The identification and treatment of women with hyperglycaemia in pregnancy: an analysis of individual participant data, systematic reviews, meta-analyses and an economic evaluation

Study design and participants

‘Born in Bradford’ is a prospective birth cohort study22 of women who delivered a live singleton baby at the Bradford Royal Infirmary, Bradford, UK. Figure 1 shows full inclusion and exclusion of women from the BiB study22 in this study. Women were excluded from all analyses if they did not complete a baseline questionnaire or the OGTT or had missing data for ethnic origin. For the main analyses of the association of gestational glucose level with perinatal outcomes and development of GDM diagnostic criteria, we excluded women who were diagnosed with GDM. GDM was defined according to modified WHO criteria operating at the time (either fasting glucose level of ≥ 6.1 mmol/l or 2-hour post-load glucose level of ≥ 7.8 mmol/l). 11,23 The cohort is broadly representative of the obstetric population in Bradford. 22 All women booked for delivery in Bradford are offered a 75-g OGTT (comprising fasting and 2-hour post-load samples) at around 26–28 weeks’ gestation, and women were recruited mainly at their OGTT appointment. At recruitment, women had their height and weight measured, completed an interviewer-administered questionnaire, and provided written consent for information to be abstracted from their medical records. Interviews were undertaken in English or in SA languages (including Urdu and Mirpuri). The analysis of glucose samples was carried out using a Siemens Advia 2400 analyser from the ADVIA^® 2400 Clinical Chemistry System (Siemens Healthcare Ltd, Camberley, UK). The coefficients of variation range between 1.73% at 3.2 mmol/l and 0.64% at 19.1 mmol/l. Ethics approval was obtained from the Bradford Research Ethics Committee (07/H1302/112). All participants provided informed written consent.

FIGURE 1.

Study sample flow chart. The criteria used in the hospital in which study participants were recruited to diagnose GDM (and hence exclude them from the analyses presented here) was either fasting glucose level of ≥ 6.1 mmol/l or 2-hour post-load glucose level of ≥ 7.8 mmol/l. BRI, Bradford Royal Infirmary.

Participants completed a morning OGTT after fasting overnight. A baseline venous blood sample was taken. Participants then consumed a standard solution containing the equivalent of 75 g of anhydrous glucose over 5 minutes. After 2 hours a second sample was taken. Fasting and post-load plasma glucose assays were undertaken immediately using a glucose oxidase method. The analyses were undertaken using the Siemens Advia 2400 analyser following a standard protocol. The coefficients of variation range between 1.73% at 3.2 mmol/l and 0.64% at 19.1 mmol/l.

We assessed associations of maternal glucose concentrations with:

• three primary outcomes – LGA (defined as BW of > 90th percentile for gestational age), infant adiposity (defined as sum of skinfolds > 90th percentile for gestational age) and C-section

• five secondary outcomes – pre-eclampsia, preterm delivery, shoulder dystocia, instrumental vaginal delivery and admission to the neonatal unit.

These outcomes are established clinical complications of GDM, and similar to the primary and secondary outcomes in the HAPO study. 6 We did not have information about cord blood C-peptide or neonatal hypoglycaemia in our cohort. We were unable to calculate percentage body fat from skinfolds as done in the HAPO study6 because no equivalent formulae exist for SA infants; thus, we used a cut-off of > 90th percentile for the sum of skinfolds. We included C-section in our analyses as, although it is not used to predict future risk of adiposity and ill health, it is an important perinatal outcome and is associated with LGA, greater infant adiposity and increased health service costs. 33

Birthweight, mode of delivery (normal vaginal, instrumented vaginal or C-section), gestational age, pre-eclampsia, shoulder dystocia and admission to the neonatal unit were obtained from hospital records. C-section was compared with all vaginal deliveries. Pre-eclampsia was defined as new-onset proteinuria (> 300 g in 24 hours) together with blood pressure (BP) of ≥ 140/90 mmHg after 20 weeks’ gestation on more than one occasion. BWs were converted into standard deviation (SD) scores standardised for gestational age and gender relative to the UK-WHO growth standard. 34,35 Infants were then categorised as either being > 90th percentile or not. 27 The UK-WHO growth standards are based on data from six counties (USA, Norway, Oman, Brazil, India and Ghana) and describe the optimum pattern of growth for all children, rather than the prevailing pattern in the UK. 35 Skinfold thickness (triceps and subscapular) were summed and the 90th percentile was established from quantile regression using six gender–ethnic groups [combining gender and ethnic origin (WB, SA, and other)] and adjusted for parity (0, 1, 2, 3+). 36 The intra-rate and inter-rate technical error of measurements for the skinfold thicknesses were, respectively, 0.22–0.35 mm and 0.15–0.54 mm for triceps, and 0.14–0.25 mm and 0.17–0.63 mm for subscapular skinfolds. 37

Statistical analyses

Associations of fasting and post-load glucose levels with outcomes were assessed by categories, and with glucose as a continuous variable (per SD). We used multivariable logistic regression with clustered sandwich estimators38 (to account for some women in the cohort having more than one pregnancy) to assess associations of fasting and post-load glucose levels with each outcome. We followed the analytical protocol used in the HAPO study6 as closely as possible, with fasting and post-load glucose concentrations divided into seven categories (see the Table 3 footnotes for definition of categories). In order to explore any extreme threshold effects, the top two categories for fasting and post-load glucose levels included about 1% and 3% of women, respectively. Models were adjusted for gestational age at OGTT, presence or absence of family history of diabetes, family history of hypertension, previous GDM, previous macrosomia, smoking status, alcohol consumption during pregnancy, maternal age and BMI, maternal education, baby gender and parity. Models for all women were additionally adjusted for ethnic origin. Models for SA women were not adjusted for alcohol consumption during pregnancy because most reported never drinking alcohol. Additionally, preterm delivery was adjusted for squared maternal BMI because of evidence of a quadratic relationship of BMI with preterm delivery. Shoulder dystocia models were not adjusted for previous GDM because of small numbers. Ethnicity was categorised as WB, SA and other ethnicity according to UK Office for National Statistics criteria. 39 Education was equivalised to UK standard attainments, and participants were included in one of five mutually exclusive categories (< 5 GCSE equivalent, 5+ GCSE equivalent, A level equivalent, higher than A level, other). 23 Parity was categorised as 0 or ≥ 1 previous pregnancies. Smoking was categorised as never, past (not during this index pregnancy), current (during this pregnancy) and alcohol as consumed during this pregnancy or not.

Maternal BMI was calculated from height measured at the time of recruitment and from weight measured at booking antenatal clinic, which was obtained from electronic hospital records. Expected date of birth (40 weeks) was estimated from a gestation ultrasound scan at ≈ 10 weeks then, using the date of OGTT and date of birth, gestational age at OGTT and birth were calculated. Infant gender was obtained from electronic hospital records, and family history of diabetes and hypertension were abstracted from paper hospital records.

We established fasting and post-load glucose thresholds for BW of > 90th percentile and standardised sum of skinfolds of > 90th percentile that equated to an OR of 1.75, using the methods of IADPSG. 8 We estimated the ORs of these outcomes at mean glucose levels and the ORs at 0.1-mmol/l intervals across the full range of fasting and 2-hour post-load glucose levels. We then plotted this range of ORs and used the plots to estimate the thresholds of fasting, and 2-hour post-load glucose that were closest to ORs for each outcome of 1.75 in both ethnic groups. These analyses were carried out with adjustment for the same potential confounders as in all multivariable regression analyses. All analyses were undertaken separately in WB and SA women, and we tested for differences in associations by including an interaction term between glucose and ethnic origin. Because women of SA origin were mainly Pakistani, we undertook a sensitivity analysis in which we repeated analyses including only Pakistani women. To maximise statistical power and minimise bias that might occur if women with missing data were excluded from analyses, we used multivariate multiple imputation with chained equations to impute missing values40 (see Appendix 1, Table 50). We repeated all analyses with the complete data cohort for comparison.

Levels of missing data range from 0% to 32% for the different variables (see Appendix 1, Table 50), and 5056 (53%) had complete data on all variables included in any analyses. To maximise statistical power and minimise bias due to excluding those with any missing data, we used multivariate multiple imputation, with chained equations to impute missing values for covariables and outcomes for the main analyses. 40 We generated 50 imputed data sets and combined these using Rubin’s rules, using the ‘mi’ commands in Stata 13 (StataCorp LP, College Station, TX, USA). Distributions of variables from pooling of the data sets with imputed variables were similar to those for observed variables (see Appendix 1, Table 50). We repeated all analyses with the complete data cohort for comparison.

Associations of fasting and post-load glucose levels with primary outcomes

Figure 2 shows the unadjusted percentage of women in each group who had each of the three primary outcomes by categories of fasting and 2-hour post-load glucose level by ethnicity and for all women. Generally, the frequency of each of the three primary outcomes increased across the seven categories of fasting and post-load glucose levels, with no evidence of a threshold at which risk markedly increases, except for the association of fasting glucose level with C-section in SA women. The higher prevalence of BW of > 90th percentile in WB infants than in SA infants is consistent across all glucose categories. Combining data for all women (i.e. including 99% of the cohort) showed monotonic relationships of fasting and post-load glucose levels up to the sixth category (see Appendix 1, Figures 45 and 46).

FIGURE 2.

Frequency of primary outcomes across glucose categories by ethnicity: WB (n = 3888) and SA (n = 4821), and for all pregnancies (N = 9509). (a) BW of > 90th percentile; (b) sum of skinfolds > 90th percentile; (c) C-section. Glucose categories are defined as follows: FPG level: category 1, < 4.3 mmol/l; category 2, 4.3–4.4 mmol/l; category 3, 4.5–4.7 mmol/l; category 4, 4.8–4.9 mmol/l; category 5, 5.0–5.2 mmol/l; category 6, 5.3–5.6 mmol/l; category 7, 5.7–6.0 mmol/l. Post-load plasma glucose level: category 1, < 4.7 mmol/l; category 2, 4.7–5.4 mmol/l; category 3, 5.5–6.2 mmol/l; category 4, 6.3–6.6 mmol/l; category 5, 6.7–7.2 mmol/l; category 6, 7.3–7.5 mmol/l; category 7, 7.6–7.7 mmol/l. FPG, fasting plasma glucose.

Regression analyses confirmed monotonic associations of glucose level with each of the primary outcomes, in each group, without (see Appendix 1, Table 52) and with adjustment for confounders (Table 4). In view of the monotonic nature of the associations, we focused our comparisons on results with fasting or post-load glucose level as a continuous variable (per 1 SD). Although there was not strong statistical evidence of differences, the point estimates suggested stronger associations of fasting and post-load glucose levels with all three outcomes, except for those of fasting glucose level with LGA and post-load glucose level with C-section. However, there was no strong statistical evidence that the associations differed between the two groups for any primary outcome (p interaction of ≥ 0.2 for all associations).

Associations of fasting and post-load glucose levels with secondary outcomes

Associations with secondary outcomes were similar in the two ethnic groups [see Appendix 1, Table 53 (unadjusted) and Table 54 (confounder adjusted)]. The frequency of pre-eclampsia, shoulder dystocia and, with a weaker magnitude, instrumental delivery, also increased across each glucose category, especially with fasting glucose level (see Appendix 1, Figures 45 and 46). Neither fasting nor post-load glucose concentrations were clearly associated with preterm delivery or admission to the neonatal unit.

Criteria for diagnosing gestational diabetes mellitus

Table 5 shows the thresholds of fasting and glucose that would result in an OR of 1.75 for BW of > 90th percentile, and sum of skinfolds of > 90th percentile in each group. Fasting and post-load glucose thresholds based on the average of BW and skinfolds of > 90th percentile for all women irrespective of ethnic origin were 5.3 mmol/l and 7.5 mmol/l, respectively. Fasting glucose thresholds based on BW or the average of BW and skinfolds of > 90th percentile were higher for WB women than for SA women (see Table 5); with skinfolds of > 90th percentile alone as the outcome, the fasting glucose threshold was the same in both ethnic groups. There was no 2-hour post-load threshold that reached an OR of 1.75 for BW of > 90th percentile in either ethnic group. A threshold for sum of skinfolds of > 90th percentile was found only in SA women (see Table 5).

Table 6 shows GDM prevalence by past and present diagnostic criteria, and the criteria derived from our data. For our study criteria, we show prevalences with the same thresholds in both ethnic groups (the thresholds derived for all women) and also ethnic-specific thresholds. Prevalence of GDM was about twice as high in SA women using any criteria range (4.1–17.4%) than in WB women (1.2–8.7%) for all non-ethnic specific criteria. Prevalence was greater in both ethnic groups with the recently derived IADPSG, NICE and our criteria than the 1999 WHO criteria. Of the three recent criteria, the NICE criteria resulted in the lowest prevalences in WB women and our criteria the highest. In SA women, the NICE criteria resulted in the lowest prevalence. If we applied criteria derived in our study for all women (i.e. not taking account of ethnic origin) to the SA women, the prevalence of GDM was the same using either IADPSG/WHO or our criteria. However, when we applied our ethnic-specific criteria, prevalence in SA women was nearly three times that in WB women (see Table 6).

Additional sensitivity analyses

The number of missing data ranged from 0% to 32% for the different variables (see Appendix 1, Table 50) and 5056 of the 9509 (53%) had complete data on all variables for the main analyses. Distributions of any variable with missing data were the same in the imputation data sets (see Table 4 and Appendix 1, Table 54) and for observed complete case data (see Appendix 1, Tables 56 and 57). There was no strong evidence for a quadratic curvilinear association between fasting or post-load glucose level and any of the primary or secondary outcomes (see Appendix 1, Table 55). The results of analyses restricted to Pakistani women did not differ from those presented for all SA women (see Appendix 1, Tables 58 and 59).

