011 and May 2013 were eligible for inclusion. Controls were randomised based on fetal crown rump length (CRL) (measurement of fetal CRL was undertaken at the time of first trimester ultrasound scan to confirm gestational age) to achieve a case:control ratio of 1:3. Women with pre-existing diabetes, pre-eclampsia, multiple pregnancies, pre-term delivery
(<37
weeks’
gestation),
miscarriage,
stillbirth,
termination,
fetal
4
chromosomal abnormality, missing clinical data and where GDM was diagnosed based on a glucose challenge test alone were excluded from analysis. Ethics approval was granted by the local hospital ethics committee (HREC/11/RPAH/472).
2.2 Maternal clinical parameters Maternal clinical risk factors were prospectively obtained at the time of combined first trimester screening between 11-13+6 weeks’ gestation and recorded in our fetal medicine database with the results of aneuploidy and pre-eclampsia screening. These included age, first trimester body mass index (BMI) (height and weight was formally measured by a midwife and BMI was then calculated in kg/m2), ethnicity and parity (classified as multiparous or nulliparous if a woman had no previous pregnancies ≥24 weeks’ gestation).
Women also completed a risk factor questionnaire at the time of first trimester screening which included family history of diabetes (defined as type 2 diabetes in a first or second degree relative and/or sibling with GDM), medical history including pre-existing hypertension and pre-eclampsia, previous history of GDM or macrosomia (defined as birth weight >4000 grams and/or >90th centile based on local population growth charts [22]), polycystic ovarian syndrome (PCOS), method of conception (spontaneous or assisted conception requiring use of ovulation drugs), medical and smoking history. Clinical risk factors for GDM were subsequently verified by review of individual electronic medical records. Pregnancy outcome data were collated from the electronic medical record and entered into the fetal medicine database as they became available.
5
2.3 Testing for GDM GDM was defined by the Australasian Diabetes in Pregnancy (ADIPS) diagnostic criteria with universal testing for GDM between 24 to 28 weeks’ gestation by either the diagnostic 2 hour 75gram oral glucose tolerance test (OGTT) (GDM diagnosed if fasting blood glucose level (BGL) ≥5.5 mmol/L and/or 1-hour BGL ≥10.5 mmol/L and/or 2-hour BGL ≥8.0 mmol/L); or a screening 50 gram glucose challenge test (GCT) and if positive (1-hour BGL ≥7.8 mmol/L), a subsequent OGTT [23]. Women at high risk for GDM (ie. non-Caucasian ethnicity, age ≥40 years, BMI ≥25 kg/m2, family history of diabetes, previous GDM or macrosomia, PCOS or concurrent medication associated with hyperglycaemia) [23] were advised to undergo early testing for GDM with an OGTT generally soon after the first antenatal appointment; repeated at 18-20 weeks’ and 24-28 weeks’ gestation if still negative. The GDM diagnostic criteria did not change during the study period.
2.4 Statistical analysis Statistical software package S+ v.8.1 (TIBCO Spotfire, Boston, USA) was used for the data analysis. Data are presented as medians, numbers (percentage) or means, as appropriate. Comparison between outcome groups was assessed by Kruskal-Wallis tests for continuous variables and Chi-square for categorical variables. Statistical significance was accepted at p<0.05.
Univariate analysis was undertaken to investigate the significance of individual maternal clinical parameters for GDM diagnosed at any time point during pregnancy. A multivariate logistic regression model was then developed for prediction of GDM; fitted with GDM status as the outcome and significant maternal clinical risk factors as
6
predictors (ie. probability of GDM=1/[1+exp(-b)], in which b is calculated as [Intercept
+
(estimate_1*variable_1)
+
(estimate_2*variable_2)
+
...+
(estimate_n*variable_n)]. The selection of variables (stepwise selection) was based on Akaike information criterion (AIC) [24]. The model was evaluated using areas under the receiver-operating characteristic (AUC) curve, and the calculated AUC for our model was compared to our current GDM clinical risk scoring system and to recently published models applied to our cohort [17, 20, 21] (Supplementary Table S1) to determine its relative predictive performance.
