bmj...13 care, university medical centre utrecht, str. 6.131, po box 85500, utrecht, the netherlands...
TRANSCRIPT
Confidential: For Review O
nly
External validation of prognostic models to predict the risk
of developing gestational diabetes
Journal: BMJ
Manuscript ID BMJ.2016.033135
Article Type: Research
BMJ Journal: BMJ
Date Submitted by the Author: 23-Apr-2016
Complete List of Authors: Lamain-de Ruiter, Marije; Universitair Medisch Centrum Utrecht, Woman and Baby Kwee, Anneke; Wilhelmina Children's Hospital, University Medical Centre, Utrecht, Department of Obstetrics Naaktgeboren, Christiana; University Medical Center Utrecht, Julius Center
for Health Sciences and Primary Care Franx, Arie; University Medical Center Utrecht, Obstetrics and gynecology Moons, Karel; Julius Center for Health Sciences and Primary Care, Epidemiology Koster, Maria; University Medical Center Utrecht, Department of Obstetrics; Erasmus MC de Groot, Inge; Tweesteden Ziekenhuis Vestiging Sint Elisabeth, Livive, Centre for Obstetrics Evers, Inge; Meander Medisch Centrum, Obstetrics Groenendaal, floris; University Medical Centre Utrecht, Department of Neonatology Hering, Yolanda; Zuwe Hofpoort Ziekenhuis, Obstetrics
Huisjes, Anjoke; Gelre Hospitals, Perinatology Kirpestein, Cornel; Ziekenhuis Rivierenland, Obstetrics Monincx, Wilma; Sint Antonius Ziekenhuis, Obstetrics Siljee, Jacqueline; Rijksinstituut voor Volksgezondheid en Milieu, Centre for Infectious Diseases Research, Diagnostics and screening (IDS) Van 't Zelfde, Annewil; Midwifery practice \'Verloskundigen Amersfoort Van Oirschot, Charlotte; Tweesteden Ziekenhuis Vestiging Sint Elisabeth, Obstetrics Vankan-Buitelaar, Simone; Midwifery practice 'GCM' Vonk, Mariska; Midwifery practice 'Het Wonder' Wiegers, Trees; Nederlands Instituut voor Onderzoek van de
Gezondheidszorg Zwart, Joost; KNOV, guidelines
Keywords: first trimester, gestational diabetes, external validation, prognostic model, head to head comparison
https://mc.manuscriptcentral.com/bmj
BMJ
Confidential: For Review O
nly
Page 1 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 1
TITLE PAGE 1
External validation of prognostic models to predict the risk of developing gestational 2
diabetes 3
Lamain-de Ruiter M1, Kwee A
2, Naaktgeboren CA
3, Franx A
4, Moons KGM
5, Koster MPH
6, on 4
behalf of the RESPECT study group 5
23 April 2016 6
Affiliations 7
1 PhD student, Department of Obstetrics, Division Woman and Baby, University Medical 8
Centre Utrecht, KE.04.123.1, PO BOX 85090, 3508 AB, Utrecht, The Netherlands 9
2 Obstetrician, Department of Obstetrics, Division Woman and Baby, University Medical 10
Centre Utrecht, KE.04.123.1, PO BOX 85090, 3508 AB, Utrecht, The Netherlands 11
3 Assistant professor of Clinical Epidemiology, Julius Centre for Health Sciences and Primary 12
Care, University Medical Centre Utrecht, Str. 6.131, PO BOX 85500, Utrecht, The Netherlands 13
4 Professor of Obstetrics, Department of Obstetrics, Division Woman and Baby, University 14
Medical Centre Utrecht, Utrecht, KE.04.123.1, PO BOX 85090, The Netherlands 15
5 Professor of Clinical Epidemiology, Julius Centre for Health Sciences and Primary Care, 16
University Medical Centre Utrecht, Str. 6.131, PO BOX 85500, Utrecht, The Netherlands 17
6 Assistant professor of Obstetrics & Gynaecology, Department of Obstetrics, Division 18
Woman and Baby, University Medical Centre Utrecht, KE.04.123.1, PO BOX 85090, 3508 AB, 19
Utrecht, The Netherlands & Department of Obstetrics and Gynaecology, Erasmus Medical 20
Centre, PO BOX 2040, 3000CA, Rotterdam, The Netherlands 21
22
23
Page 2 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 2
Collaborators in the RESPECT study group 24
de Groot I7, Evers IM
8, Groenendaal F
9, Hering Y
10, Huisjes AJM
11, Kirpestein C
12, Monincx 25
WM13
, Siljee JE14
, Van ’t Zelfde A15
, Van Oirschot CM16
, Vankan-Buitelaar S17
, Vonk MAAW18
, 26
Wiegers TA19
, Zwart JJ20
27
28
Affiliations 29
7 Midwife, ‘Livive’, Centre for Obstetrics, Tilburg, The Netherlands
30
8 Obstetrician, Department of Obstetrics, Meander Medical Centre, Amersfoort, The 31
Netherlands 32
9 Neonatologist, Department of Neonatology, Division Woman and Baby, University Medical 33
Centre Utrecht, Utrecht, The Netherlands 34
10 Midwife, Department of Obstetrics, Zuwe Hofpoort Hospital, Woerden, The Netherlands 35
11 Obstetrician, Department of Obstetrics, Gelre Hospital, Apeldoorn, The Netherlands 36
12 Midwife, Department of Obstetrics, Hospital Rivierenland, Tiel, The Netherlands 37
13 Obstetrician, Department of Obstetrics, St. Antonius Hospital, Nieuwegein, The 38
Netherlands
39
14 Senior researcher, Centre for Infectious Diseases Research, Diagnostics and Screening 40
(IDS), National Institute for Public Health and the Environment (RIVM), Bilthoven, The 41
Netherlands 42
15 Midwife, Midwifery practice ‘Verloskundigen Amersfoort’, Amersfoort, The Netherlands 43
16 Obstetrician, Department of Obstetrics, St Elisabeth Hospital, Tilburg, The Netherlands 44
17 Midwife, Midwifery practice ‘GCM’, Maarssenbroek, The Netherlands 45
18 Midwife, Midwifery practice ‘Het Wonder’, Houten, The Netherlands 46
Page 3 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 3
19 Senior researcher, Netherlands Institute for health services research (NIVEL), Utrecht, The 47
Netherlands 48
20 Obstetrician, Department of Obstetrics, Deventer Hospital, Deventer, The Netherlands 49
50
Address for correspondence 51
Name: Marije Lamain-de Ruiter 52
Phone number: +31-6-55 23 46 52 (work) 53
E-mail address: [email protected] 54
55
Grants 56
This study has been conducted with the support of The Netherlands Organization for Health 57
Research and Development (project nr 50-50200-98-060). 58
59
Word count 60
Abstract: 267 61
Main text: 3257 62
Page 4 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 4
Abstract 63
Objectives 64
To perform an external validation and direct comparison of published prognostic models for 65
early prediction of the risk of developing gestational diabetes (GDM), including predictors 66
applicable in the first trimester of pregnancy. 67
Design 68
External validation of all published prognostic models in a large scale prospective cohort 69
study. 70
Setting 71
31 independent midwifery practices and 6 hospitals in the Netherlands. 72
Participants 73
Women were included in the first trimester of pregnancy, <14 weeks of gestational age. 74
Women with pre-existing diabetes mellitus of any type were excluded. 3,723 women were 75
included for analysis of which 181 (4.9%) developed GDM in pregnancy. 76
Main outcome measures 77
Discrimination of the prognostic models was assessed by the C statistic, and calibration by 78
calibration plots. Findings were reported conform the TRIPOD statement for validation of 79
prognostic models. 80
Results 81
A systematic literature review identified 14 published prognostic models for GDM of which 82
12 models could be validated in our cohort. The C statistic ranged from 0.67 to 0.78. 83
Calibration plots showed that most models were well calibrated. The four models with the 84
highest C statistic (all >0.75) often included maternal age, maternal body mass index, history 85
Page 5 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 5
of GDM, ethnicity, and family history of diabetes as predictors. Prognostic models had a 86