Previous systematic reviews

To our knowledge, this is the first comprehensive systematic review and meta-analyses examining the association of gestational glucose level with risk of perinatal and longer-term outcomes. We have previously undertaken a review as part of a master’s degree dissertation. 55 In that dissertation, a systematic search was undertaken to identify studies that investigated the association between gestational glucose levels [measured using the OGTT or oral glucose challenge test (OGCT)] and adverse outcomes. The findings of that review suggested strong associations between fasting glucose categories and both LGA and macrosomia and these associations were weaker for 2-hour post-load glucose categories. However, that review included only studies that had been published up to March 2013, and we are aware of additional studies since then. Furthermore, there was no attempt to explore sources of heterogeneity between studies.

The aim in this study was to expand and update the previous search and analyses in order to determine associations between fasting and post-load glucose levels, and both perinatal and longer-term maternal and offspring outcomes. This section is reported in accordance with PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines. 56

Search

The original (master’s dissertation) searches were undertaken in March 2013. 55 The search strategies including interfaces and search terms used, and how they were combined, are shown in Appendix 7 (see Tables 83 and 84). Although no date or language restrictions were placed on the searches, only studies with an English language title and/or an abstract were screened for inclusion. The same search strategy (as in March 2013) was used for updating this review, with repeat searches undertaken in September 2013 and October 2014. The following databases were searched: MEDLINE^® and MEDLINE In-Process & Other Non-Indexed Citations,^® EMBASE, Cumulative Index to Nursing and Allied Health Literature (CINAHL) Plus, Cochrane Central Register of Controlled Trials (CENTRAL), Cochrane Database of Systematic Reviews (CDSR), Database of Abstracts of Reviews of Effects (DARE), Health Technology Assessment (HTA) database, NHS Economic Evaluation Database (NHS EED) and the Cochrane Methodology Register (CMR).

The search results were downloaded into an EndNote (Thomson Reuters, CA, USA) library and duplicates were removed. All records (title, publication details and abstracts if available) were screened for eligibility, independently, by two reviewers. We had previously screened the records identified by the March 2013 search; however, we rescreened these again to ensure that the screening standard was high and consistent across all searches. All studies identified as potential ‘includes’ were checked by a second reviewer. Disagreements were resolved by discussion or by a third reviewer. The reference lists of all of the included studies and any related systematic reviews identified were checked for further possible inclusions.

Inclusion and exclusion criteria

All studies reporting the association of fasting or post-load glucose level (obtained from an OGTT or OGCT) with perinatal and longer-term health outcomes in mother or offspring were potentially eligible. Associations in women diagnosed with GDM, however defined, were excluded. The included studies had to have the following characteristics.

Quality assessment

The risk of bias in the included published studies was assessed using a modified version of the Critical Appraisal Skills Programme57 (CASP) and Quality in Prognosis Studies (QUIPS) quality assessment tool. 58 These tools are designed to aid the assessment of observational studies of association and prediction. The following quality criteria were considered:

• representative nature of included population

• loss to follow-up

• consistency of glucose measurement and outcome assessment

• blinding of participants and medical practitioners to glucose level

• blinding of outcome assessors to glucose level

• selective reporting of outcomes

• adjustment of results for key confounding variables.

Each criterion was classified as being at low, high or unclear risk of bias. One reviewer performed the quality assessment; all assessments were then checked by a second reviewer.

Data extraction

Data were extracted from each publication on the following:

• glucose test used:

  • OGCT or OGTT

  • glucose load (50 g, 75 g or 100 g)

  • timing (fasting, 1-, 2- or 3-hour post load)

• glucose levels in each defined glucose category (when reported)

  • in millimoles per litre; levels presented as milligrammes per decilitre were converted to millimoles per litre

• numbers of women in each glucose category

• for each outcome reported:

  • definition of outcome

  • number of outcome events in each glucose category

  • relative risk (RR) or OR of outcomes in each glucose category relative to baseline category (if reported)

  • RR or OR per millimole per litre of glucose or per SD of glucose (if reported)

• whether women with pre-existing diabetes and GDM were excluded

• study location

• how GDM was defined

• which potential confounding factors were adjusted for

• each quality criterion.

When presented, RR or ORs adjusted for key confounding factors (such as age, BMI, parity) were extracted for each glucose category or type of glucose measure.

One reviewer performed the data extraction. A second reviewer checked the accuracy of the data extraction for all included studies, but did not independently extract data.

Contact with authors and individual participant data

Having identified all relevant published studies that fulfilled our inclusion criteria, and, given that a goal of this project was to understand GDM within a contemporary UK population, we searched for recently recruited cohorts in the UK. We identified four eligible cohorts with IPD: two had sufficiently complete case data: the BiB study22 (data provided by John Wright, Bradford Institute for Health Research, September 2013) and the Atlantic Diabetes in Pregnancy cohort (Atlantic DIP59) (data provided by Fidelma Dunne, Department of Medicine, National University of Ireland, September 2013), one cohort had insufficient complete case data and was not included (Warwick/Coventry: P Saravanan, Warwick Medical School, 2013, personal communication60) and we were unable to secure data from one other (UK HAPO cohort6); however, we have included the published estimates from the whole HAPO cohort6 wherever possible.

Both the BiB and Atlantic DIP cohorts22,59 include fasting and 2-hour post-load glucose levels, obtained as part of a 75-g OGTT. When outcomes were not reported explicitly in the data set they were derived from available data if possible. For example, macrosomia, LGA and preterm birth were calculated from BW and gestational age data.

Statistical analyses

Included studies

The search from the unpublished review55 and the updated searches together identified 11,219 potentially relevant studies following removal of duplicates. After title and abstract screening 125 publications were obtained for full-text review. After full-text screening, 57 studies (see Tables 7–9) were included in the review and 37 in the meta-analysis (including the two studies22,59 for which we had IPD) (see Appendix 2, Table 62 for excluded studies with reasons). Figure 3 shows the identification of these studies.

FIGURE 3.

The search process. a, Includes 7947 from the March 2013 search, 808 from the September 2013 search, and 2464 from the October 2014 search.

Several publications reported data from the same cohort. Four of the included publications used data from the HAPO cohort,6 but reported different outcomes. One of these publications (Pettitt et al. 41) was not included in the analyses because it reported associations with outcomes in a subset of participants that had been previously reported in the whole cohort. For the remaining publications, data from the most recent and comprehensive publication for each outcome were used. Two publications61,64 used data from the same cohort: Figueroa et al. 64 examined glucose levels at OGCT and risk of adverse outcomes, whereas Landon et al. 61 examined glucose levels at OGTT and risk of adverse outcomes. Data from both publications are therefore included in analyses.

Characteristics of eligible studies are described in Tables 7–9. The studies fall into four categories: (1) 28 studies (including BiB and Atlantic DIP)6,61,64–87 reported associations between glucose levels (from OGTT or OGCT) split into three or more categories and adverse perinatal outcomes (see Table 7); (2) 20 studies88–107 reported associations between glucose levels (from OGTT or OGCT) split into two categories with adverse perinatal outcomes (see Table 8) – these studies were mostly comparisons of women with lower glucose levels at OGCT [typically < 140 mg/dl (7.8 mmol/l)] compared with women with higher glucose levels at OGCT; (3) five studies36,41,65,108,109 reported longer-term outcomes in either mother or offspring (see Table 9) (it was not possible to pool studies reporting longer-term outcomes because they were too diverse); and (4) the remaining five studies110–113 did not present numerical data that were suitable for analysis and therefore could not be included in any of the meta-analyses (see Appendix 2, Table 63). One study114 used a 75-g OGTT in a non-fasted population. As there were no other studies that had used this test in this way, and post-load glucose levels from a non-fasted group are likely to differ to those from a fasted group, we did not include results from this study in any meta-analyses.

Quality assessment

The results of the quality assessment of the included studies are shown in Appendix 2, Table 60. In general most studies were at low risk of bias. Most studies recruited any pregnant women, without any further inclusion/exclusion criteria, so that the study population would be more representative of the whole ‘general’ obstetric population. The majority of studies were in western populations from high-income countries, with a small number from other populations, for example the Pima Indian population of Arizona. There was little loss to follow-up in most studies. Fasting and post-load glucose levels were assessed using OGCT and/or OGTT, and most outcomes were measured using standard definitions.

The main potential risk of bias was due to lack of blinding. Blinding of participants, clinicians, research staff and those assessing outcomes to glucose levels is possible; however, all staff may have been aware that glucose levels did not indicate GDM (study defined) requiring treatment. Many studies were retrospective, and in these cases it is likely that participants and medical staff were aware of the glucose levels.

Consequently, it is possible that associations of glucose levels with perinatal outcomes were ‘confounded by indication’. Women and health practitioners may have been influenced by glucose results. For example, in those women who had levels just below the thresholds for GDM diagnosis, monitoring and treatment may have differed from those whose levels were lower. Those with higher levels might have been given lifestyle advice aimed at reducing their levels, they may have been monitored more closely and there may have been a greater likelihood of intervening during labour, such as electing for C-section.

Studies generally reported all outcomes listed in their methods sections, but no single study reported data on all of our included outcomes. The studies were necessarily observational (randomisation to a given glucose level is not possible). In addition to confounding by indication, it is possible that other characteristics might have confounded the associations that we have examined. Only a minority of studies reported associations that were adjusted for what we considered to be key confounding factors (maternal age, BMI/other measure of adiposity and previous GDM).

The two cohorts for which we had IPD were judged – like most other studies described in Appendix 2, Table 60 – to be at low risk of bias. For these we had the added advantage that we were able to adjust for key confounders, although as with other studies we cannot rule out confounding by indication or residual confounding by characteristics that were unmeasured or poorly measured in those studies.

Analyses of individual participant data cohorts

The frequencies of each adverse outcome across seven fasting and 2-hour post-load glucose categories in the BiB study22 and Atlantic DIP59 are shown in Appendix 2, Figure 47. We have not included skinfolds of > 90th percentile as an outcome or any other measure of infant adiposity because it was not available in the Atlantic DIP cohort59 and in relation to the BiB study22 it has been reported in Chapter 2. For comparison we have included the HAPO study6 point estimates in the analyses presented in Appendix 2, Figures 48 and 49 (we were unable to secure IPD from the UK centres of the HAPO study6) because the results from that study6 have recently been used to develop new criteria thresholds for GDM diagnosis.

Across all categories of fasting and post-load glucose levels the frequencies of C-section, instrumental birth, LGA, macrosomia and pre-eclampsia are greater in the Atlantic DIP cohort59 than in the BiB study. 22 Preterm birth is similar in both studies, and numbers for shoulder dystocia are too few to draw conclusions.

The ORs per 1-mmol/l increase in fasting and post-load glucose levels for the BiB, Atlantic DIP and HAPO studies6,22,59 are shown in Appendix 2, Figures 48 and 49, respectively. For most outcomes the cohorts show similar results, with increases in outcome incidence as glucose levels increase, although results were not always statistically significant in the smaller Atlantic DIP cohort. 59 There are some exceptions to this. For instrumental delivery, the BiB study22 shows a positive association between outcome and glucose level, but no such association was found in the Atlantic DIP study. 59 For the Atlantic DIP study,59 risk of preterm birth reduced as fasting glucose level increased, but increased as post-load glucose level increased. This may be a chance finding and related to the low incidence of preterm birth in the Atlantic DIP study59 (3%). Associations were stronger for fasting glucose levels than 2-hour post-load glucose levels. Meta-analyses provided significant results for ORs per 1-mmol/l increases in glucose level and the majority of outcomes, with the exception of instrumental and preterm birth, for fasting glucose level.

In Appendix 2, Figures 50 and 51 show the ORs for each outcome with increasing fasting and post-load glucose categories, respectively, relative to the mean glucose level for each outcome in each of the cohorts. The dashed vertical lines show the thresholds for diagnosing GDM using the IADPSG and WHO (1999) criteria (fasting glucose levels of 5.1 mmol/l and 6.1 mmol/l; post-load glucose levels of 7.8 mmol/l and 8.5 mmol/l, respectively) and the horizontal dashed line an OR of 1.75 (the OR recommended by the IADPSG for applying GDM diagnostic thresholds).

The estimated ORs illustrated in Appendix 2, Figure 50, which are greater than the WHO fasting threshold of 6.0 mmol/l, and in Appendix 2, Figure 51, which are greater than the post-load threshold of 7.7 mmol/l, are predictions assuming that the linear trend continues at higher glucose levels (if women are not treated), and are not based on the data from the cohorts (because in both cohorts women were offered treatment if their glucose levels were greater and therefore have been excluded). For fasting glucose level, the two cohorts give similar results for C-section, LGA, macrosomia, pre-eclampsia and shoulder dystocia. For macrosomia, LGA and shoulder dystocia the IADPSG diagnostic threshold of 5.1 mmol/l corresponds reasonably closely with an OR for outcomes relative to average fasting glucose level of between 1.5 and 1.75 mmol/l. The WHO thresholds are associated with a higher OR for adverse outcome.

Trends in perinatal outcome risk with glucose levels

This section examines 28 studies presenting data on perinatal outcomes in three or more glucose categories using either an OGCT or OGTT (see Table 7). This section presents a series of figures for which the risk of a specified outcome is plotted against glucose levels. Studies are categorised according to the timing of the glucose test (fasting, 1-hour post load, 2-hour post load) and the glucose load used (50-g OGCT, 75-g OGTT, 100-g OGTT).

Appendix 2, Figure 52, shows the trend for macrosomia, and Appendix 2, Figure 53, for LGA. For both outcomes the analyses suggest the risk increases as glucose levels increase, the association seems stronger for fasting glucose level compared with post-load glucose level. The relationship appears to be linear, with no sudden increase in risk. There is considerable heterogeneity across studies in the underlying risk of macrosomia and LGA, but the trends (i.e. the slopes of the lines) appear reasonably consistent across studies. Although there are differences in the actual glucose levels according to the glucose test used, there is no evidence that the trend in risk with glucose level is different for the different glucose tests.