3. Results Maternal baseline demographic characteristics and pregnancy outcomes are summarized in Table 1. Compared to controls, women with GDM were older, shorter, predominantly of east or south Asian ethnicity, had a higher first trimester BMI and were more likely to have a family history of diabetes (all p<0.05). The prevalence of macrosomia and PCOS in the GDM cohort was 6.5% and 4.4%, respectively, lower than that reported in the literature [25, 26]; and there were no cases among controls. The majority of the cohort was nulliparous (51.6% and 58.3% for GDM and controls respectively). Multiparous women with GDM were more likely to have had previous GDM compared to controls. Conversely, there were a significantly higher proportion of multiparous controls who did not have previous GDM compared to women with GDM (all p<0.05).
Regarding pregnancy outcomes, women with GDM delivered a median of 4 days earlier and their offspring had a lower mean birthweight (3307±448 grams versus 3492±462 grams for controls, respectively; p<0.05). Almost 50% of women with
7
GDM were induced, compared to a fifth of controls (21.6%); while delivery by caesarean section was also more common among women with GDM at 28.6% versus 20.6%, respectively (all p<0.05).
Univariate logistic regression analysis is summarized in Table 2. Previous GDM (odds ratio (OR) 20.8, [95% Confidence Interval CI 10.0-43.3]) and family history of diabetes (OR 20.9, [95% CI 14.2–30.9]) were associated with the greatest risk of GDM. Multiparous women without a history of previous GDM were 43% less likely to develop GDM (OR 0.57, [95% CI 0.40-0.82]). Maternal age ≥35 years, east and south Asian ethnicity, maternal first trimester weight between 85-100 kg and/or BMI ≥30 kg/m2 were also associated with a significantly increased risk of GDM. Previous history of macrosomia and PCOS were excluded from univariate logistic regression analysis due to their absence in the control population. The logistic regression model (Table 3) showed that previous GDM and family history of diabetes, followed by south and east Asian ethnicity, were the strongest determinants for risk of GDM.
The performance of our GDM risk prediction model (variables included previous GDM, family history of diabetes, south/east Asian ethnicity, parity, maternal age and BMI (kg/m2)) in comparison to 3 recently published clinical risk prediction models and our current GDM clinical risk scoring system is shown in Fig. 1. The AUC of 0.88 [95% CI 0.85-0.92] for our model is higher than either the derived (ie. applied to our cohort) or original AUROCs for the other models. The estimated detection rate for GDM in our cohort was 80.9%, at a false positive rate (FPR) of 20%. The performance of the published clinical risk prediction models when applied to our cohort varied significantly (Table 4): van Leeuwen et al’s (2010) model [21]
8
performed worse when applied to our cohort compared to their original study cohort; Syngelaki et al’s (2014) model [17] performed better and Nanda et al’s (2011) model [20] performed similarly across both cohorts. Our existing RPAH clinical risk scoring model performed well in comparison to the multivariate models in our cohort.
4. Discussion Our study demonstrates that GDM can be accurately predicted in early pregnancy based on simple maternal clinical parameters available at the time of first trimester aneuploidy and pre-eclampsia screening between 11-13+6 weeks’ gestation. Furthermore, our multivariate GDM prediction model based on these weighted maternal risk factors demonstrates superior performance compared to other recently published models in our cohort and our existing clinical risk scoring system for GDM.
Our study also validates the currently recommended criteria for screening high risk women for GDM in early pregnancy [23]. In our large multi-ethnic cohort, the strongest predictors for GDM were previous GDM and family history of diabetes. The contribution of the latter was comparable to the risk associated with previous GDM, but was greater relative to the other studies. This may be accounted for by (1) the lower prevalence of family history of diabetes in our control cohort compared to that seen in other studies [17, 20, 21]; (2) the overall higher prevalence of GDM in our cohort; (3) the inclusion of both first and second degree relatives with diabetes, which may have over-estimated the relative contribution of family history and/or led to measurement (recall) bias; and (4) the contribution of systematic early screening for GDM based on these specific risk factors at our institution. Ethnicity, maternal age
9
and BMI also predicted GDM and were included in our model. There was a notable protective effect of multi-parity with no previous history of GDM.