similar performance in a subgroup of nulliparous women. 87
Conclusions 88
In this external validation study, most of the published prognostic models for GDM show 89
acceptable discrimination and calibration. The four models with the highest discriminative 90
abilities in our population, and which also perform well in a subgroup of nulliparous women, 91
are easy models to apply in clinical practice and therefore deserve further evaluation 92
regarding its clinical impact. 93
94
What is already known on this subject 95
- Gestational diabetes is an increasingly common complication of pregnancy and pregnancy outcome 96
can be improved through early screening, diagnosis and treatment. 97
- Numerous prognostic models for estimating the risk of developing gestational diabetes have been 98
developed, but an external validation and direct comparison in an independent large cohort of all 99
published models is lacking. 100
What this study adds 101
- This external validation study shows that in a direct comparison most published first trimester 102
prognostic models for GDM have an acceptable discrimination and good calibration. 103
- Well-performing first trimester prognostic models for GDM may be considered for implementation 104
in routine clinical care. 105
- Findings are reported conform the TRIPOD statement for validation of prognostic models. 106
107
Page 6 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 6
Introduction 108
In the field of obstetrics, the number of publications on prognostic models has more than 109
tripled in the last decade1, which reflects an increasing interest in risk-based medicine. Risk-110
based medicine aims to provide the most appropriate care to each patient, often guided by 111
outcome risk estimates based on individual patient characteristics, test results or even 112
genetic information.2 113
As a result of the obesity pandemic, the incidence of gestational diabetes (GDM), notably 114
occurring in the second or third trimester, is rising and is increasingly contributing to 115
perinatal complications, such as macrosomia, shoulder dystocia, caesarean section, and 116
neonatal hypoglycaemia.3,4
Moreover, long term sequelae of GDM are type II diabetes in the 117
mothers and obesity in their offspring.5,6
Diagnosis and treatment of GDM have been proven 118
to improve pregnancy outcomes.7,8
Some guidelines propose a population strategy for 119
diagnosing GDM (i.e. an oral glucose tolerance test (OGTT)) in each pregnant woman, where 120
others opt for a high risk strategy, an approach in which testing for GDM is only performed 121
in women with known risk factors. Both strategies include oGTTs in substantial numbers of 122
women, most of which will lead to negative results, and therefore pose a too high burden to 123
patients as well as health care resources.9 Accurate prognostic models for the risk of 124
developing GDM early in pregnancy may allow one to discriminate the high-risk from the 125
low-risk pregnancies, and to move towards more tailored care in pregnancy. In particular, 126
this may result in fewer women undergoing a burdensome diagnostic test. 127
Several prognostic models for gestational diabetes have been developed. However, these 128
prognostic models are not commonly used in routine clinical care nor are they 129
recommended by current guidelines. This may be due to the fact that external validation of 130
Page 7 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 7
these prognostic models are scarce,10–13
let alone that all these models have been directly 131
evaluated and compared based on their predictive accuracy in one single and independent 132
cohort by independent investigators. To acquire a fair comparison of their predictive 133
accuracy, and thus of their clinical value, it is essential to perform a head-to-head 134
comparison of all published prognostic models in one independent cohort.14–16
Thus, the aim 135
of our study was to perform the very first external validation and direct head-to-head 136
comparison of all published first trimester prognostic models for GDM, in a single 137
independent cohort. 138
Page 8 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 8
Methods 139
Prognostic models identified by systematic literature review 140
We performed a systematic review in which we identified 14 published prognostic models 141
for GDM that are applicable in the first trimester of pregnancy and that only consist of 142
routine and easy to obtain measures [Lamain-de Ruiter M, Kwee A, Naaktgeboren CA, Franx 143
A, Moons KGM, Koster MPH. Prediction models for the risk of gestational diabetes: a 144
systematic review]. For proper external validation it is favourable that the exact definitions 145
of the predictors included in the developed prognostic model are known as well as how they 146
were measured. Although five of the author groups of the publications of these prognostic 147
models were contacted by email for additional information on intercepts, coefficients, and 148
definitions of predictors in the model, none of them responded. Despite this lack of 149
information, it was still possible to include four of these models in our head-to-head 150
validation study, but one model had to be excluded due to missing intercept and 151
coefficients.17
Moreover, one prognostic model was excluded from analysis for utilizing 152
maternal abdominal circumference and diagnosis of polycystic ovary syndrome (PCOS) as 153
predictors, which we did not collect in our validation cohort and for which was also no proxy 154
variable available.18
Thus, a total of 12 prognostic models remained for external validation in 155
the current study.19–28
In Appendix A the equations of these 12 prognostic models as applied 156
in our cohort are shown. 157
None of the authors were involved in the development of any of these models. The results 158
are reported conform the tripod statement for transparent reporting of validation of 159
prognostic models (appendix D).29,30
160
161
Page 9 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 9
Study population of external validation cohort 162
We performed a large prospective multicentre cohort study (Risk EStimation for PrEgnancy 163
Complications to provide Tailored care (RESPECT) study) to validate the 12 prognostic 164
models for GDM. From December 2012 through January 2014 we included pregnant women 165
at their initial prenatal visit (i.e. <14 weeks of gestational age (GA)) in 31 independent 166
midwifery practices (primary care) and 6 hospitals (secondary/tertiary care) in the central 167
region of the Netherlands. We excluded women suffering from any type of diabetes mellitus 168
(DM) from the cohort. During the course of their pregnancy, participants received routine 169
antenatal care according to Dutch clinical guidelines. 170
This study was approved by the medical ethics committee of the University Medical Centre 171
Utrecht (protocol number 12-432/C) and written informed consent was obtained from all 172
participants. 173
174
Predictor assessment 175
In table 1 we have summarised predictors that were included in the 12 selected prognostic 176
models. Predictors for GDM were all measured in the first trimester at initial prenatal visit by 177
caregivers or via a self-administered questionnaire. Detailed information on predictor 178