In Appendix 2, Figure 54 shows the trends for pre-eclampsia and increasing glucose levels; Figure 55, C-section, Figure 56 instrumental birth; and Figure 57, induction of labour and increasing glucose levels. Risk of pre-eclampsia and C-section seems to increase with increasing glucose level similarly to macrosomia and LGA. Data on assisted delivery (forceps and ventouse) and induced labour are too few to draw any meaningful conclusions.

Furthermore, in Appendix 2, Figure 58 shows the trends for shoulder dystocia; Figure 59 the trends for preterm birth; and Figure 60 the trends for neonatal hypoglycaemia. Although data for all three outcomes are limited, these outcomes do not seem to be associated with increasing glucose levels. The risk of shoulder dystocia appears to increase only at the higher levels of glucose (e.g. > 6 mmol/l at 2 hours post load), although this observation is driven primarily by one study. 61

Association between 1-mmol/l increases in fasting and post-load glucose levels and risk of adverse perinatal outcomes

This section examines the trends in adverse perinatal outcomes (see Appendix 2, Figures 52–60) and increasing glucose levels represented as ORs for each outcome per 1-mmol/l increase in glucose level, calculated using the logistic regression models described above (see Methods). Each glucose load/test is considered separately, and when there are sufficient studies for any outcome, results are combined in meta-analyses.

50-g oral glucose challenge test

Figure 4 shows the OR per 1-mmol/l increase in 1-hour post-load glucose for outcomes using the 50-g OGCT. The associations between C-section, LGA, macrosomia, neonatal hypoglycaemia, pre-eclampsia and shoulder dystocia are statistically significantly associated with 1-mmol/l increases in glucose level. Preterm birth does not seem to be associated with increases in glucose level, or does PIH/pre-eclampsia (some studies combine these two outcomes).

FIGURE 4.

Odds ratio for 1-mmol/l increases in 1-hour post-load glucose for 50-g OGCT and adverse outcomes.

75-g oral glucose tolerance test

Figure 5 shows the OR per 1-mmol/l increase in fasting glucose for all outcomes. The risk of the majority of adverse outcomes increases with increasing fasting glucose level, with statistically significant associations for LGA, macrosomia, pre-eclampsia, C-section, induced labour and neonatal hypoglycaemia. There is no evidence of increasing odds with increasing glucose levels for shoulder dystocia or instrumental birth (forceps and ventouse). There seems to be a negative association between increasing fasting glucose levels and preterm birth, suggesting that, as glucose levels increase, the odds of a preterm birth reduces by 23% (OR 0.77, 95% CI 0.62 to 0.96).

FIGURE 5.

Odds ratio for 1-mmol/l increases in fasting glucose level for 75-g OGTT and adverse outcomes.

Appendix 2, Figure 61, shows the OR per 1-mmol/l increase in 1-hour post-load glucose for reported outcomes without meta-analysis because of the limited number of studies included. The HAPO study6 reports BW of > 90th percentile and Pettitt et al. 41 report LGA; however, Pettitt et al. report results for a subset of participants from the whole HAPO study. 6 The analyses suggest that the 2-hour post-load glucose (Figure 6) associations are weaker than the fasting glucose associations (see Figure 5), and the statistically significant associations between increasing fasting glucose and reduced odds of preterm birth and increased odds of C-section are lost.

FIGURE 6.

Odds ratio for 1-mmol/l increases in 2-hour post-load glucose level for 75-g OGTT and adverse outcomes.

100-g oral glucose challenge test

Figure 7 shows the OR per 1-mmol/l increase in fasting glucose for all outcomes using the 100-g OGTT, and Figure 8 shows the results for 2-hour post-load glucose without meta-analysis because of the limited number of studies. Only one study61 reported 1-hour 100-g results for outcomes including cord C-peptide and LGA. One study80 performed the OGTT at 9 weeks rather than at the more usual 26–28 weeks. Fasting results (the only levels reported by this study80) seem consistent with the fasting glucose results reported at the more conventional 26–28 weeks. 83

FIGURE 7.

Odds ratio for 1-mmol/l increases in fasting glucose for 100-g OGTT and adverse outcomes.

FIGURE 8.

Odds ratio for 1-mmol/l increases in 2-hour post-load glucose for 100-g OGTT and adverse outcomes.

It is difficult to draw conclusions from the results reported by these studies, given the limited data, but the findings are similar to the post-load associations of the 75-g test results. Fasting glucose and associated outcomes should not be affected by the subsequent glucose load; however, differences in glucose load and subsequent post-load glucose associations may be.

Combining 75-g and 100-g glucose test results

To increase the number of studies and participants included in the comparisons, we combined the results for the 75-g and 100-g OGTTs. We therefore assumed that the association between outcomes and increases in glucose were the same for both tests. The results of meta-analyses combining these tests are shown in Figure 9.

FIGURE 9.

Combined 75-g and 100-g OGTT fasting glucose, 1-hour glucose levels, 2-hour glucose levels and adverse outcomes.

Combined results (75-g and 100-g OGTT) are similar to those for the 75-g OGTT alone. For fasting glucose there are statistically significant increases in risk of C-section, LGA, macrosomia, pre-eclampsia, neonatal hypoglycaemia, induction (of labour) and shoulder dystocia. Glucose levels do not appear to be associated with preterm birth or assisted (instrumental) delivery. The increase in odds can be substantial, with a more than doubling in odds per 1-mmol/l increase in fasting glucose level for pre-eclampsia and for LGA.

The results for the 2-hour post-load glucose levels are weaker than fasting associations, although CIs are narrower, particularly for preterm birth and pre-eclampsia, The association of increasing 2-hour post-load glucose levels with C-section just misses significance (OR 1.10, 95% CI 0.96 to 1.25).

The associations are weaker when the 75-g and 100-g test results are combined compared with the 75-g test results alone, reflecting the weaker associations when the 100-g test is administered.

Testing the linearity assumption

We viewed the shape of the associations between the log odds of outcomes and increasing glucose levels or categories for fasting and 2-hour OGTT (combining 75-g and 100-g tests) and for the 1-hour OGCT, for each outcome, these associations appeared generally linear. We tested the assumption of a linear association by including a squared term for the glucose levels in the regression models (see Methods). If the associations were quadratic curvilinear (rather than linear) we would expect there to be a statistically significant association between outcome risk and the square of the glucose level. The results of this analysis for all outcomes for the fasting and 2-hour 75-g OGTT and the OGCT are presented in Appendix 2, Table 61. Data were too limited to repeat this analysis for the 100-g OGTT alone.

The small number of studies limits the ability to draw conclusions; however, for the majority of outcomes, the association between outcome and the square of glucose was not statistically significant – it is therefore reasonable that, given the visual evidence (see Appendix 2, Figures 52–60), the association between glucose levels and outcomes is linear. There were, however, statistically significant curvilinear associations between fasting and 2-hour post-load glucose levels (75-g OGTT) and 1-hour post-load glucose level (50-g OGCT) and PIH/pre-eclampsia, but two of these associations were negative, suggesting that the odds of pre-eclampsia ‘levels off’ slightly at higher glucose levels rather than continuing to increase, and one association (fasting glucose, 75-g OGTT) was positive. This inconsistency in direction suggests that these may be chance findings, and therefore caution is advised when considering these results. There was a positive association with fasting glucose (75-g OGTT) and neonatal hypoglycaemia and preterm birth, but few studies were included, and, again, these results should be interpreted with caution (see Appendix 2, Table 61).

Analyses of adjusted odds ratios

Of the included studies, 106,61,64,66,68–70,72,80,82 reported ORs for outcomes adjusted for potential confounding factors such as maternal BMI and previous GDM. This section presents these ORs relative to the baseline glucose category for macrosomia64,70,80 LGA6,61,64,72,82 C-section6,68,80,82 and pre-eclampsia66,68,69 which were the only outcomes reported across studies.

Appendix 2, Figure 62, shows the results for macrosomia; Figure 63, LGA; Figure 64, C-section; and Figure 65, pre-eclampsia. The limited data make drawing conclusions difficult; however, there seems to be a general trend of increasing odds (when results are adjusted for selected potentially confounding variables) of each adverse perinatal outcome with each 1-mmol/l increase in glucose level. Compared with post-load glucose level, fasting is more strongly associated with odds of a perinatal outcome. For example the odds of LGA doubles with each 1-mmol/l increase in fasting glucose level, whereas the odds of LGA doubles with each 3 mmol/l increase in 2-hour post-load glucose level.

Meta-analysis of studies with two oral glucose challenge test or oral glucose tolerance test categories

Fifteen studies reported associations between two glucose categories and adverse perinatal outcomes. The characteristics of these studies are presented in Table 8. Ten of these studies examined associations between glucose levels following a 50-g OGCT and perinatal outcomes. Outcomes were compared between those who did not meet the criteria for having an OGTT to those who did meet those criteria, but who were not diagnosed with GDM, that is, two groups were compared [< 130 mg/dl vs. ≥ 130 mg/dl post challenge glucose level or (for some studies) < 140 mg/dl vs. ≥ 140 mg/dl post challenge glucose level]. Three studies compared lower levels of glucose with higher levels 2 hours after a 75-g or 100-g OGTT, whereas two other studies compared women with no elevated glucose levels at any time following an OGTT, with women with one elevated glucose level. For all of these comparisons, the numbers of outcomes in the two groups were used to calculate ORs for outcomes comparing one group with a perceived lower risk with another group with a perceived higher risk.

The results of these meta-analyses are shown in Figures 10 and 11. Women in the group with higher glucose levels following an OGCT have a statistically significant increased risk of C-section, LGA, macrosomia and pre-eclampsia than women with lower glucose levels. Other outcomes were reported in only one or two studies; therefore, these associations are less clear (see Figure 10).

FIGURE 10.

Meta-analysis for ORs of outcomes comparing those with OGCT negative results with those with OGCT positive results.

FIGURE 11.

Meta-analysis for ORs of outcomes comparing those with one OGTT elevated glucose level with those with no elevated OGTT glucose levels.

Results from studies comparing lower levels of glucose with higher levels at 2 hours following a 75-g or 100-g OGTT, and results from studies comparing women with no elevated glucose levels at any time after an OGTT to women with one elevated glucose level, were not pooled in meta-analyses. This was because of the differences in the glucose tests used and the timings of the glucose measurements. Generally, however, there was a suggestion that women with one elevated glucose level at OGTT are at higher risk of C-section and macrosomia than women with no elevated glucose levels, although the CIs are wide and often include the null value. There was no evidence of a difference between groups in the odds of preterm birth.

Studies of longer-term and anthropometric outcomes

Six studies22,36,41,65,108,109 reported longer-term and/or anthropometric outcomes, either measures of adiposity or incidence of diabetes. The characteristics of these studies are presented in Table 9.

Three studies7,36,65 (see Chapter 2) reported neonatal obesity (skinfold thickness and percentage body fat of > 90th centile), one study41 reported obesity at age 2 year and one study108 reported obesity at age 5–7 years. One study,109 in Pima Indians, reported longer-term incidence of diabetes in offspring.

The associations of glucose levels with each outcome for each study are shown in Figure 12. Data from one study115 were presented only as ORs of association, so this study could not be included in this figure.

FIGURE 12.

Associations between glucose level and longer-term and anthropometric outcomes: risk of morbidity. Glucose level measured at (a) fasting; (b) 1 hour; and (c) 2 hours post load. Colour indicates the category of morbidity. ● = results for 50-g OGCT; ▲ = results for 75-g OGTT.

Data are too few to perform a meta-analysis; however, the HAPO36 (2009) model II results (model II, adjusted for age, BMI, BMI2, height, mean arterial BP, gestational age at OGTT, smoking, alcohol use, hospitalisation prior to delivery, and any family history of diabetes) suggest a strong association between glucose levels and sum of skinfolds (flank, triceps and subscapular) of > 90th centile (fasting OR per SD 1.39, 95% CI, 1.33 to 1.47; 1-hour post-load glucose level OR 1.42, 95% CI 1.35 to1.49); 2-hour post-load glucose level OR 1.36, 95% CI 1.30 to 1.43) and body fat of > 90th centile (fasting OR 1.35, 95% CI 1.28 to 1.42; 1-hour post-load glucose level OR 1.44, 95% CI 1.37 to 1.52; 2-hour post-load glucose level OR 1.35, 95% CI 1.29 to 1.42). Aris et al. 65 and the BiB study7 (see Chapter 2) report a linear association between skinfold thickness or body fat percentage of > 90th percentile with increasing glucose levels (fasting and 2-hour post load).

There is limited evidence of an association between increasing maternal glucose levels and longer-term child obesity and overweight. Hillier et al. 108 report the risk of childhood obesity and overweight is greater for infants of women with the highest OGCT glucose levels than for those with the lowest, and that risk is greater for infants of women with GDM than infants of women without GDM (child weight > 95th centile, OR 1.82, 95% CI 1.15 to 2.88). Pettitt et al. 41 found no evidence of an association between glucose levels and overweight and obesity at the age of 2 years. 41

Pettitt et al. 109 reported a strong association between increasing maternal glucose level and increased risk of offspring diabetes, but this was in an unusual population (Pima Indians) therefore the results may not generalise to European populations.

Other identified studies not included in the meta-analyses

Several studies could not be included in the analyses: one study110 reported the association between fasting, 1-hour and 2-hour OGTT glucose levels and outcomes only as an OR per SD. The study110 reports statistically significant increases in risk of LGA, C-section, preterm birth, shoulder dystocia and gestational hypertension with increasing glucose levels.

One study111 examined the incidence of type 2 diabetes in Pima Indians (similarly to Pettitt et al. ,109 and possibly for the same cohort, so is excluded) and concluded that diabetes incidence in offspring increases as maternal glucose levels increase, the 2-hour post-load OGTT glucose level was used, diabetes was recorded at ages from 10 to 25 years.