In addition, our multivariate model performs better than other recently published multivariate models and our existing clinical risk scoring system. The AUC for our model was 0.88, such that screening for GDM identifies 70.2, 80.0 and 89.3% of cases at a false positive rate of 10, 20 and 40%, respectively. Differences in the performance of the models may be accounted for by differences in study cohort characteristics and size; prevalence, diagnostic and screening criteria for GDM; and whether risk factors were treated as continuous rather than categorical variables in the multivariate models [17]. The superior performance of multivariate models in comparison to clinical risk scoring systems has been shown previously [17]. In our cohort, we found that the performance of our clinical risk scoring system was comparable to that seen with the multivariate prediction models. Nevertheless, there are important expected clinical benefits associated with the observed incremental predictive performance of our multivariate model compared to our existing clinical risk scoring system for GDM (AUC 0.88 vs. 0.85), primarily relating to the significant improvement in detection rate. Specifically, the detection rate at a 10% fixed FPR was 70.2% for our multivariate model compared to 60.2% for our clinical risk scoring system; and such an improvement in early pregnancy screening accuracy for GDM would enable better risk stratification, targeting high risk women for preventive interventions or early diagnostic testing. Conversely, implementation of our multivariate model change may streamline subsequent clinical care pathways by better identifying low risk women who do not require further screening or diagnosis for GDM.
10
It is also important to note that the comparable performance of our clinical risk scoring system is likely accounted for by the fact that the threshold for ‘high risk’ for GDM in can be fulfilled by a single risk factor (eg. previous GDM), over-estimating the association of certain risk factors and leading to its better than expected performance. Further, previous PCOS and macrosomia are risk factors for our clinical risk scoring system; and these were absent in the control cohort. In comparison, the clinical risk scoring system published by Teede et al (2011) [16] demonstrated an AUC of only 0.70 [95% CI 0.68-0.73] in their cohort. Unfortunately, we were unable to determine the derived AUROC in our cohort for this clinical risk scoring system due to differences in ethnic classification in our dataset.
There are other potential limitations to our study. Our model’s performance may have been over-estimated, although comparison with other multivariate models within our cohort confirmed its superiority. Nevertheless, validation in a larger cohort utilizing a similar diagnostic strategy is required; this would also enable further evaluation of the predictive utility of previous PCOS and macrosomia which were not included in our final multivariate model. More specifically, while maternal demographic data was collected prospectively, a history of PCOS and macrosomia were not integral to assessing the risk for early onset pre-eclampsia (the original focus of the data set) and thus we must question the quality of this specific data given their absent prevalence among controls. Reassuringly however, only one of the previously published models included these factors (birthweight >90th centile in Nanda et al (2011) [20]), and their exclusion would most likely contribute to an under-estimate rather than over-estimate of our model’s performance.
In addition, measurement bias in self-reporting of
11
family history of diabetes may have over-estimated the accuracy of both our model and existing clinical risk scoring system, however our model’s performance was comparable to that of Syngelaki et al (2014) [17], which also included family history of diabetes as a risk factor.
Strengths of our study include our large multi-ethnic cohort with prospectively collected data obtained at the time of aneuploidy screening between 11-13+6 weeks’ gestation. Importantly, we diagnosed GDM via the 2 hour OGTT in accordance with international guidelines: in early pregnancy for high risk women and as part of universal screening at 24-28 weeks’ gestation [11, 12].
In conclusion, our model provides a highly effective means of screening for GDM in early pregnancy, using simple maternal clinical characteristics at the time of first trimester screening. This model allows the estimation of an individual’s specific a priori risk of GDM, thereby addressing the poor sensitivity and specificity characterizing traditional risk scoring approaches. This method of screening for GDM is not only easily applicable but can potentially be combined with early pregnancy screening for aneuploidy and pre-eclampsia for comprehensive, cost effective pregnancy risk assessment in the first trimester. Future studies should seek to prospectively validate these findings in other cohorts using the revised IADPSG diagnostic criteria for GDM [11] and potentially further improve the performance of the model with the addition of biomarkers implicated in the pathogenesis of GDM [20, 27]. The impact of early preventive and intervention strategies following positive early screening for GDM using these novel prediction models should also be explored.
12
Funding sources This research did not receive any specific grant funding from funding agencies in the public, commercial or not-for-profit sectors. Competing interests None Author contributions A.N.S. researched and compiled data, wrote manuscript. H.A analysed data. G.P.R. and J.W reviewed/edited manuscript. H.K analysed data, reviewed manuscript. P.W compiled data, reviewed/edited manuscript. M.S analysed data, reviewed/edited manuscript. J.H. researched data, contributed to discussion, reviewed/edited manuscript. A.N.S is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
13
References 1.
HAPO Study Group. Hyperglycaemia and adverse pregnancy outcomes. New Eng J Med. 2008;358:1991-2002.
2.