definition and measurement can be found in Appendix B. 179
180
Outcome assessment 181
GDM was diagnosed by a 75 grams two-hour oral glucose tolerance test (OGTT) between 24 182
and 28 weeks of gestation according to the WHO 1999 guidelines as the presence of either a 183
fasting glucose level of ≥7.0 mmol/L (126 mg/dl) or a glucose level of ≥7.8 mmol/L (140 184
Page 10 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 10
mg/dl) after two hours.31,32
Women were offered an OGTT if risk factors or any signs of GDM 185
were present. Without risk factors or signs women were not tested and considered as not 186
having GDM. 187
Body mass index (BMI) in first trimester >30kg/m2, history of GDM, history of macrosomia 188
(birth weight above 95th
Dutch population centile)33
, family history of DM (first degree), non-189
western ethnicity, history of unexplained intrauterine foetal death, and polycystic ovary 190
syndrome (PCOS) were considered as risk factors for GDM. Polyhydramnios and macrosomia 191
were considered as possible signs of GDM. 192
For studies validating prognostic models there is no solid sample size recommendation, but a 193
minimum of 100 patients with events and at least 100 patients without events has been 194
suggested.34
195
196
Statistical Analysis 197
Predictor and outcome information was missing for some patients in the validation cohort 198
and these data were not missing completely at random, as can be derived from table 2. To 199
avoid biased validation of the models, we imputed the missing values using multiple 200
imputation.35
201
To start, we applied the “original” prognostic models, exactly as they were published, to our 202
study cohort when the full prediction rule, including its intercept, was available (appendix A). 203
Next, to allow for fair comparison of the prognostic models, we adjusted the intercepts of 204
the models to the cohort at hand, so that the mean predicted probability in each model was 205
equal to the observed outcome frequency. This is known as “recalibration-in-the-large” and 206
was performed by fitting a logistic regression model using the linear predictor as the only 207
Page 11 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 11
covariate in a new model (and offsetting it so that only the intercept would be 208
estimated).36,37
We then performed “logistic recalibration” by fitting logistic regression 209
models using the linear predictor as the only covariate. This resulted in an updated 210
calibration slope and intercept.36,38
211
We validated the original and recalibrated models by calculating the predicted probabilities 212
of GDM for each individual and comparing these with their observed outcomes. We assessed 213
discrimination using Harrell’s C statistic, which is equivalent to the area under the ROC 214
curve.39
It verifies whether participants with a higher predicted risk for GDM are indeed 215
more likely to have the disease. 216
Calibration refers to the agreement of predicted probabilities with observed proportions and 217
was assessed using calibration plots. When a model has perfect calibration, the predicted 218
probabilities equal the observed proportions. Thus, when a model is well calibrated, the 219
calibration plot has an intercept of 0 and a slope of 1. Some calibration plots have fewer 220
than 10 points because it was not possible to split the predicted probabilities into 10 groups. 221
This was the case for models with only a few categorical variables were included (e.g. sum 222
score models) in which a limited probabilities were possible. 223
A history of GDM is an important predictor in most models, but obviously not applicable for 224
nulliparous women. Therefore, discrimination and calibration of all 12 models were also 225
assessed in a subgroup analysis of nulliparous women. 226
All analyses were carried out on each of the multiple imputed datasets and Rubin’s rules 227
were used to combine the results into summary estimates. Analyses were performed using 228
the mice and rms packages of R-3.1 for windows (http://cran.r-project.org). 229
Page 12 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 12
Results 230
RESPECT cohort 231
3,723 women were included for analysis of which 1,655 (44%) were nulliparous (figure 1). 232
Table 2 shows the baseline characteristics of these women. GDM was diagnosed in 181 233
(4.9%) women, 33 (18%) of which needed insulin for glycaemic control. In the nulliparous 234
subgroup, 71 women (4.3%) developed GDM. 235
236
Calibration of the prognostic models 237
Three original publications provided the full prediction rule (Gabbay-Benziv 2014, Savona-238
Ventura 2013, and Van Leeuwen 2010) of which two models showed good calibration 239
(Gabbay-Benziv 2014, Van Leeuwen 2010) (figure 2). The model of Savona-Ventura et al. had 240
a poor calibration and tended to overestimate the risk of GDM. 241
Calibration plots were also drawn for each recalibrated model (figure 3). Most recalibrated 242
models showed good calibration, with the calibration line closely following the ideal 243
calibration line, except for four models. The models of Eleftheriades 2014, Naylor 1997, and 244
Tran 2013 seemed to overestimate the risk for women with an observed high risk of GDM, 245
whereas the model of Pintaudi 2014 seemed to underestimate the risk for these women. 246
When we compared the calibration plots of the original models with the recalibrated models 247
all three calibration plots improved. 248
249
Discrimination 250
C statistics for the original and recalibrated models ranged from 0.67 to 0.78 (table 3). The 251
four models with the highest C statistic (Gabbay-Benziv 2014, Nanda 2011, Teede 2011, and 252
Page 13 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 13
Van Leeuwen 2010) included maternal age, BMI, history of GDM, ethnicity, and family 253
history of DM as predictors. The poorest discriminating models were the models containing 254
the fewest predictors (Eleftheriades 2014, Savona-Ventura 2013, and Tran 2013). 255
256
Subgroup analysis 257
Discrimination for nulliparous women was worse as compared to the overall population for 258
four prognostic models (Gabbay-Benziv 2014, Nanda 2011, Naylor 1997, and Teede 2011) 259
(table 3). For all other models the C statistic was higher for nulliparous compared to all 260
women. Calibration of the prognostic models was also acceptable to good in the nulliparous 261
subgroup (figure 4). 262
Page 14 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 14
Discussion 263
Statement of principal findings 264
A total of 12 first trimester prognostic models for GDM were selected by comprehensive 265
review of literature and were compared head-to-head for their predictive accuracy in our 266
population-based cohort of 3,723 women. The model with the highest discriminative ability, 267
by Nanda et al., is any easy model to apply in clinical practice. Predictors in this particular 268
prognostic model (i.e. maternal age, BMI, ethnicity, history of GDM and history of 269
macrosomia) are easy to measure and their categorization of ethnicity is widely applicable. 270
Calibration was good for all models and improved by recalibrating the models to our 271
population. Although obstetric history is an important predictor in most models, the 272
prognostic models for GDM also performed well in nulliparous women. 273