One study71 reported preliminary results, so it was excluded because it was superseded by a later publication included in the analyses. 112 The earlier study71 reported 2-hour post-load 75-g OGTT glucose levels and incidence of LGA, shoulder dystocia, C-section and preterm birth.

One study114 reported a significant association between the 2-hour post-load glucose level following a 75-g non-fasted OGCT and macrosomia. The study114 was excluded because the 75-g OGCT is a glucose load not normally used in a non-fasted state and there were no other studies using this test in this way.

One study113 compared the risk of outcomes in women without elevated glucose levels to those with a single elevated glucose level at either 1, 2 or 3 hours following a 100-g OGTT. The study113 reported that, compared with women without an elevated glucose level, women with one elevated glucose level were at higher risk of adverse outcomes including C-section, pre-eclampsia and neonatal hypoglycaemia. The associations were stronger for women with an elevated glucose level at 1-hour post-load compared with 2 or 3 hours. There were no women with an elevated fasting glucose level and normal post-load glucose level.

Studies examining the association between three or more graded increases in glucose level and risk of perinatal adverse outcomes

There was a positive association between LGA, macrosomia, C-section plus pre-eclampsia and increasing glucose levels. There was evidence for an increase in risk in shoulder dystocia, neonatal hypoglycaemia, instrumental (assisted) birth and induced labour, although data were more limited for these outcomes. There was no evidence of an association between preterm birth and increasing maternal glucose levels. The associations, in terms of OR per 1-mmol/l increase in glucose level, were stronger for fasting glucose levels than post-load glucose levels.

Fewer studies examining glucose levels and adverse outcomes used the 100-g OGTT than the 75-g OGTT; however, our findings suggest that associations are similar for the two tests. Equally, the post-load glucose results for the 1-hour 50-g OGCT were broadly consistent with the post-load glucose results of the 75-g and 100-g OGTT. We combined the fasting results from the 75-g and 100-g OGTT together and the results from the 2-hour post-load 75-g and 100-g OGTT. Combining results for fasting glucose measurements is reasonable, because fasting glucose levels will not be affected by subsequent glucose load. The validity of combining results for post-load glucose level is more uncertain, as it assumes that the association of risk between adverse outcomes and glucose level is independent of the test’s glucose load.

Our analysis generally suggests the odds of an adverse outcome increases linearly with glucose levels. Therefore, there appears to be a continuum of risk across glucose levels, with no sudden increase in risk at any glucose level, suggesting that there is no glucose level below which there is no increased risk, and no clear glucose threshold that can distinguish between women at low risk from those at a substantially increased risk of having an adverse outcome.

We excluded women from our analysis with diagnosed GDM, however defined, because these women would have been offered treatment and treatment to reduce hyperglycaemia will influence the natural association between glucose level and outcome risk (see Chapter 2).

Most of the analyses were based on unadjusted raw numbers of women with outcomes in different categories. The risks may therefore be over or underestimated because of potential confounding. For example, the risk of macrosomia is recognised as being higher for obese women than ‘normal’ weight women irrespective of glucose levels. 116 In the few studies presenting adjusted analyses to correct for possible confounding, there was evidence that risk of adverse outcomes increased (similarly to unadjusted analyses) as glucose levels increased, and we have also shown this using the BiB study data,22 as described in Chapter 2. However, we cannot rule out confounding by indication or residual confounding by characteristics that were unmeasured or poorly measured in studies.

Studies examining the association between graded increases in glucose level and risk of longer-term adverse outcomes

There were few studies investigating longer-term outcomes for the mother or infant. Two studies41,108 examined glucose levels and risk of offspring obesity and reported variable results. One study109 reported an association with diabetes in childhood; however, the women were Pima Indians, who are at greater risk of diabetes, and therefore these results may not generalise to a European population.

Studies examining the association between two categories of glucose level and risk of adverse outcomes

Several studies present glucose levels at OGCT divided into two categories (typically women considered to have OGCT positive results vs. OGCT negative results). Our analyses showed that elevated OGCT levels (those indicating an OGTT should be offered) compared with ‘normal’ OGCT results were associated with an increased risk of C-section, LGA, macrosomia and pre-eclampsia, suggesting that this test does identify women at increased risk of these adverse outcomes. A limited number of studies presented results for the OGTT with just two glucose groups, but data were too few to draw any firm conclusions.

Strengths and limitations

We identified a large number of high-quality studies examining associations between maternal glucose levels and adverse perinatal outcomes; most studies included > 1000 women. Studies tended to report similar adverse outcomes and therefore we were able to pool estimates in meta-analyses. We were able to examine (because they were available and reported) a variety of outcomes including macrosomia, LGA, C-section, pre-eclampsia, neonatal hypoglycaemia and shoulder dystocia, across the whole glucose spectrum.

Included studies were necessarily observational (randomisation to a given glucose level is not possible), therefore it is conceivable that other characteristics might have confounded the associations that we have examined and may be responsible for the results we present.

Unfortunately, five studies71,110,111,113,114 did not include sufficient data or information to allow inclusion in the meta-analyses and therefore our estimates may have been different if these studies were able to be included. However, as most studies suggest a positive relationship between increasing glucose level and adverse outcome, the inclusion of these studies would be unlikely to change results.

Five studies36,41,65,108,109 examined maternal glucose level and longer-term outcomes, although they investigated diabetes and adiposity, the timing of the measurements and methods of assessment varied, preventing pooling of estimates. Data for the 100-g OGTT, and particularly the 1-hour post-load measure, were more limited than the 75-g OGTT, therefore less confidence should be placed on these results.

Conclusion

Our meta-analyses suggest an increasing risk for the majority of reported adverse perinatal outcomes including macrosomia, LGA, C-section, pre-eclampsia, neonatal hypoglycaemia and shoulder dystocia, across the whole spectrum of glucose levels. Associations between risk of an outcome and graded increases in glucose level seem to apply to all glucose loads (50-g, 75-g and 100-g) and at all measurement times (fasting, and 1-hour and 2-hour post load), although the strength of these associations varies. Associations were stronger for fasting glucose levels than post-load glucose levels and for the 75-g OGTT compared with the 100-g OGTT.

Search strategy

Searches were undertaken in July 2014 in MEDLINE, MEDLINE In-Process & Other Non-Indexed Citations, EMBASE, the Maternity and Infant Care database and CENTRAL. No date restrictions were applied to the searches; citations were restricted to English language only (see Appendix 7, Table 84).

Title and abstract screening and full-text screening were performed in duplicate by two reviewers with disagreements resolved by consensus or by a third reviewer.

Three cohort studies were eligible and provided data at the individual participant level:

• the BiB study22 (John Wright, Bradford Institute for Health Research, September 2013)

• the Atlantic DIP study59 from the Irish Atlantic seaboard (Fidelma Dunne, Department of Medicine, National University of Ireland, September 2013)

• the Warwick/Coventry cohort,60 unpublished data from Warwick Hospital, George Eliot Hospital, Nuneaton and University Hospital Coventry (Ponnusamy Saravanan, Warwick Medical School, September 2013).

Because IPD were available from an Irish cohort (Atlantic DIP59), we have considered prevalence of GDM in the UK and Ireland together.

Inclusion/exclusion criteria

This review sought to identify all cohorts of pregnant women in whole, or in part, in the UK or Republic of Ireland who were assessed for GDM.

The included studies had to have the following characteristics.

Quality assessment

We assessed the characteristics of all of the publication/study criteria (including the population, location and publication year) that were used to diagnose GDM and derive prevalence estimates.

Data extraction

The following data were extracted from each publication:

• year of publication

• location of the study

• details of the population characteristics, for example ethnicity, age, BMI distribution (if reported)

• details of the OGTT methods and diagnostic criteria used

• total number of women with and without GDM, or the prevalence of GDM

• prevalence of GDM in participant subgroups, such as ethnic group or BMI group.

For the IPD, the prevalence of GDM was calculated, based on the reported OGTT glucose measurements, with GDM diagnosed according to a range of diagnostic criteria as described earlier in Table 1. Prevalence was also calculated by ethnic group (white, SA or ‘Other’) and by age categories using the modified WHO 1999 criteria11 (fasting glucose level of ≥ 6.1 mmol/l and 2-hour post-load glucose level of ≥ 7.8 mmol/l).

Synthesis methods

Prevalences of GDM, along with their 95% CI, were estimated from the data for each study. These prevalence estimates are shown on forest plots. Studies were categorised by GDM diagnostic criteria and year of publication, in order to investigate the effect of these factors (see Figures 14 and 15).

Meta-analyses of the prevalence data were considered, but not performed because of the heterogeneity across the studies, particularly the diversity of diagnostic criteria used to diagnose GDM.

Quality assessment and included studies

A summary of GDM diagnostic criteria are presented in the introduction to this report (see Table 1). Table 10 summarises the 10 published studies42,44,118,127–129,131–133 and the three IPD cohorts included in this review.

Prevalence of gestational diabetes mellitus by year the study was undertaken and gestational diabetes mellitus criteria used

Figure 14 shows prevalence by year and GDM criteria used by each study. Using data from the three IPD cohorts we calculated GDM prevalence according to the most commonly used GDM diagnostic criteria presented in Table 1; 1-hour post-load glucose levels (75-g OGTT) were not available for the BiB,22 Atlantic DIP59 and Warwick/Coventry cohorts,60 therefore prevalences may be underestimated for criteria that include a 1-hour glucose level [American Diabetes Association (ADA), IADPSG, NDDG (National Diabetes Data Group)]. These prevalence estimates are shown in Figure 15. The Atlantic DIP study59 has higher prevalence estimates for all diagnostic criteria. NDDG criteria are the most conservative, having the highest glucose thresholds. The WHO 1980, WHO 1999, ADA and Australasian Diabetes in Pregnancy Society (ADIPS) criteria produce similar prevalence estimates, despite their different glucose threshold criteria. The IADPSG criteria give the highest prevalence estimates for the IPD cohorts, similarly to published estimates, as a result of the lower fasting glucose threshold.

FIGURE 14.

Prevalence of GDM by year the study was undertaken and GDM criteria used. DPSG (EASD), Diabetic Pregnancy Study Group (of the European Association for the Study of Diabetes).

FIGURE 15.

Estimated prevalence according to different GDM criteria in the IPD cohorts. See Table 1 for criteria thresholds.

Prevalence of gestational diabetes mellitus by ethnicity

Two published studies44,118 report prevalence of GDM by ethnicity (Table 11). Both of these studies44,118 were undertaken when recommended criteria thresholds were higher (1992 and 1995) than those now suggested by the IADPSG and consequently report lower GDM prevalence than would be expected today. Both studies,44,118 however, report differing GDM prevalence by ethnicity, with women of Asian and SA origin having the highest rates. Koukkou et al. 118 do not provide more information on the origin of the Asian women in their study (they were recruited from an inner-city London hospital), so they could be of any number of Asian ethnicities.

Prevalence of GDM by ethnicity was calculated using the three IPD cohorts. These data are summarised in Table 12.

Prevalence of gestational diabetes mellitus by age

The published studies provided insufficient data to estimate prevalence by age, but we have been able to calculate estimates using the IPD cohorts. The results are summarised in Table 13. GDM prevalence appears to increase as age category increases in all three cohorts. A logistic regression confirmed this, with a statistically significant increase in odds of GDM of 1.08, 95% CI 1.08 to 1.10 per year.

Prevalence of gestational diabetes mellitus by timing of oral glucose tolerance test

The BiB22 IPD included information on the timing of the OGTT, in terms of gestational age.

We examined results (numbers in parenthesis) by the following gestational age categories (in weeks plus days): < 25 (438), 25–25 plus 6 days (1733), 26–26 plus 6 days (5695), 27–27 plus 6 days (1133), 28–28 plus 6 days (529), 29–29 plus 6 days (276), 30–30 plus 6 days (263) and ≥ 31 (364). A logistic regression analysis found no evidence that the prevalence of GDM changed according to the timing of the test (OR 1.00, 95% CI 0.96 to 1.04).

Strengths and limitations

We identified 13 studies22,42,44,59,60,118,126–129,131–133 undertaken over 25 years in varied areas of England and Ireland. We were able to demonstrate how prevalence changed over these 25 years and how participant characteristics and criteria influence prevalence. The studies were large, all included > 1000 women and all reported their inclusion and GDM criteria. Our IPD provided valuable information that was not available from published estimates, and showed that, even in contemporary cohorts, GDM prevalence can vary considerably between groups with varying maternal characteristics, including ethnicity.

Few published studies included populations at high risk of GDM because of their ethnicity therefore the inclusion of the BiB cohort22 is extremely valuable. Estimates of prevalence for SA women in the Atlantic DIP cohort59 are uncertain because there were few women of SA ethnicity in that cohort, and even fewer with diagnosed GDM. We undertook several subgroup comparisons; however, these results should be interpreted cautiously given that the studies were not designed or powered to detect differences in prevalence across subgroups. The prevalence of GDM in WB women in the BiB study22 is lower (5%) than that in the Atlantic DIP59 (9%) and the Warwick/Coventry60 (8%) cohorts, even although all used the same diagnostic criteria. The Atlantic DIP study59 (like the BiB study22) universally offered an OGTT, whereas the Warwick/Coventry studies60 selectively tested their population, but both cohorts59,60 had similar and higher GDM prevalence in their white populations than the BiB study,22 and it is unclear why this is so.

We did not identify any eligible studies that included Scottish or Welsh cohorts, therefore, although we intended to present data on UK prevalence of GDM, our data represent England, Northern Ireland and the Republic of Ireland (as we were able to include the Atlantic DIP cohort59).

Conclusions

The prevalence of GDM is increasing in the UK; the offer of an OGTT to all women, the lowering of diagnostic thresholds, and increases in the proportion of women at risk, either because of their ethnicity or increasing weight or age, are all contributing factors (which is examined in the Chapter 5). Within a narrow gestational time frame we have demonstrated that timing of OGTT does not seem to influence prevalence; however, we had few women tested at < 25 or > 31 weeks of gestation and therefore caution should be taken when interpreting these findings. We showed that populations of older women or women whose ethnicity conveys a high risk of diabetes will have higher GDM prevalence.