Barker D. Mothers, babies and diseases in later life. London: BMJ Publishing Group;1994
3.
Crowther CA, Hiller JE, Moss JR, et al. Effect of treatment of gestational diabetes mellitus on pregnancy outcomes. N Engl J Med. 2005;352(24):24772486.
4.
Landon MB, Spong CY, Thom E, et al. A multicenter, randomized trial of treatment for mild gestational diabetes. N Engl J Med. 2009;361(14):13391348.
5.
Sovio U, Murphy H, Smith GCS. Accelerated fetal growth prior to diagnosis of gestational diabetes mellitus: a prospective cohort study of nulliparous women. Diabetes Care, published online ahead of print April 9 2016.
6.
Logan KM, Emsley RJ, Jeffries S, et al. Development of early adiposity in infants of mothers with gestational diabetes mellitus. Diabetes Care, published online ahead of print May 12 2016.
7.
Zhang C, Tobias DK, Chavarro JE, et al. Adherence to healthy lifestyle and risk of gestational diabetes mellitus: prospective cohort study. BMJ. 2014;349:g5450.
8.
Thangaratinam S, Rogozinska E, Jolly K, et al. Effects of interventions in pregnancy on maternal weight and obstetric outcomes: meta-analysis of randomised evidence. BMJ. 2012;344:e2088.
14
9.
Oostdam N, van Poppel MN, Wolters MG, van Mechelen W. Interventions for preventing gestational diabetes mellitus: a systematic review and metaanalysis. J Womens Health (Larchmt). 2011;20(10):1551-1563.
10.
Koivusalo SB, Rono K, Klemetti MM, et al. Gestational Diabetes Mellitus Can Be Prevented by Lifestyle Intervention: The Finnish Gestational Diabetes Prevention Study (RADIEL): A Randomized Controlled Trial. Diabetes Care. 2016;39(1):24-30.
11.
International Association of Diabetes in Pregnancy Study Groups. Metzger BE, Gabbe SE, Persson B, et al. International association of diabetes and pregnancy study groups recommendations on the diagnosis and classification of hyperglycemia in pregnancy. Diabetes Care. 2010;33(3):676-682.
12.
World Health Organisation. Diagnostic criteria and classification of hyperglycaemia first detected in pregnancy: a World Health Organization Guideline. Diabetes Res Clin Pract. 2014;103(3):341-363.
13.
Vandorsten JP, Dodson WC, Espeland MA, et al. NIH consensus development conference: diagnosing gestational diabetes mellitus. NIH Consens State Sci Statements. 2013;29(1):1-31.
14.
Cosson E, Benchimon M, Carbillon L, et al. Universal rather than selective screening for gestational diabetes mellitus may improve fetal outcomes. Diabetes Metab. 2006; 32(2):140-146.
15.
Naylor CD, Sermer M, Chen E, et al. Selective screening for gestational diabetes mellitus. Toronto Trihospital Gestational Diabetes Project Investigators. N Engl J Med. 1997;337(22):1591-1596.
15
16.
Teede HJ, Harrison CL, Teh WT, et al. Gestational diabetes: development of an early risk prediction tool to facilitate opportunities for prevention. Aust N Z J Obstet Gynaecol. 2011;51(6):499-504.
17.
Syngelaki A, Pastides A, Kotecha R, et al. First-Trimester Screening for Gestational Diabetes Mellitus Based on Maternal Characteristics and History. Fetal Diagn Ther. 2015;38(1):14-21.
18.
Australian Institute of Health and Welfare. Diabetes in pregnancy: its impact on Australian women and their babies. Diabetes series no. 14. Cat. No. CVD 52. Canberra: AIHW; 2010.
19.
Flack JR, Ross GP, McElduff A. Recommended changes to diagnostic criteria for gestational diabetes: impact on workload. Aust N Z J Obstet Gynaecol. 2010;50(5): 439-443.
20.
Nanda S, Savviddou M, Syngelaki A, et al. Prediction of gestational diabetes mellitus by maternal factors and biomarkers at 11 to 13 weeks. Prenat Diagn. 2011;31(2): 135-141.
21.
van Leeuwen M, Opmeer BC, Zweers EJ, et al. Estimating the risk of gestational diabetes mellitus: a clinical prediction model based on patient characteristics and medical history. BJOG. 2010;117(1):69-75.
22.