274
Strengths and limitations 275
This has been the first external validation study that comprises almost all published first 276
trimester prognostic models for GDM in one single cohort study, allowing for head-to-head 277
comparison of these models. Our study had a large sample size and many cases of GDM. It 278
was performed using a prospective multicentre approach, and included an unselected 279
population of women from primary care (low-risk) as well as secondary/tertiary care (high-280
risk) within a geographically defined area. Additionally, missing data was handled by multiple 281
imputation, which is the most preferable method.34
282
However, some limitations of our study need to be addressed. First, according to Dutch 283
guidelines a high risk strategy was adhered. To prevent unnecessary testing in study 284
participants, women without predefined risk factors only underwent an OGTT in case of any 285
Page 15 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 15
symptoms of GDM. This strategy may have led to an underestimation of GDM cases in low-286
risk women. A study on the performance of similar strategies estimated that 7.3% of GDM 287
diagnoses may have been missed.40
However, this possible underestimation is unlikely to 288
have influenced the discriminative ability of the validated models since the C statistic is a 289
rank order insensitive to systematic errors in calibration such as differences in outcome 290
incidence.41
Moreover, this potential underestimation is also unlikely to have influenced our 291
inferences on the predictive accuracy of the models since we have recalibrated the 292
models, which accounts for any differences in overall incidence between the original model 293
development studies and our external validation study. 294
A second limitation of our study was that we were not able to include two published 295
prognostic models in our external validation. For one model information on the prediction 296
rule was not available despite contacting the authors. The other study was published after 297
the start of data collection for our validation cohort and information on some predictors 298
(maternal abdominal circumference and presence of PCOS) was not collected. 299
300
Comparison with other studies 301
Validation studies on prognostic models for GDM are scarce and our study differs from the 302
few previously published validation studies10–13
since we have performed a head-to-head 303
comparison where others validated a single prognostic model or only a small selection of the 304
prognostic models for GDM. However, our findings are similar to the findings of these 305
external validation studies, except for the external validation of the Van Leeuwen 2010 306
model by Lovati et al.10
In that study, the Van Leeuwen 2010 model yielded a poor C statistic 307
(0.60), in contrast to the other external validation studies, including our current study, that 308
Page 16 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 16
showed C statistics between 0.75 and 0.77.12,26
This might be due to the case-control study 309
design chosen by Lovati et al. in which it is not possible to adjust for observed outcome 310
frequency. 311
312
Clinical implications and conclusions 313
The use of accurate first trimester prognostic models for GDM will avoid the need to 314
perform an OGTT in all or many women, as is now recommended by various international 315
guidelines.42,43
A comparison of performance between the best discriminating prognostic 316
models and current strategies allows weighing the pros and cons (e.g. missed cases) and will 317
help to choose the model to be implemented into clinical practice. The decision which of the 318
four best models to implement in clinical practice may also depend on population 319
characteristics, the availability of predictors, and the incidence of GDM, and may therefore 320
be country- or region-specific. 321
Implementation of prognostic models for GDM early in pregnancy provides room for 322
preventive measures, i.e. lifestyle modification interventions such as diet and exercise 323
counseling.44,45
Additionally, metformin is likely to play a role in the prevention of GDM in 324
the near future.46,47
Early prevention, screening, diagnosis, and treatment of GDM when 325
necessary, can and will most likely reduce the rates of caesarean section, neonatal 326
hypoglycaemia and macrosomia and long term neonatal complications.4 327
To conclude, most of the 12 previously published prognostic models for GDM that have been 328
validated in this study show acceptable to good discrimination and calibration. Four models 329
outstand with C statistics of at least 0.75 (Gabbay-Benziv 2014, Nanda 2011, Teede 2011, 330
and Van Leeuwen 2010). We recommend that these four models will be further investigated 331
Page 17 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 17
for implementation in clinical practice. The model with the highest discriminative ability by 332
Nanda et al. is an easy model to apply in clinical practice, as the model consists of 333
straightforward predictors: maternal age, BMI, history of GDM, history of macrosomia and 334
ethnicity. Once prognostic models for GDM are applied in routine clinical care, further 335
research is recommended on the effects on clinical impact, actual development of GDM and 336
subsequent pregnancy outcomes. 337
Page 18 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 18
References 338
1. Kleinrouweler CE, Cheong-See FM, Collins GS, et al. Prognostic models in obstetrics: 339
available, but far from applicable. Am J Obstet Gynecol 2016; 214: 79–90.e36. 340
2. James JE. Personalised medicine, disease prevention, and the inverse care law: more 341
harm than benefit? Eur J Epidemiol 2014; 29: 383–90. 342
3. DeSisto CL, Kim SY, Sharma AJ. Prevalence estimates of gestational diabetes mellitus 343
in the United States, Pregnancy Risk Assessment Monitoring System (PRAMS), 2007-344
2010. Prev Chronic Dis 2014; 11: E104. 345
4. Metzger BE, Lowe LP, Dyer AR, et al. Hyperglycemia and adverse pregnancy outcomes. 346
N Engl J Med 2008; 358: 1991–2002. 347
5. Bellamy L, Casas J-P, Hingorani AD, Williams D. Type 2 diabetes mellitus after 348
gestational diabetes: a systematic review and meta-analysis. Lancet 2009; 373: 1773–349
9. 350
6. Clausen TD, Mathiesen ER, Hansen T, et al. High prevalence of type 2 diabetes and 351
pre-diabetes in adult offspring of women with gestational diabetes mellitus or type 1 352
diabetes: the role of intrauterine hyperglycemia. Diabetes Care 2008; 31: 340–6. 353
7. Alwan N, Tuffnell DJ, West J. Treatments for gestational diabetes. Cochrane database 354
Syst Rev 2009; 8: CD003395. 355
8. Viana LV, Gross JL, Azevedo MJ. Dietary intervention in patients with gestational 356
diabetes mellitus: a systematic review and meta-analysis of randomized clinical trials 357
on maternal and newborn outcomes. Diabetes Care 2014; 37: 3345–3355. 358
9. Werner EF, Pettker CM, Zuckerwise L, et al. Screening for gestational diabetes 359
mellitus: are the criteria proposed by the international association of the Diabetes and 360
Pregnancy Study Groups cost-effective? Diabetes Care 2012; 35: 529–35. 361
10. Lovati E, Beneventi F, Simonetta M, et al. Gestational diabetes mellitus: Including 362