Screening options

There are two ‘screening’ methods generally used to identify women who are at increased risk of developing GDM; (1) the 50-g OGCT, which is similar to the OGTT, but does not require an overnight fast: one plasma glucose level is obtained 1 hour following the consumption of a 50-g glucose drink – women with a positive test (above a predefined glucose level) are offered an OGTT; and (2) maternal characteristics/risk factor assessment, which involves the assessment of maternal characteristics to identify increased risk of GDM: family history of diabetes; being of an ethnicity with a high prevalence of diabetes; previous history of having a macrosomic infant or GDM; or BMI of ≥ 30 kg/m² are risk factors recommended for use by NICE18 – when one or more risk factors are identified then NICE recommends that an OGTT is offered.

Diagnostic testing

Gestational diabetes mellitus is generally diagnosed using an OGTT. The OGTT is normally conducted in the morning following an overnight fast. A baseline plasma glucose sample is obtained, the woman then consumes a drink containing, typically, 75 g or 100 g of glucose, and then at hourly intervals plasma glucose is measured. The frequency of measurement depends on the glucose load and local policy. Women with an ‘elevated’ glucose level at one or more measurements are classified as having GDM.

There are some limitations to the OGTT as a diagnostic test: (1) a negative OGTT result does not mean a woman will not develop GDM later in pregnancy – because, as gestation progresses, insulin resistance may increase, repeat glucose testing therefore may be required; (2) the linear positive graded association across the whole spectrum of maternal glucose and risk of adverse outcomes has made the identification of clear diagnostic glucose level thresholds for GDM difficult;6,7 and (3) the reproducibility of the OGTT is around only 75%. 20,21

The accuracy of a screening or diagnostic test relates to its ability to distinguish between those with the condition (GDM) and those who do not have the condition; this can be presented in terms of a test’s sensitivity and specificity, predictive values, likelihood ratios, and the area under the receiver operating characteristic (ROC) curve.

Several health-care agencies, including NICE and the ADIPS (see Table 14), have recommended that pregnant women should have their risk of GDM evaluated by assessment of maternal characteristics/risk factors. Those with one or more maternal characteristics/risk factors should be offered a diagnostic OGTT (or alternative test: see Table 14, ADA entry).

Table 14 shows a selection of the risk factors recommended for use to guide diagnostic testing.

Recommended risk factors vary considerably; however, all strategies suggest that early pregnancy (first trimester) screening and testing for GDM or previously undiagnosed type 2 diabetes in those women classified as being at particularly high risk. NICE recommends that first trimester testing is offered to women who have had previous GDM, whereas the ADA and ADIPS recommend early testing of women with a variety of risk factors. Of these three institutions, the ADA recommends that all of those not identified as having GDM (or previously undiagnosed diabetes) in the first trimester should be tested in the third trimester. 136 NICE and ADIPS recommend universal risk factor screening and selective OGTT in the third trimester. 9,18

Methods

Two cohorts with IPD were eligible and agreed to share their data:

• The BiB cohort (John Wright, Bradford Institute for Health Research, September 2013) At the time of recruitment to the BiB study,22 all women planning to give birth at the Bradford Royal Infirmary were offered a 75-g OGTT (irrespective of risk factors). The WHO 199911 (modified) criteria were used to diagnose GDM (fasting glucose level of ≥ 6.1 mmol/l, 2-hour post-load glucose level of ≥ 7.8 mmol/l).

• The Atlantic DIP Cohort59 (Fidelma Dunne, Department of Medicine, National University of Ireland, September 2013) Women at participating hospitals in the south-west of Ireland were all offered a 75-g OGTT (irrespective of risk factors) and, as with the BiB study,22 the WHO 1999 (modified) criteria11 were used to diagnose GDM (fasting glucose level of ≥ 6.1 mmol/l, 2-hour post-load glucose level of ≥ 7.8 mmol/l).

An OGTT was offered to all women in both cohorts. Uptake of the offer varied between the two cohorts [63% (the BiB study22) vs. 58% (the Atlantic DIP study59)]. 124,137

We have examined the following characteristics because they are associated with a greater risk of GDM development, and their use as indicators for OGTT is recommended by institutions including NICE18 and the ADA. 136

• age

• obesity, measured by BMI

• parity (multiparous vs. primiparous)

• ethnicity (white, SA or other)

• family history of diabetes

• GDM in previous pregnancy

• macrosomic baby (≥ 4 kg) in previous pregnancy.

Age was examined yearly from 20 to 40 years and BMI at every 1.0-kg/m² unit increase from 15.0 to 40.0 kg/m². For age and BMI combined, however, we present results for age ≥ 25 years and ≥ 30 years and BMI of ≥ 25 kg/m² and ≥ 30 kg/m².

Ethnicity was coded as white, SA or other [the majority of women were either of white European (in the BiB22 and Atlantic DIP59 studies) or SA ethnicity (the BiB22 study)] and parity was coded as primiparous (first pregnancy) or multiparous (second or subsequent pregnancy).

Statistical analyses

For each risk factor the following were calculated with their SEs and 95% CIs:

• Sensitivity:

  • The proportion of women with GDM who had the risk factor (i.e. proportion of GDM cases correctly identified by the test).

• Specificity:

  • The proportion of women without GDM who did not have the risk factor.

• Positive rate:

  • The proportion of women with the risk factor (i.e. proportion who would be offered an OGTT).

Most existing GDM screening guidelines recommend offering an OGTT to any woman who has at least one risk factor from a set of risk factors (see Table 14). To investigate the screening potential of this approach we considered the risk factors in the list above, with age ≥ 25 years or ≥ 30 years, and ≥ BMI 25 kg/m² or ≥ 30 kg/m². For each of the 287 possible combinations of these risk factors we calculated whether or not each woman had at least one of the risk factors, and then estimated the sensitivity, specificity and positive rate associated with having one or more risk factors.

From this set of 287 possible combinations of risk factors we removed those that were ‘dominated’ by others. A screening test is dominated if there is at least one other ‘test’ with both higher sensitivity and specificity, which would be preferred to the dominated test. Sensitivity and positive rate for the remaining non-dominated tests were plotted in ROC space.

We examined screening based on a predicted risk of GDM, similar to screening strategies used to identify those at risk of cardiovascular disease. 138 A logistic regression model was fitted to the data from both cohorts, regressing GDM incidence against the risk factors. The resulting log ORs from this regression model were used to calculate a predicted risk of GDM for each woman in the data set. The sensitivity and positive rate for predicting GDM at each percentage point of risk from 1% to 80% was calculated and plotted in ROC space. The same analyses were conducted on the separate and pooled data sets for comparison.

Results

Risk factors as predictors for gestational diabetes mellitus

We next consider risk factor screening, by which women are offered an OGTT if they have at least one positive maternal characteristic/risk factor among a set. The results for these analyses are provided in Figure 16, which shows the percentage of GDM cases identified (sensitivity) against the percentage of women offered an OGTT (positive rate) for all of the sets of risk factors that were not ‘dominated’ by others.

FIGURE 16.

Screening performance of one or more risk factor for identifying GDM. Colour indicates the number of risk factors in each set. ● = results for the Atlantic DIP study;59 ▲ = results for the BiB study. 22

Figure 16 shows that in order to identify 80% of women with GDM (80% sensitivity) then approximately 60% of women would have to be offered an OGTT (40% specificity). To identify 90% of women with GDM, about 70% of women would need to be offered an OGTT, and for 95%, 80% of women need to be offered an OGTT. Therefore, most women would need to be tested to identify the majority of women with GDM using risk factors to identify those at higher risk. Figure 16 shows that strategies that identify one risk factor or one out of two risk factors seem to detect < 60% of GDM cases. To detect ≥ 90% of GDM cases requires that at least three or four risk factors are considered (in Figure 16, ‘Number of risk factors’ refers to the following: 1 = one risk factor; 2 = at least one risk factor out of two; 3 = at least one risk factor out of three; and 4 = at least one risk factor out of four).

Both cohorts have similar estimates of sensitivity and specificity across the range of possible risk factor screening strategies, that is, the points in Figure 16 all lie on approximately the same curve for both cohorts. Importantly, however, the risk factors used to achieve, for example, a sensitivity of 90% differs between cohorts. Table 16 provides examples of risk factors, included in screening strategies, with sensitivity of between 90% and 95% (so they detect almost all cases of GDM) for the two cohorts, separately and combined. Age and BMI are the most commonly occurring risk factors, with family history of diabetes, prior GDM and non-white ethnicity also common. In the BiB cohort22 offering an OGTT to anyone either aged ≥ 25 years or with a BMI of ≥ 30 kg/m² detects 92% of all GDM cases (because the majority of women in the BiB cohort are aged > 25 years of age or have a BMI of > 30 kg/m²), including other factors, does not substantially increase the detection rate, but there are few women left to test and only an 8% increase was needed to achieve a 100% detection rate.

TABLE 16 - Performance of age and BMI categories for the identification of GDM

Risk factors included	Sensitivity	Specificity	Positive rate
BiB cohort22
Aged ≥ 25 years, BMI ≥ 30 kg/m2	90.4	28.7	72.7
Aged ≥ 25 years, BMI ≥ 30 kg/m2, prior GDM	90.4	28.6	72.8
Aged ≥ 25 years, BMI ≥ 30 kg/m2, diabetes	91.6	23.2	77.7
Aged ≥ 25 years, BMI ≥ 30 kg/m2, diabetes, prior GDM	91.6	23.1	77.7
Aged ≥ 30 years, BMI ≥ 30 kg/m2, non-white	94.3	21.3	79.8
Aged ≥ 30 years, BMI ≥ 30 kg/m2, non-white, prior GDM	94.3	21.3	79.9
Aged ≥ 25 years, BMI ≥ 25 kg/m2, diabetes	94.4	16.9	83.8
Aged ≥ 25 years, BMI ≥ 25 kg/m2, diabetes, prior GDM	90.4	28.7	72.7
Atlantic DIP cohort59
BMI ≥ 25 kg/m2, non-white	90.1	36.8	66.0
Aged ≥ 30 years, BMI ≥ 30 kg/m2	90.8	28.6	73.4
Aged ≥ 30 years, BMI ≥ 30 kg/m2, non-white	93.9	26.0	76.0
Cohorts combined
Aged ≥ 30 years, BMI ≥ 30 kg/m2, diabetes	90.0	24.6	76.4
Aged ≥ 30 years, BMI ≥ 25 kg/m2, diabetes, prior GDM	90.3	24.6	76.5
BMI ≥ 25 kg/m2, non-white	92.0	24.0	77.3
BMI ≥ 25 kg/m2, non-white, prior GDM	92.1	24.0	77.3
Aged ≥ 25 years, BMI ≥ 30 kg/m2	93.2	23.3	78.0
Aged ≥ 25 years, BMI ≥ 30 kg/m2, prior GDM	93.2	23.3	78.1
Aged ≥ 30 years, BMI ≥ 30 kg/m2, non-white	94.1	22.7	78.7
Aged ≥ 30 years, BMI ≥ 30 kg/m2, non-white, prior GDM	94.1	22.7	78.7
Aged ≥ 25 years, BMI ≥ 25 kg/m2	95.9	16.5	84.5
Aged ≥ 25 years, BMI ≥ 25 kg/m2 prior GDM	95.9	16.5	84.5

Risk prediction models

The association between each risk factor and GDM, in terms of the OR, is shown in Table 17. All risk factors, except multiparity, were statistically significantly associated with GDM. The results were generally consistent across the two cohorts. Having GDM in a previous pregnancy is the most dominant risk factor, associated with a five-fold increase in the odds of GDM development in the current pregnancy. Non-white ethnicity is also a strong indicator of risk. Multiparity was associated with lower risk (see Table 17).

TABLE 17 - Odds ratio for the association between risk factors and GDM

Risk factor	BiB22		Atlantic DIP59
	OR	95% CI	OR	95% CI
Age (per year)	1.09	1.08 to 1.1	1.10	1.07 to 1.12
BMI (per kg/m2)	1.06	1.05 to 1.08	1.13	1.11 to 1.15
Ethnicity (non-white)	2.32	1.90 to 2.83	5.16	3.85 to 6.91
Multiparity	0.89	0.73 to 1.08	0.74	0.58 to 0.96
Family history of diabetes	1.36	1.14 to 1.63	1.42	1.17 to 1.80
Previous macrosomia	1.54	1.12 to 2.13	–	–
Previous GDM	5.90	3.78 to 9.22	–	–

The ROC curve of sensitivity against positive rate using predicted risk to screen for GDM is shown in Figure 17 for both the BiB22 and Atlantic DIP59 cohorts. Both cohorts provide similar results. To detect 90% of GDM cases based on a risk model requires that 70% of pregnant women undergo an OGTT.

FIGURE 17.

Sensitivity and positive rate when using a risk prediction model to predict GDM.

Figure 18 compares the screening performance of the risk prediction model with screening performance based on age alone and based on oral glucose tolerance testing of anyone with at least one risk factor using data from the BiB cohort22 only. Figure 18 shows that using a risk prediction model to screen for GDM generally provides improved performance compared with screening based on counting positive risk factors alone. This is because the sensitivity is generally higher at all specificities and positive rates, so the number of women who would be offered an OGTT could be reduced while detecting the same number of GDM cases. However, at high specificities (where most GDM cases are detected) there is little performance difference between using a risk prediction model and counting risk factors.

FIGURE 18.

Screening performance using risk prediction compared with having one positive risk factor, or using age alone.