Beeby PJ, Taylor LK. New South Wales population-based birthweight percentile chart. J Paediat Child Heal. 1996;32:512-518.
23.
Nankervis A, Moses R, Ross GP, Callaway L, Porter C et al. ADIPS Consensus Guidelines for the Testing and Diagnosis of Hyperglycaemia in Pregnancy in Australia and New Zealand. Updated November 2014. Available from: http://adips.org/downloads/2014ADIPSGDMGuidelinesV18.11.2014_000.pdf.
16
24.
Akaike H. Information theory and an extension of the maximum likelihood principle. Proceedings of the Second International Symposium on Information Theory, ed. Csaki. 1973, Budapest: Akademiai Kiado.
25.
Kousta E, Cela E, Lawrence N, et al. The prevalence of polycystic ovaries in women with a history of gestational diabetes. Clin Endocrinol (Oxf). 2000;53(4):501-507.
26.
Anttila L, Karjala K, Pentilla RA, et al. Polycystic ovaries in women with gestational diabetes. Obstet Gynecol. 1998;92(1):13-16.
27.
Theriault S, Giguere Y, Masse J, et al. Early prediction of gestational diabetes: a practical model combining clinical and biochemical markers. Clin Chem Lab Med. 2016;54(3):509-518.
17
Table 1 Maternal demographics and pregnancy outcomes in the screening population
Variables Maternal age (years) Maternal weight (kg) Maternal height (cm BMI (kg/m2) Ethnicity (%) Caucasian East Asian South Asian Other* Smoker (%) Conception (%) Spontaneous Ovulation induction In vitro fertilization Family history of diabetes** (%) Parity (%) Nulliparous Parous with previous GDM Parous with no previous GDM Previous GDM (%) Polycystic Ovarian Syndrome (%) Previous Macrosomia*** (%) Gestation at delivery (days) Birthweight (grams) Gender (%) Male Female Mode of Delivery (%) Normal**** Caesarean Section Induction
GDM (n=248) 33 (30-36) 64.4 (58.2-75.4) 162.0 (158.0-167.5) 24.5 (22.5-28.3)
Controls (n=732) 32 (29-35) 63.7 (57.4-71.7) 164.0 (159.5-169.0) 23.3 (21.6-26.1)
P value < 0.05 NS < 0.05 < 0.05 < 0.05
108 (43.5%) 91 (36.7%) 47 (19.0%) 2 (0.8%) 4 (1.6%)
530 (72.4%) 134 (18.3%) 54 (7.4%) 14 (1.9%) 21 (2.9%)
235 (94.8%) 0 (0%) 13 (5.2%) 138 (55.6%)
702 (95.9%) 2 (0.3%) 28 (3.8%) 51 (7.0%)
< 0.05
128 (51.6%) 52 (21.0%) 68 (27.4%) 59 (23.8%) 11 (4.4%) 16 (6.5%) 275 (271 - 280) 3307±448
427 (58.3%) 8 (1.1%) 297 (40.6%) 9 (1.2%) 0 (0%) 0 (0%) 279 (273 – 285) 3492±462
NS < 0.05 < 0.05 < 0.05 < 0.05 < 0.05 < 0.05 < 0.05
121 (48.8%) 127 (51.2%)
353 (48.2%) 379 (51.8%)
NS NS < 0.05
177 (71.4%) 71 (28.6%) 120 (48.4%)
581 (79.4%) 151 (20.6%) 158 (21.6%)
NS NS
Values are presented as medians (IQR), numbers (percentages) or means ± SD, as appropriate. *Middle Eastern, African, or combination of south/east Asian and Caucasian. **First or second degree relative with type 2 diabetes and/or sister with GDM. ***Birthweight >4000 grams and/or >90th centile based on local population growth charts [23]. ****Normal delivery: vaginal or assisted (ventouse, forceps). NS: Non-significant.