serum pregnancy-associated plasma protein-A testing in the clinical management of 363
primiparous women? A case-control study. Diabetes Res Clin Pract 2013; 100: 340–364
347. 365
11. Van Leeuwen M, Opmeer BC, Zweers EJK, et al. External validation of a clinical scoring 366
system for the risk of gestational diabetes mellitus. Diabetes Res Clin Pract 2009; 85: 367
96–101. 368
12. Theriault S, Forest J-C, Masse J, Giguere Y. Validation of early risk-prediction models 369
for gestational diabetes based on clinical characteristics. Diabetes Res Clin Pract 2014; 370
103: 419–25. 371
13. Syngelaki A, Pastides A, Kotecha R, et al. First-Trimester Screening for Gestational 372
Diabetes Mellitus Based on Maternal Characteristics and History. Fetal Diagn Ther 373
2015; 38: 14–21. 374
14. Steyerberg EW, Moons KG, van der Windt DA, et al. Prognosis Research Strategy 375
(PROGRESS) 3: prognostic model research. PLoSMed 2013; 10: e1001381. 376
15. Justice AC. Assessing the Generalizability of Prognostic Information. Ann Intern Med 377
Page 19 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 19
1999; 130: 515. 378
16. Collins GS, Moons KG. Comparing risk prediction models. BMJ 2012; 344: e3186. 379
17. Savvidou M, Nelson SM, Makgoba M, et al. First-trimester prediction of gestational 380
diabetes mellitus: examining the potential of combining maternal characteristics and 381
laboratory measures. Diabetes 2010; 59: 3017–22. 382
18. Popova P V, Grineva EN, Gerasimov AS, et al. The new combination of risk factors 383
determining a high risk of gestational diabetes mellitus. Minerva Endocrinol 2014; 40: 384
239–47. 385
19. Caliskan E, Kayikcioglu F, Ozturk N, Koc S, Haberal A. A population-based risk factor 386
scoring will decrease unnecessary testing for the diagnosis of gestational diabetes 387
mellitus. Acta Obstet Gynecol Scand 2004; 83: 524–530. 388
20. Eleftheriades M, Papastefanou I, Lambrinoudaki I, et al. Elevated placental growth 389
factor concentrations at 11-14 weeks of gestation to predict gestational diabetes 390
mellitus. Metabolism 2014; 63: 1419–25. 391
21. Gabbay-Benziv R, Doyle LE, Blitzer M, Baschat AA. First trimester prediction of 392
maternal glycemic status. J Perinat Med 2014; 43: 283–9. 393
22. Nanda S, Savvidou M, Syngelaki A, Akolekar R, Nicolaides KH. Prediction of gestational 394
diabetes mellitus by maternal factors and biomarkers at 11 to 13 weeks. Prenat Diagn 395
2011; 31: 135–141. 396
23. Naylor CD, Sermer M, Chen E, Farine D. Selective screening for gestational diabetes 397
mellitus. N Engl J Med 1997; 337: 1591–1596. 398
24. Pintaudi B, Vieste GD, Corrado F, et al. Improvement of selective screening strategy 399
for gestational diabetes through a more accurate definition of high-risk groups. Eur J 400
Endocrinol 2014; 170: 87–93. 401
25. Savona-Ventura C, Vassallo J, Marre M, Karamanos BG. A composite risk assessment 402
model to screen for gestational diabetes mellitus among Mediterranean women. Int J 403
Gynaecol Obstet 2013; 120: 240–4. 404
26. Syngelaki A, Bredaki FE, Vaikousi E, Maiz N, Nicolaides KH. Body mass index at 11-13 405
weeks’ gestation and pregnancy complications. Fetal Diagn Ther 2011; 30: 250–65. 406
27. Tran TS, Hirst JE, Do MAT, Morris JM, Jeffery HE. A clinical model for early predicting a 407
vietnamese woman at risk of gestational diabetes mellitus. J Paediatr Child Health 408
2012; 48: 53. 409
28. Van Leeuwen M, Opmeer BC, Zweers EJ, et al. Estimating the risk of gestational 410
diabetes mellitus: a clinical prediction model based on patient characteristics and 411
medical history. BJOG 2010; 117: 69–75. 412
29. Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent Reporting of a 413
multivariable prediction model for Individual Prognosis Or Diagnosis (TRIPOD): The 414
TRIPOD Statement. Ann Intern Med 2015; 162: 55. 415
30. Correction: Transparent Reporting of a multivariable prediction model for Individual 416
Prognosis Or Diagnosis (TRIPOD): The TRIPOD Statement. Ann Intern Med 2015; 162: 417
600. 418
Page 20 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 20
31. World Health Organization. Definition, diagnosis and classification of diabetes mellitus 419
and its complications. Part 1: Diagnosis and classification of diabetes mellitus. 1999. 420
32. Nederlandse Diabetes Federatie (Dutch Diabetes Federation). NDF Zorgstandaard 421
Diabeteshttp://www.zorgstandaarddiabetes.nl/wp-content/uploads/2015/08/NDF-422
Zorgstandaard-zwangerschapsdiabetes-2015.pdf (2015). 423
33. Visser GHA, Eilers PHC, Elferink-Stinkens PM, Merkus HMWM, Wit JM. New Dutch 424
reference curves for birthweight by gestational age. Early Hum Dev 2009; 85: 737–44. 425
34. Moons KGM, de Groot JAH, Bouwmeester W, et al. Critical appraisal and data 426
extraction for systematic reviews of prediction modelling studies: the CHARMS 427
checklist. PLoS Med 2014; 11: e1001744. 428
35. Donders ART, van der Heijden GJMG, Stijnen T, Moons KGM. Review: a gentle 429
introduction to imputation of missing values. J Clin Epidemiol 2006; 59: 1087–91. 430
36. Steyerberg E, Borsboom G, van Houwelingen, HC; Eijkemans M, Habbema J. Validation 431
and updating of predictive logistic regression models: a study on sample size and 432
shrinkage. Stat Med 2004; 23: 2567–2586. 433
37. Toll, DB; Janssen, KJ; Vergouwe, Y; Moons K. Validation, updating and impact of 434
clinical prediction rules: a review. J Clin Epidemiol 2008; 61: 1085–1094. 435
38. Janssen KJM, Vergouwe Y, Kalkman CJ, Grobbee DE, Moons KGM. A simple method to 436
adjust clinical prediction models to local circumstances. Can J Anaesth = J Can 437
d’anesthésie 2009; 56: 194–201. 438
39. Harrel F. Regression modeling strategies: with applications to linear models, logistic 439
regression, and survival analysis. New York: Springer-Verlag, 2001. 440
40. Teh WT, Teede HJ, Paul E, et al. Risk factors for gestational diabetes mellitus: 441
Implications for the application of screening guidelines. Aust New Zeal J Obstet 442
Gynaecol 2011; 51: 26–30. 443
41. Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of prediction 444
models: a framework for traditional and novel measures. Epidemiology; 21: 128–38. 445
42. American Diabetes Association. Standards of medical care in diabetes--2011. Diabetes 446
Care 2011; 34 Suppl 1: S11–61. 447
43. Nankervis A, McIntyre H, Moses R, et al. ADIPS Consensus Guidelines for the Testing 448
and Diagnosis of Gestational Diabetes Mellitus in Australia. Australas Diabetes 449
Pregnancy Soc 2008; 1–8. 450
44. Bain E, Crane M, Tieu J, et al. Diet and exercise interventions for preventing 451
gestational diabetes mellitus. Cochrane database Syst Rev 2015; 4: CD010443. 452
45. Sanabria-Martínez G, García-Hermoso A, Poyatos-León R, et al. Effectiveness of 453
physical activity interventions on preventing gestational diabetes mellitus and 454
excessive maternal weight gain: a meta-analysis. BJOG 2015; 122: 1167–74. 455
46. Ainuddin JA, Kazi S, Aftab S, Kamran A. Metformin for preventing gestational diabetes 456
in women with polycystic ovarian syndrome. J Coll Physicians Surg Pak 2015; 25: 237–457
41. 458
47. Zhuo Z, Wang A, Yu H. Effect of metformin intervention during pregnancy on the 459
Page 21 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 21
gestational diabetes mellitus in women with polycystic ovary syndrome: a systematic 460