Methods

Results

Included studies

The database searches identified 5867 citations (3140 after deduplication). After title and abstract screening 181 publications were retrieved for full-text screening; 47 of these were excluded (see Appendix 4, Table 65). Ninety-seven studies reported associations between risk factors and GDM incidence, but did not consider multiple risk factor screening, and eight studies examined the effect of screening based on a single risk factor (although some studies reported more than one risk factor, they were not considered in combination). Because the analysis in the first section of this chapter suggests that single risk factor screening is not the most efficient strategy, these studies have not been included and they are not considered further. Five publications were identified through reference checking of related reviews and eight from other searches conducted for this report.

One hundred and thirty-four studies reported the association between maternal characteristics/risk factors and GDM, and 29 of these reported data on risk factors, 24 of which had sufficient data to allow inclusion in the analyses. Details of the identification process are presented in Figure 19.

FIGURE 19.

The search process.

Studies of multiple risk factor screening

Of the 29 included studies,93,122,139–165 24 provided sufficient data for screening performance (sensitivity, specificity) to be calculated; these 24 studies are summarised in Table 18. The studies were conducted in a variety of countries and used different criteria for diagnosing GDM; therefore different studies will produce different GDM prevalences. Six studies93,122,139,143,146,157 assessed the screening performance of existing guideline recommendations [NICE, ADA, American College of Obstetricians and Gynecologists (ACOG), ADIPS, Irish, French]. Seven studies93,122,139,142,143,146,157 counted the number of risk factors for each woman. Six studies140,141,149,151,155,164 used a risk prediction model or a risk score to determine the results of risk factor screening and five studies148,150,152,158,165 examined various risk factors.

TABLE 18 - Characteristics of included multiple risk factor studies

C&C, Carpenter and Coustan; GCT, glucose challenge test.

A total of 5235 women were included in the study; 2992 had an OGTT performed. DPSG, Diabetic Pregnancy Study Group.

Study	Year	Country	GDM diagnosis criterion	Total women	No. with GDM	Risk factor screening strategy
Avalos122	2013	Ireland	IADPSG	5500	681	Irish guideline recommendations
Caliskan142	2004	Turkey	NDDG	422	14	Number of risk factors
Cosson143	2013	France	WHO	18,755	2710	French guideline recommendations
Cypryk144	2008	Poland	WHO	2180	510	Number of risk factors
Danilenko-Dixon146	1999	USA	NDDG	18,504	564	ADA guideline recommendations
Jensen145	2003	Denmark	DPSG	2992a	83	Number of risk factors
Jiménez-Moleón93	2002	Spain	NDDG	1962	65	ADA and ACOG guideline recommendations
Marquette147	1985	USA	C&C	434	12	Number of risk factors
Moses148	1998	Australia	ADIPS	2907	183	Age, BMI, ethnicity
Nanda149	2011	UK	WHO	11,464	297	Risk model
Naylor164	1997	US	NDDG or C&C	1571	69	Risk score
Ostlund150	2003	Sweden	WHO	3616	61	‘Traditional risk factors’
Phaloprakam151	2009	Thailand	C&C	469	127	Risk score
Pintaudi152	2014	Italy	IADPSG	1015	113	‘Standard risk factors’
Sacks153	1987	USA	ADA	4116	138	Number of risk factors
Savona-Ventura165	2013	Mediterranean	ADA	1368	119	Based on age, BMI and diastolic BP
Shamsuddin154	2001	Malaysia	OGTT levels reported	768	191	Number of risk factors
Shirazian155	2009	Iran	ADA	924	68	Risk score
Sunsaneevithayakul156	2003	Thailand	Not reported	9325	235	Number of risk factors
Teh157	2011	Australia	ADIPS	2426	250	NICE, ADA and ADIPS guideline recommendations
Van Leeuwen141 (A)	2010	Netherlands	OGTT/GCT levels reported	995	24	Risk model
Van Leeuwen140 (B)	2009	Netherlands	WHO	1266	47	Risk score
Williams158	1999	US	NDDG	25,118	148	Based on age, BMI, ethnicity, family history
Yang139	2002	China	WHO	9471	171	ADA guideline

Figure 20 shows the estimates of sensitivity and positive rate for the studies in Table 18 plotted against each other in ROC space. In Figure 20 (see also Figure 21) the shape of the points indicates the type of screening method used (● = existing guidelines; ▲ = counting numbers of risk factors; + = use of a risk prediction model or score; and ▪ = other methods).

FIGURE 20.

Screening performance (sensitivity and positive rate) for the included studies.

Figure 21 presents the results for studies93,122,139,143,146,157 reporting the performance of current screening guidelines. The vertical lines here show the 95% CIs for sensitivity and positive rate.

FIGURE 21.

Screening performance of existing risk factor screening guidelines.

Figure 22 shows the sensitivity and specificity for those studies140,141,149,151,155,164 evaluating risk prediction models or risk scores. Each study has multiple points because the studies reported results at various levels of risk. Results are reasonably consistent across studies, with all points lying approximately on a common ROC curve, suggesting that no specific risk scoring method is superior to another. Increasing sensitivity reduces specificity, for example to achieve a sensitivity of 80%, specificity is approximately 45%; to achieve a sensitivity of 90%, the specificity is approximately 35%. So to identify greater numbers of women with GDM requires offering an OGTT to increasing numbers.

FIGURE 22.

Screening performance of risk prediction or scoring models.

Discussion

Risk factor screening aims to identify as many ‘at risk’ women as possible so that diagnostic testing can be offered to those most likely to test positive, while preventing unnecessary testing in those least likely to test positive. A screening test with high sensitivity and specificity is therefore beneficial. Screening for GDM based on risk factors can take a variety of forms, including offering an OGTT to women with just one risk factor important in that population, or with one or more risk factors out of a set, which is the approach recommended by several institutions providing guidance, including NICE,19 or by calculating a predicted risk or risk score. This review identified 29 eligible studies, 93,122,139–165 which were methodologically diverse. We were unable to demonstrate superiority of any one strategy over another.

Certain risk factors are increasing in the pregnant population (e.g. obesity and advanced maternal age) therefore in the future it may be that the majority of women will have at least one risk factor and fulfil many criteria for diagnostic testing. Risk factor screening in such a population would require most women to be tested, but a proportion of women with hyperglycaemia/GDM would still be missed. Screening women for risk factors makes additional demands on consultation time and risk factors may not be recognised; however, even although they may be relatively easy to identify. Generally, the risk factor screening strategies examined in this review use only maternal characteristics; occasionally, however, as in the Nanda et al. study149 biochemical markers are included; these may increase screening complexity and costs, without necessarily improving case detection rates.

Search strategy

This review updates five existing systematic reviews of treatments for GDM: Alwan et al. (Cochrane review),168 Hartling et al. (Annals of Internal Medicine 2013),1 Horvath et al. (BMJ 2010),169 Falavigna (Diabetes, Research and Clinical Practice 2012)170 and Gui et al. (PLOS ONE). 171 The search strategies of all five reviews1,168–171 included randomised controlled trials (RCTs), one review also included observational studies which were ineligible in our review and therefore these studies were not considered. 1 One review compared the effects of metformin with insulin only,171 the four remaining reviews compared any treatment for GDM. It is likely that search strategies and assessment of eligibility differed between the reviews (these details are not published); however, the reviews seemed to take a broad approach to potential inclusion (RCTs, women with GDM and any treatments) and generally excluded and included the same trials, although there were slight variation when multiple publications of the same trial were identified.

The search strategies (see Appendix 7, Table 86) were designed to identify records of RCTs added to search sources since the most recent search date of the review by Alwan et al. 168 (July 2011). Strategies were developed using a combination of subject indexing terms and free-text search terms in the title and abstract fields, to identify relevant trials related to GDM and impaired glucose tolerance (IGT) in pregnancy. Where database functionality allowed, results were limited to records added to the database since 2011, using appropriate fields such as the entry date field in MEDLINE. Searches were first conducted in September 2013 and updated in October 2014 using the same search strategies. The databases searched were MEDLINE and MEDLINE In-Process & Other Non-Indexed Citations, EMBASE, and CENTRAL. Results of the searches were downloaded into EndNote X7 bibliographic management software and duplicate records were removed using several algorithms. In addition to database searches, reference checking of included journal articles and related systematic reviews were undertaken. For full details of all database search strategies, including interfaces used, search dates and result numbers, see Appendix 7, Table 86.

All trials included in the existing systematic reviews were obtained. Title and abstract screening and then full-text screening were performed by two reviewers with disagreements resolved by consensus, or by a third reviewer.

Inclusion/exclusion criteria

This review identified RCTs in which a treatment designed to lower blood glucose in women with GDM was examined in comparison with routine or standard antenatal care or an alternative treatment designed to lower blood glucose. The other inclusion criteria were as follows.

Quality assessment

The risk of bias of the included trials was assessed using the Cochrane risk of bias tool,172 which considers the following characteristics:

• sequence generation

• allocation concealment

• blinding of participants and medical staff to treatment allocation

• blinding of the assessors of outcomes to the treatment allocation

• completeness of outcome reporting (e.g. loss to follow-up)

• selective reporting of outcomes

• Other sources of bias (not addressed by the above domains).

Each criterion was classified as being at low, high or unclear risk of bias. One reviewer performed the quality assessment, which was checked by a second.

Data extraction

Data were extracted from each publication on the following:

• maternal age at randomisation

• gestational age at randomisation and/or oral glucose tolerance testing

• ethnicity

• BMI

• what test, and diagnostic criteria were used to diagnose GDM

• details of treatment and control used.

Statistical data for each reported outcome were extracted, when reported:

• numbers of women in treatment and control groups

• numbers of morbidities in each group

• OR or RR for comparison between groups (with 95% CIs)

• mean and SD for the outcome in each group.

One reviewer performed the data extraction, which was checked by a second reviewer.

Synthesis methods

Existing reviews

This review updates five existing systematic reviews (see Chapter 5, Methods). Here we present a short summary of the existing reviews.

Included trials

Trial publications from the five identified systematic reviews1,168–171 were obtained for screening. The two searches in September 2013 and October 2014 identified 6450 citations (2985 and 3555 citations, respectively). Following deduplication of titles (and abstracts where available), 3645 citations were reviewed (including citations of trials included in the previous reviews). Of these, 158 were judged potentially eligible based on title and abstract. After obtaining the full text and assessing eligibility, 48 trials were included (46 in the meta-analyses: two trials reported insufficient data to allow inclusion) and 110 were excluded (see Appendix 5, Table 67). The full details of the search process are presented in the flow chart (Figure 23).

FIGURE 23.

Flow chart of the search process.

Of the included trials, 23175–195 compared drug treatments: 10 trials175–183,196 compared metformin with insulin, eight trials184–191 compared glibenclamide with insulin, two trials192,193 compared glibenclamide with metformin, one trial compared a metformin–glibenclamide combination with insulin195 and one trial194 compared glibenclamide in addition to diet therapy with placebo in addition to diet therapy. Ten trials51,52,197–203 compared combinations of diet modification, glucose monitoring and insulin use to routine obstetric care. Five trials204–208 compared different insulin formulations. Of the remaining nine trials,200–202,209–214 five trials201,202,211–213 were comparisons of different diets. One trial200 compared types of dietary education, one trial214 compared diet to insulin, one trial210 compared exercise to insulin, and one trial209 compared exercise to diet. None of these nine trials200–202,209–214 was included in any meta-analysis because of trial diversity. The included trials are summarised in Table 19.

The trials included women with GDM who were diagnosed following a 75-g or 100-g OGTT, a variety of threshold criteria were used, including the Carpenter and Coustan criteria (C&C) or those of the NDDA,14 WHO,11 ADA136 and local guidelines (criteria were specified in the publications). 177,182 Inclusion criteria for women in the dietary modification trials were more varied. Some of these trials included women with a positive OGCT, but negative OGTT (therefore not diagnosed as GDM by usual criteria); some included women with ‘mild or borderline’ GDM (usually defined as having fasting glucose below the threshold, but post-load glucose above the threshold for diagnosing GDM) and women defined as having ‘IGT’ (although current diagnostic criteria would consider these women to have GDM).

Quality assessment

The quality assessment results are provided in Appendix 5, Table 68. In general, reporting of many aspects of trial quality was poor. The randomisation procedure was rarely described, and so its quality could often not be assessed, although all trials were described in the publications as being ‘randomised’ trials. Blinding of participants and medical staff was generally not reported, but as most of these trials include some administration of insulin, it is probable that most clinicians could not be blinded. Most trials had reasonably complete outcome data and loss to follow-up was low. Most trials presented results for all their prespecified outcomes; therefore selective reporting was assessed as minimal. The large number of possible outcomes, however, means that the possibility that some trials collected data on outcomes, but did not report them, cannot be ruled out.

Trials comparing metformin and insulin

There were 10 trials that compared metformin with insulin. 175–183,196 In general, women were eligible if they were unable to achieve glycaemic control with dietary and lifestyle modification. Therefore, there is the possibility that those included may have more severe insulin resistance or may be less compliant or find adhering to lifestyle interventions more difficult than those requiring only dietary intervention outside of a trial. The specific criteria for the addition of insulin are not reported in most trials, although some trials do report that supplemental insulin was prescribed if glycaemic control was not achieved by participants receiving metformin. It is possible that thresholds for what is defined as ‘good’ control differed between trials and between sites in multicentre trials. Included trials are summarised in Table 20. Criteria for diagnosing GDM vary across trials as does the screening strategy used; these differences should be considered when interpreting the results from the meta-analyses. Meta-analyses are shown by outcome; not all trials report each outcome.

TABLE 20 - Trials comparing metformin and insulin

NZ, New Zealand.

It is assumed that, unless otherwise reported, the screening strategy as advocated by the criteria used was adhered to.

Criteria did not correspond to any standard diagnostic criteria. GDM was diagnosed using 2-hour 75-g OGTT after an overnight fast if one or more capillary plasma glucose thresholds 5.3 (fasting), 11.0 (1 hour) and 9.6 (2 hours) mmol/l was equalled or exceeded.