18
Table 2 Prevalence and logistic regression analysis for the prediction of GDM from maternal clinical characteristics
Variable Age (years) <30 30-34 ≥35 Ethnicity Caucasian East Asian South Asian Other* Weight (kg) <75 75-85 85-100 >100 BMI (kg/m2) <20 20-24.9 25-29.9 30-34.9 >35 Parity Nulliparous Parous (Hx GDM -) Parous (Hx GDM +) Unknown Family history of diabetes** No Yes Unknown
Number of women
Prevalence of GDM (%)
Odds ratio
LCI
UCI
P-value
254 322 404
21.3 24.2 28.7
Reference 1.18 1.49
0.80 1.03
1.76 2.16
0 0.2 0.02
638 225 101 16
16.6 40.4 46.5 12.5
Reference 3.33 4.27 0.70
2.38 2.74 0.16
4.67 6.65 3.12
0 <0.0001 <0.0001 0.03
776 101 66 37
23.6 28.7 34.8 35.1
Reference 1.31 1.73 1.76
0.82 1.02 0.88
2.07 2.95 3.52
0 0.13 0.02 0.06
84 523 255 74 44
20.2 22.6 25.9 41.9 36.4
0.87 Reference 1.20 2.47 1.96
0.49 0.85 1.49 1.03
1.54 1.70 4.10 3.75
0.6 <0.0001 0.31 0.0004 0.04
551 347 64 18
22.7 14.4 85.9 1.0
Reference 0.60 20.80 -
0.40 10.00 -
0.8 43.33 -
0 0.001 <0.0001 -
769 189 22
11.4 73.0 1.0
Reference 20.90 -
14.20 -
30.90 -
0 <0.0001 -
LCI: Lower (95%) Confidence Interval. UCI: Upper (95%) Confidence Interval. SE: Standard error. *Middle Eastern, African, or combination of south/east Asian and Caucasian. **First or second degree relative with type 2 diabetes and/or sister with GDM.
19
Table 3 Multivariate analysis for the clinical risk prediction model Odds Ratio
LCI
UCI
Estimate
SE
p-value
(Intercept)
-
-
-
-5.9515
1.0700
<0.0001
Previous GDM
17.5049
6.9317
44.2060
2.8624
0.4727
<0.0001
East Asian
4.7117
2.8731
7.7267
1.5500
0.2524
<0.0001
South Asian Family history of diabetes* Parity
5.4921
2.8718
10.5032
1.7033
0.3308
<0.0001
21.4935
13.5238
34.1598
3.0678
0.2364
<0.0001
0.5579
0.3513
0.8859
-0.5836
0.2359
0.007
Maternal age
1.0360
0.9836
1.0912
0.0354
0.0265
0.09
1.0883
1.0409
1.1378
0.08462
0.0227
<0.0001
2
BMI (kg/m )
LCI: Lower (95%) Confidence Interval. UCI: Upper (95%) Confidence Interval. SE: Standard error. *First or second degree relative with type 2 diabetes and/or sibling with GDM.
20
Figure 1 Sensitivity and specificity of the new model compared to other models for GDM prediction within our cohort
Green: van Leeuwen et al. (2010) [21]; Yellow: Nanda et al. (2011) [20]; Brown: Existing clinical risk scoring system; Orange: Syngelaki et al. (2014) [17]; Blue: New multivariate model.
21
Table 4 Comparison of the performance of the GDM prediction models in our cohort compared to the original cohort Original
Detection rate (%) at
AUC
FPR of:
Derived AUC Prediction Model [95% CI] [95% CI]
10%
20%
40%
70.2
80.9
89.3
63.6
77.7
89.5
46.0
59.3
79.4
29.0
41.3
58.5
60.2
73.4
86.2
0.88 New multivariate model [0.85-0.92] Syngelaki et al. (2015)
0.87
0.82
[17]
[0.84 – 0.90]
[0.82–0.83]
0.78
0.79
[0.75 – 0.82]
[0.76–0.82]
van Leeuwen et al.
0.64
0.77
(2010) [21]
[0.61 – 0.69]
[0.69–0.85]
Existing clinical risk
0.85
scoring model
[0.82-0.88]
Nanda et al. (2011) [20]
NA
22
AUC: Area under the receiver operating characteristic curve. 95% CI: Associated 95% Confidence Interval. FPR: False Positive Rate. Derived AUC: Applied to our cohort. Original AUC: Previously published AUC.
23
Highlights: • Recent studies show that lifestyle intervention commenced in early pregnancy can prevent the subsequent onset of gestational diabetes mellitus (GDM) • The best screening test for GDM in early pregnancy is however unknown • We report that a multivariate GDM prediction model based on maternal clinical risk factors can accurately predict GDM in early pregnancy • We also compare our model’s performance to that of other recently published GDM prediction models, demonstrating the superior performance of our model • We hope that our findings will improve early pregnancy screening programmes for GDM, facilitating early intervention and prevention strategies
24