review and meta-analysis. J Diabetes Res 2014; 2014: 381231. 461
462
Page 22 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 22
Tables and figures 463
Figure 1. Flow chart of participants of the RESPECT cohort 464
Figure 2. Calibration plots original prognostic models 465
Figure 3. Calibration plots recalibrated prognostic models 466
Figure 4. Calibration plots of recalibrated prognostic models in nulliparous subgroup 467
Table 1. Summary of predictors per model 468
Table 2. Baseline characteristics stratified by variables that were available for imputation 469
Table 3. Predictive performance of prognostic models for gestational diabetes on the 470
RESPECT cohort 471
Appendix A. Full equations for GDM risk prediction models as applied in the validation 472
cohort 473
Appendix B. Description of predictors 474
Appendix C. Transparency declaration 475
Appendix D. TRIPOD checklist 476
Page 23 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 23
Table 1. Summary of predictors per model 477
Predictors Ca
lisk
an
20
04
Ele
fth
eri
ad
es
20
14
Ga
bb
ay
-Be
nzi
v 2
01
4
Na
nd
a 2
01
0
Na
ylo
r 1
99
7
Pin
tau
di
20
14
Sa
vo
na
-Ve
ntu
ra 2
01
3
Sh
ira
zia
n 2
00
9
Sy
ng
ela
ki
20
11
Te
ed
e 2
01
1
Tra
n 2
01
3
Va
n L
ee
uw
en
20
10
To
tal
Maternal age X X X X X X X X X X 10
Weight X 1
BMI, pre-pregnancy X 1
BMI X X X X X X X X X 9
Blood pressure X X 2
Hx of GDM X X X X 4
Family hx of DM X X X X X 5
Hx of chronic hypertension X X 2
Ethnicity X X X X X X 6
Parity X X 2
Poor obstetric outcome X 1
Hx of macrosomia X X X X 4
Method of conception X 1
Smoking X 1
Glucose X X 2
Total 5 2 5 5 3 4 3 3 8 5 3 5
BMI, body mass index; hx; history, GDM, gestational diabetes; DM, diabetes mellitus 1st
or 2nd
degree. 478 All predictors are measured in the first trimester at the initial prenatal visit, unless otherwise specified. 479
Page 24 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 24
Table 2. Baseline characteristics stratified by variables that were available for imputation 480
Characteristic
Missing
n(%)
Complete cases
(n =2603)
At least one
missing value
(n=1120) p value
Overall
RESPECT cohort
(n=3723)
Age, yrs 168 (4.5%) 30.9(4.34) 30.7(3.91) 0.286 30.8 (4.2)
BMI pre-pregnancy,
kg/m2
46 (1.2%) 23.3 [21.2, 26.3] 23.0 [20.9, 25.9] 0.013 23.2 [21.1, 26.2]
BMI, kg/m2 182 (4.8%) 23.8 [21.6, 26.9] 23.4 [21.3, 26.3] 0.010 23.7 [21.5, 26.7]
Systolic BP, mmHg 65 (1.7%) 114 (12) 115 (12) 0.480 115 (12)
Diastolic BP, mmHg 64 (1.7%) 67( 8) 67 (8) 0.092 67 (8)
Glucose, mmol/L 171 (4.5%) 4.7 [4.3, 5.1] 4.7 [4.4, 5.1] 0.343 4.7 [4.4, 5.1]
Ethnicity,
- White
- African
- Asian
- Mixed
- Other
732 (19.7%)
1665 (89.0%)
17 (0.9%)
30 (1.6%)
44 (2.4%)
115 (4.4%)
1066 (95.2%)
2 (0.2%)
11 (1.0%)
15 (1.3%)
26 (2.3%)
<0.001*
3387 (91.0%)
30 (0.8%)
53 (1.4%)
77 (2.1%)
176 (4.7%)
Education,
- Low
- Middle
- High
223 (6.0%)
198 (7.6%)
825 (31.7%)
1357 (57.0%)
52 (4.6%)
362 (32.3%)
706 (63.0%)
0.004*
270 (7.3%)
1273 (34.29%)
2180 (58.6%)
Smoking during
pregnancy
0 (0%) 258 (9.9%) 73 (6.5%) 0.001* 334 (9.0%)
History of chronic
hypertension
1 (0.0%) 43 (1.7%) 14 (1.2%) 0.440 57 (1.5%)
Family history of DM 1 (0.0%) 389 (15.0%) 154 (13.8%) 0.368 543 (14.6%)
Method of conception
- Spontaneous
- Ovulation drugs
- IVF
30 (0.8%)
2396 (93.1%)
61 (2.4%)
82 (3.2%)
1033 (92.2%)
38 (3.4%)
28 (2.5%)
0.199
3429 (92.9%)
99 (2.7%)
110 (3.0%)
Nulliparous 4 (0.0%) 1143 (44.0%) 509 (45.4%) 0.429 1655 (44.5%)
History of GDM 0 (0.0%) 47 (1.8%) 12 (1.1%) 0.133 59 (1.6%)
History of macrosomia
(>90th
percentile)
0 (0.0%) 146 (5.6%) 84 (7.5%) 0.034* 230 (6.2%)
Recurrent
miscarriages (≥2)
4 (0.0%) 173 (6.7%) 59 (5.3%) 0.125 232 (6.2%)
History of foetal death 0 (0.0%) 58 (2.2%) 16 (1.4%) 0.140 74 (2.0%)
GDM in pregnancy
- Insulin dependent
263 (7.0%) 116 (5.0%)
20 (0.8%)
53 (4.7%)
13 (1.2%)
0.839
0.327
181 (4.9%)
33 (0.9%)
Gestational age at
delivery, days
342 (9.2%) 280 [273, 285] 280 [274, 286] 0.357 280 [273, 285]
Sex, male 358 (9.6%) 1154 (50.7%) 569 (52.3%) 0.400 1902 (51.1%)
Birth weight, g
- percentile
- >90th
percentile
372 (10.0%) 3504
[3200, 3860]
55 [30, 77]
256 (12.0%)
3540
[3216, 3880]
57 [32, 80]
140 (13.2%)
0.127
0.066
0.367
3520
[3190, 3880]
55 [30, 79]
494 (13.3%)
yrs, years; BMI, body mass index; BP, blood pressure; DM, diabetes mellitus; IVF, in vitro fertilization; GDM, 481 gestational diabetes; Data are n, n(%), mean (SD), or median [IQR]. The column ‘overall RESPECT cohort’ 482 includes imputed data for those with missing values. * Significant at the P < 0.005 level. 483
Page 25 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
23 April 2016 25
Table 3. Predictive performance of prognostic models for gestational diabetes on the 484
RESPECT cohort 485
Prediction model
C statistic
Development
C statistic
Recalibrated
C statistic
Nulliparous
Caliskan
2004
NR 0.73
[0.69-0.76]
0.74
[0.68-0.80]
Eleftheriades
2014
0.73
[0.65-0.81]
0.70
[0.65-0.74]
0.73
[0.66-0.79]
Gabbay-Benziv
2014
0.81
[0.77-0.87]
0.75
[0.71-0.79]
0.72
[0.66-0.79]
Nanda
2011
0.79
[0.76-0.82]
0.78
[0.74-0.82]
0.76
[0.70-0.83]
Naylor
1997
0.69
NR
0.72
[0.69-0.76]
0.71
[0.65-0.77]
Pintaudi
2014
NR 0.72
[0.68-0.75]
0.73
[0.67-0.79]
Savona-Ventura
2013
0.89
[0.86-0.91]
0.68
[0.64-0.72]
0.72
[0.65-0.78]
Shirazian
2009
NR 0.71
[0.67-.75]
0.73
[0.66-0.79]
Syngelaki
2011
NR 0.71
[0.66-0.75]
0.76
[0.66-0.79]
Teede
2011
0.70
NR
0.77
[0.73-0.81]
0.76
[0.69-0.82]
Tran
2013
0.63
[0.60-0.65]
0.67
[0.63-0.72]
0.69
[0.63-0.76]
Van Leeuwen
2010
0.77
[0.69-0.85]
0.75
[0.71-0.78]
0.77
[0.71-0.84]
The C statistic of ‘development’ shows the C statistics as reported in the original publication if available. The C 486 statistics ‘recalibrated’ shows the C statistics per model, recalibrated to the RESPECT cohort. The C statistics 487 ‘Nulliparous’ shows the C statistic per model when applied to a subgroup of only nulliparous. 488 Data are presented in mean [95% confidence interval]. NR = not reported. 489
Page 26 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
Figure 1. Flow chart of participants of the RESPECT cohort 3,723 women were included for 145x80mm (300 x 300 DPI)
Page 27 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
Figure 2. Calibration plots original prognostic models
Three original publications pr
176x72mm (300 x 300 DPI)
Page 28 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
Figure 3. Calibration plots recalibrated prognostic models
Calibration plots were also dr
176x291mm (300 x 300 DPI)
Page 29 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
Figure 4. Calibration plots of recalibrated prognostic models in nulliparous subgroup Calibration of the prognostic 176x291mm (300 x 300 DPI)
Page 30 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nlyAppendix A. Full model equations for gestational diabetes as applied to the RESPECT cohort
Caliskan 2004
The probability of developing gestational diabetes was calculated as:
X = 1 (if adverse outcome(i.e. recurrent (≥2) abortions & previous IUFD)) + 1 (if age ≥25 yrs)
+ 1 (if BMI ≥25 kg/m2) + 1 (if family history of DM, first degree) + 1 (if parous with
previous LGA above 90th
percentile)
Eleftheriades 2014
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = α + 0.058 (weight, kg) + 0.182 (age, yrs)
Gabbay-Benziv 2014
The probability of developing gestational diabetes was calculated as: 1/(1+e-(X)), where
X = -11.569 + 0.064 (age, yrs) + 0 (if white race) + 2.026 (if Asian race) + 0.083 (if African