The diagnostic cut-off values of plasma glucose up to December 2008 were the following: fasting ≥ 4.8 mmol/l, 1-hour ≥ 10.0 mmol/l and 2-hour ≥ 8.7 mmol/l, and thereafter ≥ 5.3, ≥ 10.0 and ≥ 8.6 mmol/l, respectively.

Conference abstract.

Reference	Year	Location	Population	Criteria used to diagnose GDM	Screening strategya
Hague176	2003	Australia	30	ADIPS	Risk based
Hassan175	2012	Pakistan	150	WHO	50-g OGCT
Ijäs177	2010	Finland	100	Reported in paperb	Risk based
Mesdaghinia178	2013	Iran	200	ADA	50-g OGCT
Moore179	2007	USA	63	NDDG	50-g OGCT
Niromanesh196	2012	Iran	160	C&C	50-g OGCT
Rowan180	2008	Australia/NZ	751	ADIPS	Risk based
Spaulonci181	2013	Brazil	94	ADA	Universal OGTT
Tertti182	2013	Finland	217	Finnish criteriac (changed during trial)	Risk-based screening then OGTT or universal OGTT testing
Zinnat183	2013	Bangladesh	450	Not reportedd	Not reportedd

Because of the large volume of treatments and outcome comparisons, we have generally combined trials and presented pooled estimates in forest plots. We have provided an example of one outcome (usually macrosomia) and treatment comparison by individual trials (results across trials of all outcomes are available from the contact author on request).

Figure 24 shows the results of the meta-analysis of trials comparing metformin and insulin treatment and risk of macrosomia. There was no evidence of heterogeneity in this analysis (I² = 2%).

FIGURE 24.

Metformin vs. insulin: macrosomia.

Figure 25 shows the results of the meta-analysis of trials for all dichotomous perinatal outcomes reported. There is a significantly reduced risk of macrosomia, neonatal hypoglycaemia and PIH (although for PIH this is marginal) for women given metformin compared with those who were given insulin. There is a suggestion that those who were given metformin rather than insulin are less likely to develop pre-eclampsia or have a baby admitted to the NICU, although results were not statistically significant. Metformin use was associated with a statistically significantly increased risk of instrumental birth (vaginal delivery by forceps or ventouse) compared with insulin use.

FIGURE 25.

Metformin vs. insulin: dichotomous outcomes.

Heterogeneity varied across these analyses but was generally low to moderate (I² = 0–60%); however, trials reporting C-section had high heterogeneity (I² = 71%).

Figure 26 shows the results of the meta-analysis of trials for all continuous outcomes: gestational age at birth, BW and 5-minute Apgar score for metformin vs. insulin treatment. All outcomes were similar between the two groups. Heterogeneity across trials was low.

FIGURE 26.

Metformin vs. insulin: continuous outcomes. GA, gestational age.

Trials comparing glibenclamide and insulin

Eight trials compared glibenclamide to insulin. 184–191 Details of these trials are given in Table 21.

TABLE 21 - Trials comparing glibenclamide and insulin

It is assumed that, unless otherwise reported, the screening strategy advocated by the criteria used was adhered to.

Reference	Year	Location	Population	Criteria used to diagnose GDM	Screening strategya
Anjalakshi184	2007	India	23	WHO	Universal OGTT
Bertini185	2005	Brazil	70	WHO	Not reported
Lain186	2009	USA	99	ADA – C&C	50-g OGCT then OGTT
Langer187	2000	USA	404	C&C	50-g OGCT then OGTT
Mukhopadhyay188	2012	India	60	WHO	Universal OGTT (2-hour post load > 7.7 mmol/l only)
Ogunyemi189	2007	USA	97	Not reported	Not reported
Silva190	2007	Brazil	68	WHO	Universal OGTT
Tempe191	2013	India	64	C&C	50-g OGCT then OGTT

Figure 27 shows the results of the meta-analysis of trials for all dichotomous perinatal outcomes for glibenclamide compared with insulin treatment. The analysis suggests that insulin is more effective in reducing the odds of an adverse outcome compared with glibenclamide (LGA, macrosomia, neonatal hyperglycaemia and pre-eclampsia) or has similar effects (NICU admission, preterm birth and C-section); however, the increases were not statistically significant and the CIs were wide.

FIGURE 27.

Glibenclamide vs. insulin: dichotomous outcomes.

Figure 28 shows the results of the meta-analysis of trials for all continuous outcomes – gestational age at birth, BW and Apgar score at 5 minutes – for glibenclamide vs. insulin treatment. Infants whose mothers were given glibenclamide were generally born heavier than infants of mothers given insulin (approximately 120 g). Gestational age at birth and Apgar score at 5 minutes were both similar between the two groups.

FIGURE 28.

Glibenclamide vs. insulin: continuous outcomes. GA, gestational age.

Trials comparing glibenclamide and metformin

Two trials directly compared metformin to glibenclamide (Table 22). 192,213

TABLE 22 - Trials comparing glibenclamide and metformin

It is assumed that, unless otherwise reported, the screening strategy advocated by the criteria used was adhered to.

Reference	Year	Location	Population	Criteria used to diagnose GDM	Screening strategya
Moore192	2010	USA	149	C&C	50-g OGCT then OGTT
Silva193	2012	Brazil	200	WHO	Not reported

Figure 29 shows the risk of dichotomous outcomes for glibenclamide treatment compared with metformin treatment, and Figure 30 shows the risk of continuous outcomes for glibenclamide treatment compared with metformin treatment. Given the limited number of trials in this analysis with few women (349) included, it is difficult to draw meaningful conclusions. For several outcomes, only one trial reported data, and, even when data from two trials were combined, the results were generally non-significant, with wide CIs.

FIGURE 29.

Glibenclamide vs. metformin: dichotomous outcomes.

FIGURE 30.

Glibenclamide vs. metformin: continuous outcomes. GA, gestational age.

Network meta-analysis comparing glibenclamide, insulin and metformin

For the network analysis, all trials comparing one pharmacological treatment with another have been included: metformin, glibenclamide and insulin (Figure 31). Only dichotomous outcomes reported in at least two glibenclamide trials were included to ensure there were sufficient trials in each network analysis to produce meaningful results. Separate network analyses were performed for each outcome; their results are shown in Figure 32.

FIGURE 31.

Network meta-analyses, relationship of comparisons.

FIGURE 32.

Network meta-analysis comparing metformin, glibenclamide and insulin. First better – treatment listed first in the outcome column is superior; second better – treatment listed second in the outcome column is superior.

The network analyses suggest that, compared with insulin, metformin reduces the risk of all reported outcomes with the exception of C-section. However, comparisons include the null value, and the CIs are wide and include either an increase or a decrease in risk.

Glibenclamide compared with insulin or metformin treatment seems to increase the risk of macrosomia and LGA. There is no evidence of a difference between glibenclamide and insulin or glibenclamide and metformin treatment for the remaining outcomes. These results are similar to the results from the direct meta-analyses. The loss of estimate precision (and statistical significance) for comparisons in the network meta-analyses compared with the direct (head-to-head) meta-analyses may be a result of the increased heterogeneity within group comparisons across trials.

Table 23 shows the estimated probability of the effectiveness of each treatment at reducing the risk of each outcome, derived from the network meta-analysis. This analysis suggests that for all of the outcomes, with the exception of C-section, metformin is most likely to be the most effective treatment, and for most outcomes the probability that metformin is most effective is reasonably high.

TABLE 23 - Estimated probability (%) of a treatment being the most effective in reducing the risk of a dichotomous outcome

Outcome	Treatment
	Insulin	Metformin	Glibenclamide
LGA	17.2	82.7	0.1
Macrosomia	3.5	96.4	0
NICU admission	1.4	50.3	48.3
Neonatal hypoglycaemia	7.9	89.5	2.7
C-section	12.6	9.4	78
Pre-eclampsia	11.5	74.6	13.9

Insulin does not ever seem to be the best treatment, with very low probabilities of being best for each outcome. However, it should be noted that supplemental insulin was often added to treatment with metformin in the trials that report management of blood glucose when glycaemic control was not achieved with diet, lifestyle and metformin. This analysis therefore suggests metformin is superior (in terms of its ability to influence the risk of associated adverse outcomes) to insulin and glibenclamide as a first-line treatment, rather than a standalone treatment.

Trials comparing different insulin preparations

Five trials compared different insulin preparations:195,205,207,210,218 three compared analogue to human insulin,195,210,218 one different numbers of doses per day,207 and one gave insulin only to women with the most elevated glucose levels. 205 Details of these trials are given in Table 25.

TABLE 25 - Trials comparing different insulin preparations

Reference	Year	Location	No.	Experimental insulin	Control insulin	Criteria used to diagnose GDM
Balaji195	2012	India	323	Analogue insulin (Aspart BIAsp)	Human insulin	WHO
Di Cianni210	2007	Italy	96	Analogue (Aspart) or lispro	Human insulin	C&C
Jovanovic218	1999	USA	42	Lispro (humalog)	Human insulin	C&C
Kjos205	2001	USA	98	Insulin if fetal abdominal circumference > 70th centile and/or FPG > 6.7	Insulin irrespective of fetal growth or glucose levels	FPG 5.8–6.7 mmol/l
Nachum207	1999	Israel	392	Four times daily administration	Twice-daily administration	NDDG

The differences in the composition of the insulin preparations used by the trials precluded their inclusion in a meta-analysis. As an alternative, summary results for each trial and each outcome are presented in forest plots, without pooling across trials. Figure 33 shows the effects of the differing insulin preparations on dichotomous outcomes, and Figure 34 shows the effect on continuous outcomes. For the majority of outcomes, there are no statistically significant differences in the effectiveness of different insulin preparations.

FIGURE 33.

The effect of different insulin preparation on dichotomous outcomes.

FIGURE 34.

The effect of different insulin preparation on continuous outcomes. GA, gestational age.

Trials comparing different types of diet modification

Ten trials51,52,196–198,217–221 compared diet modification or advice, possibly alongside glucose monitoring and insulin use (although this was often not reported) to routine antenatal care (usually no specific diet modification or insulin treatment) (Table 26).

TABLE 26 - Trials comparing diet modification (with insulin if needed) and routine antenatal care

The trial was conducted in the United Arab Emirates; no details of GDM criteria reported.

Not internationally recognised criteria.

Reference	Year	Location	No. included	Criteria used to diagnose GDM	Screening strategy	Insulin use in diet group
Bevier202	1999	USA	103	Positive OGCT, negative OGTT	Universal 50-g OGCT only	If needed
Bonomo203	2005	Italy	300	Positive OGCT, negative OGTT	Risk factors and 50-g OGCT	Not reported
Crowther51	2005	UK/Australia	1000	WHO 199911	Risk factors or 50-g OGCT	If needed
Deveer197	2013	Turkey	100	ACOG-positive OGCT, negative OGTT	Universal 50-g OGCT and OGTT	Not reported
Elnour200	2006	UAE	180	Not reporteda	Not reported	If needed
Garner201	1997	Canada	299	bHatem 1988221	Universal 75-g OGCT and selective OGTT test result	If needed
Landon52	2009	USA	958	ADA 200612	Universal 50-g OGCT and selective OGTT	If needed
Li198	1987	Hong Kong	158	NDDG 197915	Risk factors and selective OGTT	Not reported
O’Sullivan211	1966	USA	615	O’Sullivan 196416	OGCT or risk factors	Only in treated group
Yang199	2003	China	150	WHO 1998222	Universal OGCT and selective OGTT	If needed

In seven trials, insulin was reported as being used if required; three trials197,198,203 did not report insulin use. Two trials220 reported secondary analyses of previously published trials;52,211 both trials220 are therefore excluded from the analyses.

Across the included trials, the dietary interventions included specific diets, individualised advice from a dietitian, or more general advice (the compositions of the dietary interventions were generally well reported by trials). Two of the included reports were secondary analyses of data from the Crowther trial,51 one223 of which was a secondary analysis, examining longer-term infant (4- to 5-year-olds) obesity risk (BMI z-scores) and so was excluded from our meta-analyses.

For the meta-analysis the varying forms of dietary modification and advice were assumed to be equivalent, and any potential differences in insulin use were not considered. The forest plot for the meta-analysis of trials reporting macrosomia as an outcome is presented in Figure 35 (results across trials of all other outcomes are available from the authors on request). This analysis suggests that diet modification halves the incidence of macrosomia compared with routine care (whatever that care may be). The three trials that did not report insulin use report similar results to those trials reporting that supplemental insulin was used when required. Heterogeneity was moderate (I² = 32%), driven by the two small trials with extreme outcomes. 197,202

FIGURE 35.

The effect of diet modification on macrosomia incidence.

Figure 36 shows the risk of dichotomous outcomes and diet modification compared with routine care, and Figure 37 shows the risk of continuous outcomes and diet modification compared with routine care. Diet modification seems to reduce the risk of LGA, macrosomia, shoulder dystocia and pre-eclampsia by around 50%, with a more modest 15% reduction in the incidence of C-section compared with routine care. Diet modification compared with routine care also seems to reduce BW by approximately 120 g. There is no evidence that diet modification reduces the incidence of neonatal intensive care admission, neonatal hypoglycaemia, induced labour, preterm birth or Apgar score at 5 minutes. Heterogeneity varied across outcomes, with I² ranging from 0% to 67%.

FIGURE 36.

The effect of diet modification on dichotomous outcomes.

FIGURE 37.

The effect of diet modification on continuous outcomes. GA, gestational age.

Subgroup analyses: diet modification trials by definition of gestation diabetes mellitus

Unlike the trials evaluating pharmacological treatments (which tended to use recommended tests, criteria and thresholds for diagnosing GDM) the tests, criteria and thresholds used to define a comparison group’s ‘GDM status’ varied considerably across the 10 trials included in the meta-analyses of diet modification trials. These differences are briefly summarised in Table 27.