race) + 1.661 (if other nonwhite race) + 2.144 (history of GDM) + 0.034 (systolic
blood pressure, mmHg) + 0.082 (BMI, kg/m2)
Nanda 2011
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = α + 0.058 (age, yrs) + 0.113 (BMI, kg/m2) + 0 (if Caucasian ethnicity) + 0.888 (if Asian
ethnicity) + 0 (nulliparous) + 3.723 (if parous with previous GDM) + 0.67 (parous with
previous LGA above 90th
percentile)
Naylor 1997
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = α + 0 (age ≤30 yrs) + 0 (age 31-34 yrs) + 0.47 (age ≥35 yrs) + 0 (BMI ≤22.0 kg/m2) + 0.588
(BMI 22.1-25.0 kg/m2) + 1.163 (BMI ≥25.1 kg/m
2) + 0 (if white race) - 0.357 (if black
Page 31 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nlyrace) + 1.569 (if Asian race) + 0.47 (if other race)
Pintaudi 2014
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = α + 0 (class 4) + 1.361 (class 3) + 1.856 (class 2) + 3.091 (class 1) + 1.281 (previous LGA
with birth weight ≥4500 gr) + 0.588 (if family history of DM, first degree)
Class 1: random glucose >5.1 mmol/L
Class 2: random glucose >4.4-≤ 5.1 mmol/L & pre-pregnancy BMI >24.4 kg/m2
Class 3: random glucose >4.4-≤ 5.1 mmol/L & pre-pregnancy BMI ≤24.4 kg/m2
Class 4: random glucose ≤4.4 mmol/L
Savona-Ventura 2013
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = -4.144 + 3.142 (if glucose >5.0 mmol/L) + 0.758 (if age ≥30 yrs) + 0.543 (if diastolic blood
pressure ≥80 mmHg)
Shirazian 2009
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = α + 0 (if age ≤24 yrs) + 0.512 (if age 25-29 yrs) + 1.515 (if age ≥30 yrs) + 0 (if BMI pre-
pregnancy ≤24.9 kg/m2) + 0.513 (if BMI pre-pregnancy 25.0-29.9 kg/m
2) + 0.892 (if
BMI pre-pregnancy ≥30.0 kg/m2) + 0.842 (if family history of DM, first degree)
Syngelaki 2011
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = α + 0.014 (BMI, kg/m2) + 0.068 (age, yrs) + 0 (if Caucasian race) + 0.344 (if African race) +
1.050 (if Asian race) + 0.174 (if mixed race) + 0 (if spontaneous conception) + 0.432
(if conception with ovulation drugs) + 0.312 (if conception via IVF) + 0.020 (if
Page 32 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nlysmoking) – 0.010 (if chronic hypertension) + 0 (if nulliparous) – 0.211 (if parous
without previous LGA above 95th
percentile) + 0.663 (if parous with previous LGA
above 95th
percentile)
Teede 2011
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = α + 0 (if age <25 yrs) + 0.92 (if age 25-29 yrs) + 1.22 (if age 30-34 yrs) + 1.69 (if age 35-39
yrs) + 1.95 (if age ≥40 yrs) + 0 (if BMI <20 kg/m2) + 0.53 (if BMI 20.0-24.9 kg/m
2) +
0.69 (if BMI 25.0-26.9 kg/m2) + 0.83 (if BMI 27.0-29.9 kg/m
2) + 1.28 (if BMI 30.0-34.9
kg/m2) + 1.82 (if BMI ≥35.0 kg/m
2) + 1.31 (if Asian race) + 0.06 (if African race) + 0.37
(if other race) + 0.53 (if family history of DM, first degree) + 2.39 (if history of GDM)
Tran 2013
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = α + 0.351 (age, yrs) + 0.131 (BMI, kg/m2)
Van Leeuwen 2010
The probability of developing gestational diabetes was calculated as: eX/(1+e
X), where
X = -6.1 + 0.83 (if non-Caucasian race) + 0.57 (if family history of DM, first degree) – 0.67 (if
parous without history of GDM) + 0.5 (if parous with history of GDM) + 0.13 (BMI in
kg/m2 with BMI <22 transformed to 22, if > 30 transformed to 30)
IUFD, intra uterine fetal death; yrs, years; BMI, body mass index; DM, diabetes mellitus; LGA,
large for gestational age; kg, kilograms; GDM, gestational diabetes; gr, grams; IVF, in vitro
fertilization
Page 33 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nlyAppendix B. Description of predictors
Predictor measurement in validation cohort
Gestational age was calculated based on the first-trimester crown rump length
measurement at ultrasound examination (dating scan) using the formula of Robinson and
Fleming.1 Anthropometric variables were all measured at initial visit in the first trimester;
maternal body weight was measured in kilograms, height was measured in centimetres.
Body mass index (BMI) was calculated as weight (kilograms) divided by the squared height
(meters) of the subject. Blood pressure was measured in mmHg following standard
procedures.2 Mean arterial pressure (MAP) was calculated as (1/3 * systolic blood pressure)
+ (2/3 * diastolic blood pressure). A random glucose was measured in mmol/L. Maternal
characteristics as well as medical and obstetrical history were obtained through a self-
administered questionnaire. The following variables were relevant for this study: age (years),
ethnicity (Caucasian, African, Hindustani, Moroccan, Turkish, Middle Eastern, Asian, other
western, other nonwestern, and mixed), level of education (low, primary education or lower
level; middle, secondary education; high, tertiary education or higher level), cigarette
smoking during pregnancy (yes or no), method of conception (spontaneous, use of ovulation
drugs or in vitro fertilization), history of chronic hypertension (yes or no) and parity
(nulliparous with no previous pregnancy beyond 16 weeks or parous women), family history
of DM (first degree relative, yes or no), birth weight (grams) and birth weight percentile of
infants of previous pregnancies (based on national reference curves adjusted for parity, GA,
sex and ethnicity)3, previous pregnancies with GDM (yes or no), two or more spontaneous
abortions (yes or no), prior intra-uterine fetal death after 20 weeks of gestation (yes or no).
Page 34 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nlyAdjustments for external validation
Several continuous variables were transformed into categorical variables, according to
definitions in the original prediction model. This was the case for: age (Caliskan 2004, Naylor
1997, Shirazian 2009, Teede 2011), BMI (Caliskan 2004, Naylor 1997, Shirazian 2009, Teede
2011), and birth weight percentile of prior infants (Caliskan 2004, Syngelaki 2011). For one
model (Caliskan 2004) the cut-off for birth weight percentile of prior infants was not
specified nor provided on request. The most frequent used definition was applied (≥90th
percentile).
For family history of DM different definitions were used: one model (Shirazian 2009) defined
it as a relative with DM type II, degree not specified. A second model (Teede 2011) specified
a first degree relative, type of diabetes not specified. Another model considered family
history of DM positive if a 1st or 2nd degree relative has type I of type II DM (Van Leeuwen
2011). For the RESPECT-study participants family history of first degree relatives was
administered, there was no distinction made between type of DM. Therefore, the predictor
family history of DM was considered positive in all models if a first degree relative has any
type of DM.
Ethnicity was divided into ten subgroups in the original RESPECT-study. Most studies
(Gabbay-Benziv 2014, Naylor 1997, Syngelaki 2011) divided ethnicity into four groups:
White, Asian, Black, other. Ethnicity was recoded into these categories according to the
prediction models. Two studies (Nanda 2011 & Teede 2011) distinguished different Asian
types, for the RESPECT-study this was not possible. The most frequent Asian type was
Chinese, therefore we choose for east Asian of Chinese Asian predictor coefficients. Other
Asian type predictor coefficients were thereby excluded from the model.
Page 35 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nlyOne model (Teede 2011) included ‘poor obstetric outcome’ in the prediction model. This
variable was not further specified, and showed no significant contribution in the original
prediction model. Therefore this predictor was excluded from the model.
References
1. Robinson HP, Fleming JE. A critical evaluation of sonar ‘crown-rump length’ measurements. Br J Obstet Gynaecol 1975; 82: 702–10.
2. De Boer J, Zeeman K, Verhoeven C. Hypertensive disorders in pregnancy, labour and post partum period. 2011.
3. Visser GHA, Eilers PHC, Elferink-Stinkens PM, Merkus HMWM, Wit JM. New Dutch reference curves for birthweight by gestational age. Early Hum Dev 2009; 85: 737–44.
Page 36 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nlyAppendix C. Transparency declaration
Corresponding author
The corresponding author has the right to grant on behalf of all authors and does grant on
behalf of all authors, an exclusive licence on a worldwide basis to the BMJ Publishing Group
Ltd to permit this article (if accepted) to be published in BMJ editions and any other BMJPGL
products and sublicences such use and exploit all subsidiary rights, as set out in our licence.
Declaration of competing interests All authors have completed the Unified Competing
Interest form at www.icmje.org/coi_disclosure.pdf (available on request from the
corresponding author) and declare no support from any organization for the submitted work; no
relationships with companies that might have an interest in the submitted work in the
previous 3 years; no other relationships or activities that could appear to have influenced the
submitted work.
Details of contributors MPHK, AK, AF, KGMM and the RESPECT study group had the original
idea for the study and were involved in writing the original study protocol. The RESPECT
study group and MLdR were involved in data collection. CAN and MLdR performed data
analysis. MLdR, CAN, and MPHK wrote the first draft of the manuscript, which was
subsequently revised by AF, AK and KGMM. All authors participated in the final approval of
the manuscript. MPHK and AF are the guarantors of this study.
Page 37 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nlyStatement of access to all of the data All authors had full access to all of the data (including
statistical reports and tables) in the study and take responsibility for the integrity of the data
and the accuracy of the data analysis.
Transparency declaration The lead author affirms that the manuscript is an honest,
accurate, and transparent account of the study being reported; that no important aspects of
the study have been omitted; and that any discrepancies from the study as planned (and, if
relevant, registered) have been explained.
Ethical approval This study was approved by the medical ethics committee of the University
Medical Center Utrecht (protocol number 12-432/C) and written informed consent was
obtained from all participants.
Details of funding This study has been conducted with the support of The Netherlands
Organization for Health Research and Development (project nr 50-50200-98-060). The
funding source no role in the design, conduct, analyses, or reporting of the study or in the
decision to submit the manuscript for publication.
Data sharing statement Data sharing: patient level data and full dataset and technical
appendix and statistical code are available from the corresponding author. Informed consent
was not obtained but the presented data are anonymized and the risk of identification is
low.
Page 38 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
Appendix D. TRIPOD Checklist
Section/Topic Item Checklist Item Page
Title and abstract
Title 1 D;V Identify the study as developing and/or validating a multivariable prediction model, the target population, and the outcome to be predicted.
1
Abstract 2 D;V Provide a summary of objectives, study design, setting, participants, sample size, predictors, outcome, statistical analysis, results, and conclusions.
4
Introduction
Background and objectives
3a D;V Explain the medical context (including whether diagnostic or prognostic) and rationale for developing or validating the multivariable prediction model, including references to existing models.
6
3b D;V Specify the objectives, including whether the study describes the development or validation of the model or both.
7
Methods
Source of data
4a D;V Describe the study design or source of data (e.g., randomized trial, cohort, or registry data), separately for the development and validation data sets, if applicable.
8+9
4b D;V Specify the key study dates, including start of accrual; end of accrual; and, if applicable, end of follow-up.
9+11
Participants
5a D;V Specify key elements of the study setting (e.g., primary care, secondary care, general population) including number and location of centres.
9
5b D;V Describe eligibility criteria for participants. 9
5c D;V Give details of treatments received, if relevant. NA
Outcome 6a D;V
Clearly define the outcome that is predicted by the prediction model, including how and when assessed.
10
6b D;V Report any actions to blind assessment of the outcome to be predicted. NA
Predictors
7a D;V Clearly define all predictors used in developing or validating the multivariable prediction model, including how and when they were measured.
9
7b D;V Report any actions to blind assessment of predictors for the outcome and other predictors.
9
Sample size 8 D;V Explain how the study size was arrived at. 11
Missing data 9 D;V Describe how missing data were handled (e.g., complete-case analysis, single imputation, multiple imputation) with details of any imputation method.
11
Statistical analysis methods
10a D Describe how predictors were handled in the analyses. NA
10b D Specify type of model, all model-building procedures (including any predictor selection), and method for internal validation.
NA
10c V For validation, describe how the predictions were calculated. App A +B
10d D;V Specify all measures used to assess model performance and, if relevant, to compare multiple models.
12
10e V Describe any model updating (e.g., recalibration) arising from the validation, if done. 11 + App B
Risk groups 11 D;V Provide details on how risk groups were created, if done. NA
Development vs. validation
12 V For validation, identify any differences from the development data in setting, eligibility criteria, outcome, and predictors.
App A
Results
Participants
13a D;V Describe the flow of participants through the study, including the number of participants with and without the outcome and, if applicable, a summary of the follow-up time. A diagram may be helpful.
13
13b D;V Describe the characteristics of the participants (basic demographics, clinical features, available predictors), including the number of participants with missing data for predictors and outcome.
Table 2
13c V For validation, show a comparison with the development data of the distribution of important variables (demographics, predictors and outcome).
NA
Model development
14a D Specify the number of participants and outcome events in each analysis. NA
14b D If done, report the unadjusted association between each candidate predictor and outcome.
NA
Model specification
15a D Present the full prediction model to allow predictions for individuals (i.e., all regression coefficients, and model intercept or baseline survival at a given time point).
NA
15b D Explain how to the use the prediction model. NA
Model performance
16 D;V Report performance measures (with CIs) for the prediction model. Table 3
Model-updating 17 V If done, report the results from any model updating (i.e., model specification, model performance).
Table 3
Discussion
Limitations 18 D;V Discuss any limitations of the study (such as nonrepresentative sample, few events per predictor, missing data).
16
Interpretation 19a V
For validation, discuss the results with reference to performance in the development data, and any other validation data.
16
19b D;V Give an overall interpretation of the results, considering objectives, limitations, results from similar studies, and other relevant evidence.
15
Implications 20 D;V Discuss the potential clinical use of the model and implications for future research. 17
Other information
Supplementary information
21 D;V Provide information about the availability of supplementary resources, such as study protocol, Web calculator, and data sets.
App A-C
Funding 22 D;V Give the source of funding and the role of the funders for the present study. App C
Page 39 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960
Confidential: For Review O
nly
Appendix D. TRIPOD Checklist
*Items relevant only to the development of a prediction model are denoted by D, items relating solely to a validation of a prediction model are
denoted by V, and items relating to both are denoted D;V. We recommend using the TRIPOD Checklist in conjunction with the TRIPOD
Explanation and Elaboration document.
Page 40 of 39
https://mc.manuscriptcentral.com/bmj
BMJ
123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960