TABLE 27 - Tests, criteria and thresholds used by included trials in diet modification trials

Reference	Year	Criteria or test used to diagnose GDM	Thresholds
Bevier202	1999	Positive OGCT	Positive OGCT and negative OGTT
Bonomo203	2005	Positive OGCT	Positive OGCT and negative OGTT
Crowther51	2005	WHO	Mild GDM fasting < 7.8 mmol/l and 2-hour post load 7.8–11.1 mmol/l
Deveer197	2013	Positive OGCT	Positive OGCT and negative OGTT
Elnour200	2006	Not reported	‘GDM’ diagnosed ≤ 20 weeks’ gestation
Garner201	1997	‘Hatem’ trial specific	GDM fasting > 7.5 mmol/l 2-hour > 9.6 mmol/l
Landon52	2009	ADA	Mild GDM fasting < 5.3 mmol/l and two or more of: 2-hour > 10.0 mmol/l and 2-hour > 8.6 mmol/l, 3-hour > 7.8 mmol/l
Li198	1987	WHO	IGT fasting < 7.9 mmol/l and 2-hour post load 7.8–11.1 mmol/l
O’Sullivan211	1966	O’Sullivan	GDM fasting > 6.1 mmol/l 1-hour> 9.4 mmol/l, 2-hour> 6.7 mmol/l, 3-hour > 6.1 mmol/l
Yang199	2003	WHO 1998	Fasting ≥ 7.0 mmol/l and 2-hour post load 7.8–11.1 mmol/l

Three trials included women with a positive OGCT and a negative OGTT; therefore, they did not meet any criteria for current GDM diagnoses. Some early trials, for example Li (1987) classify women as having IGT or mild GDM. Today, however, IGT and ‘mild’ GDM are viewed within the same spectrum of hyperglycaemia and are therefore usually classified as GDM.

In order to investigate the impact of these potentially different populations we performed subgroup analyses (irrespective of glucose load and glucose thresholds used), dividing the trials into four categories:

• GDM Women had GDM according to a ‘standard’ diagnostic criteria, e.g. ADA, WHO or C&C.

• Mild GDM Women described as having mild GDM.

• IGT Women described as having IGT.

• Negative OGTT Women had positive OGCT but negative OGTT.

Figure 38 shows the effect of diet modification on risk of macrosomia, with trials grouped by glucose level/severity classification [most severe first (GDM) down to positive OGCT, negative OGTT]. Although data within each group are limited, there is no evidence that the effect of diet modification varies substantially according to the glucose level/severity classification. Figure 39 summarises the results for the dichotomous outcomes LGA, macrosomia, C-section and pre-eclampsia, the outcomes where there is strongest evidence of a benefit of diet modification. Again, there is no evidence of any treatment effect differences between trials, although, again, data are limited.

FIGURE 38.

Impact of diet modification on macrosomia, by degree of glucose intolerance [GDM, IGT, mild GDM or (positive OGCT) negative OGTT].

FIGURE 39.

Impact of diet modification on dichotomous outcomes, by degree of glucose intolerance [GDM, IGT, mild GDM or (positive OGCT) negative OGTT].

Because women with ‘mild’ GDM and IGT may both be categorised as having GDM under current diagnostic guidelines we combined these groups with those described as having GDM in a further subgroup analysis. We compared this new GDM group to women without GDM (those with elevated glucose at OGCT and with a subsequently ‘normal’ OGTT). The results of this subgroup analysis are shown in Figure 40. Diet modification (with insulin if needed), irrespective of the severity of the hyperglycaemia identified or the test used to identify it, seems to be effective in reducing the risk of adverse perinatal outcomes.

FIGURE 40.

Impact of diet modification comparing women with GDM to those with only a positive OGCT and negative OGTT.

‘Other’ diet and exercise trials

Nine trials204,206,208,209,212,215–217,219 were too methodologically diverse to allow pooling of data. Six trials206,209,212,215,217,219 compared two different types of diets, one trial204 compared exercise to insulin use, one trial compared diet and insulin to diet alone, and one compared exercise with diet216 (Table 28).

TABLE 28 - ‘Other’ diet and exercise trials

DASH, Dietary Approaches to Stop Hypertension. GI, glycaemic index.

Author	Year	Location	No. included	Experimental group	Control group	Criteria used to diagnose GDM
Asemi215	2014	Iran	52	‘DASH’ diet (high fruit, veg, wholegrain and dairy)	Control diet	ADA
Bo216	2014	Italy	200	Diet and exercise	Diet only	Not reported
Bung204	1991	USA	41	Exercise bike use	Insulin	Not reported
Cao217	2012	China	275	Individual diet education	Standard group diet education	Not reported
Cypryk209	2007	Poland	30	High-carbohydrate diet	Low carbohydrate diet	WHO
Louie219	2011	Australia	99	Low-GI diet	High fibre moderate GI diet	ADIPS
Moreno-Castilla206	2013	Spain	152	Low-carbohydrate diet	Control diet	NDDG
Rae212	2000	Australia	124	Low-calorie diabetic diet	Standard diabetic diet	OGTT fasting > 5.5 mmol/l or 2-hour > 7.9 mmol/l (glucose load not reported)
Thompson208	1990	USA	108	Diet and insulin	Diet only	NDDG

As commented on above, these trials compared very different interventions (diet, exercise and insulin preparations), therefore meta-analysis has not been undertaken; instead, summary results for each trial and each outcome are presented in forest plots (without pooling across trials).

Figure 41 shows the effect of the differing diets, by trial, on the risk of dichotomous outcomes, and Figure 42 shows the effect of the differing diets, by trial, on the risk of continuous outcomes. Results are varied; however, there is no evidence that any one particular type of diet improves all outcomes reported. See Table 28 for information on type of intervention and participant numbers.

FIGURE 41.

Effect of diet interventions on dichotomous outcomes.

FIGURE 42.

Effect of diet interventions on continuous outcomes. GA, gestational age.

In the three remaining trials,204,208,216 a mixture of interventions was evaluated. Bo et al. 216 gave the same diet to four groups of women; the first group also received diet recommendations, the second group was advised to walk briskly for 20 minutes each day, the third group received behavioural dietary recommendations and the fourth group received both of the second and third groups’ interventions. Bo et al. 216 reported that exercise reduced postprandial glucose levels and a composite measure of maternal and neonatal complications, whereas behavioural interventions had no effect. Bung et al. 204 compared exercise and diet to insulin and diet alone in a group of women who had ‘failed’ to achieve ‘normal’ glucose levels on diet alone within a week of GDM diagnosis. Bung et al. 204 reported no differences between groups in outcomes; however, macrosomia rate was doubled in the group receiving insulin (two vs. four), which seems contrary to results expected; numbers in the trial were small (34), however, possibly accounting for the results. Thompson et al. 208 compared diet with insulin to diet alone, reporting that the group that was given diet advice with insulin had infants with lower BW and incidence of macrosomia compared with the group that was given diet advice alone. Group rates of C-section, shoulder dystocia and neonatal hypoglycaemia were similar.

Pharmacological treatments

For women requiring pharmacological intervention to reduce hyperglycaemia (often when diet alone had proved ineffective), metformin as first-line treatment seems to be at least as effective as insulin in reducing the risk of adverse perinatal outcomes, and metformin and insulin seem to be more effective than glibenclamide.

Although the treatment effects of metformin and insulin were similar, there was a trend for metformin to perform better, or at least no worse, than insulin for most outcomes reported. For example, metformin reduces the risk of macrosomia by 25% compared with insulin. Although not statistically significant, there was some evidence that treatment with metformin may be associated with a greater risk of preterm birth and an Apgar score of < 7 at 5 minutes compared with insulin treatment. Glibenclamide, performed less well than insulin for several outcomes, for example the infants of mothers who were given glibenclamide were, on average, 120 g heavier at birth compared with the infants of mothers who were given insulin; the risk of LGA and macrosomia is also greater for those given glibenclamide compared with insulin, but not significantly so. The network meta-analysis confirmed the direct trial meta-analyses findings and suggests that there is a high probability that metformin is the most effective treatment, compared with insulin or glibenclamide, for reducing the risk of most adverse perinatal outcomes examined within this review.

Metformin, in addition to performing well, may also be preferred by women, as it is administered orally and can be stored at room temperature as opposed to insulin, which requires subcutaneous injection and refrigerated storage. Metformin is associated with gastrointestinal upset, however, which may affect compliance; unfortunately, few trials report side effects or participant satisfaction, quality of life or well-being, which should be examined by future trials.

Dietary modification

Dietary modification generally reduces the risk of most reported perinatal outcomes compared with ‘usual or routine’ antenatal care. For example, the risk of macrosomia is halved (RR 0.46, 95% CI 0.36 to 0.60) and the risk of C-section is reduced by 14% (RR 0.86, 95% CI 0.77 to 0.95) with dietary modification compared with usual/routine antenatal care. If trials are adequately powered and methodologically robust (particularly in terms of adequacy of randomisation) it is reasonable to accept that diet modification does improve adverse outcome rates compared with routine care (without special diet modification); however, risk of bias was generally high or unclear, suggesting poor trial quality with respect to these indicators.

There was no evidence that the effect of diet modification varied according to the types of women (or severity of hyperglycaemia) included in the trials. For LGA, macrosomia and C-section, diet modification seemed to be equally effective in women with a negative OGTT (usually following a positive OGCT) compared with women with diagnosed GDM or ‘mild’ GDM. This finding suggests that modifying the diet of women who do not have GDM (as currently defined) may be as effective in reducing risks as in women with diagnosed GDM. This is supported by trials of diet and lifestyle modification that have been undertaken in women who do not have GDM. 224 However, compliance with diet advice may be less (because women may be less amenable to change if they do not believe themselves at risk) and thus any beneficial effects may be reduced. The finding that dietary modification can reduce the risk of most adverse outcomes in women with lower glucose levels (below those currently diagnostic of GDM at the time the trials were conducted) is important given the recommendations of the IADPSG and our analyses findings (detailed in Chapter 2), which suggest that lower thresholds are required to identify the majority of infants at risk of LGA and high adiposity at birth. As we have explained in Chapter 2, these outcomes (LGA and high adiposity at birth) are associated with increased obesity and cardiometabolic risk. It is assumed that treatment of GDM, however, will reduce the risk of these longer-term outcomes, although that remains to be substantiated by large RCTs with longer-term follow-up.

Although our meta-analyses suggest that diet modification is effective in reducing the risk of the majority of reported adverse perinatal outcomes compared with routine antenatal care, nine trials could not be included because of differences between comparisons and interventions (see Table 28). These trials often included small numbers reducing the reliability of their results, for example although Asemi et al. 215 reported that their ‘Dietary Approaches to Stop Hypertension’ (DASH) diet significantly lowered BW, reduced macrosomia, C-section rate and need for insulin compared with a control diet; only 52 women were included, and head circumference and ponderal index were also significantly reduced in the infants of women consuming the DASH diet compared with control diet, suggesting a detrimental effect. The remaining trials showed varying results by outcome (see Figures 41 and 42).

Analogue and human insulin

Although the number of trials was limited, the analyses suggest that analogue and human insulin are equally effective.

Frequency of insulin administration

One trial207 investigated frequency of insulin administration and reported a statistically significant reduction in neonatal hypoglycaemia when insulin was administered four times daily in comparison with twice daily. However, the number of women included were few (274) and, consequently, there were few events (one in the four-times administration group; eight in the two-times administration group).

We found that few trials included in this review reported negative treatment effects and it is possible that negative outcomes such as gastrointestinal upset associated with metformin use may reduce compliance and treatment effects.

Conclusions

Treatment of GDM with diet and lifestyle and pharmacological interventions seems to reduce the risk of most reported perinatal adverse outcomes. Diet modification alone seems to reduce the risk of adverse outcomes even in women with glucose levels below those currently diagnostic of GDM. Given the graded linear association between glucose levels and adverse outcomes (reported in Chapters 2 and 3) and findings from a systematic review of trials in non-GDM populations,224 the provision of dietary advice for all pregnant women (irrespective of their glucose levels at OGTT) may be beneficial in terms of reducing the risk of adverse outcomes across the whole glucose spectrum. Dietary advice, however, may be costly, especially if specialist advice is provided above the dietary advice that could be given by obstetricians and midwives during ‘routine’ antenatal appointments. It is also possible that women who view themselves as ‘normal’ may be less compliant with dietary advice than women who are aware that they have GDM and who appreciate that dietary modification may improve their and their infant’s health outcomes and reduce their need for supplementary pharmacological intervention, especially insulin. Women requiring pharmacological intervention in addition to diet and lifestyle, however, may also be more insulin resistant or their insulin resistance may be more refractory than women who require only diet and lifestyle advice.

Supplemental metformin in addition to diet and lifestyle modification (if required to normalise glucose levels) is as effective as insulin and therefore should be the first-line pharmacological treatment of choice, as it is at least as effective as insulin and may be preferred by women because it does not require injection, although it should be remembered that trials generally used insulin in the metformin group if hyperglycaemia was not ‘well’ controlled. The results of this review provide reassurance, however, that a ‘step-up’ approach of first providing dietary and lifestyle advice then adding supplementary metformin or insulin if glucose levels are not adequately controlled is a reasonable and effective approach to take.

Overview

A decision tree model was developed to evaluate the cost-effectiveness of alternative strategies of combined screening, diagnosis and treatment of hyperglycaemia during pregnancy in the UK for a time horizon of 3 months in the base-case analysis. The best-performing strategy is identified by backward induction. At the first step the best-performing diagnostic glucose thresholds are identified (defining the best-performing diagnostic strategy). The second step is a full incremental comparison of all strategies composed variously of screening, diagnosis and treatment, but for which the diagnostic glucose thresholds are set at those identified at the first stage. Results are expressed in terms of costs and quality-adjusted life-years (QALYs). The perspective of the analysis is that of NHS and personal social services. The key modelling assumptions in the economic analysis are listed in Appendix 6, Table 69.