Screening for hyperglycaemia in pregnancy: a rapid update for the National Screening Committee

Article history

The research reported in this issue of the journal was commissioned by the HTA programme as project number 08/82/01. The contractual start date was in January 2009. The draft report began editorial review in August 2009 and was accepted for publication in April 2010. As the funder, by devising a commissioning brief, the HTA programme specified the research question and study design. The authors have been wholly responsible for all data collection, analysis and interpretation, and for writing up their work. The HTA editors and publisher have tried to ensure the accuracy of the authors’ report and would like to thank the referees for their constructive comments on the draft document. However, they do not accept liability for damages or losses arising from material published in this report.

Declared competing interests of authors

None

Copyright statement

© 2010 Queen’s Printer and Controller of HMSO. This journal is a member of and subscribes to the principles of the Committee on Publication Ethics (COPE) (http://www.publicationethics.org/). This journal may be freely reproduced for the purposes of private research and study and may be included in professional journals provided that suitable acknowledgement is made and the reproduction is not associated with any form of advertising. Applications for commercial reproduction should be addressed to: NETSCC, Health Technology Assessment, Alpha House, University of Southampton Science Park, Southampton SO16 7NS, UK.

2010 Queen’s Printer and Controller of HMSO

The previous Health Technology Assessment review

Screening for gestational diabetes has long been a controversial topic. Even the definition of what is included varies. Strictly speaking gestational diabetes should refer to diabetes which comes on during pregnancy and resolves after delivery, but both the Scottish Intercollegiate Guidelines Network (SIGN) Guideline No. 1161 and the National Institute for Health and Clinical Excellence (NICE) guideline on diabetes in pregnancy2 have used the World Health Organization (WHO) description of diabetes with first onset in pregnancy. In the USA the definition does not require post-natal resolution, so the US label of gestational diabetes mellitus (GDM) can include true type 2 diabetes mellitus (T2DM) with onset, or just diagnosis, during pregnancy.

A previous Health Technology Assessment (HTA) report3 reviewed the literature on screening for GDM, published up to the middle of 2000. It also assessed the case for screening against the criteria of the National Screening Committee (NSC) (available on NSC website4). The main findings were that:

• There was still debate about what was meant by GDM. The threshold for diagnosis was not clear, and what was called GDM usually included impaired glucose tolerance. There were many different definitions, perhaps reflecting the history of GDM, which was originally defined on the basis of the mother’s risk of later development of T2DM. In recent years the focus has been more on harms to the baby.

• The criteria for gestational diabetes, as defined by the WHO in 1998,5 used a cut-off of 7.0 mmol/l for fasting and of 7.8 mmol/l for the 2-hour post oral glucose tolerance test (OGTT) levels, which meant that a much larger range of hyperglycaemia was included than in non-gestational diabetes, including impaired glucose tolerance (IGT) as well as diabetes. Various studies showed that what was classed as GDM using non-pregnant diagnostic levels was in fact mainly IGT. This failed to take into account the physiological rise in post-prandial glucose (PPG) levels in pregnancy. In a Swedish study,6 the mean 2-hour plasma glucose (PG) level was 8.0 mmol/l; if the WHO criteria were applied, 18% would have had GDM.

• There was almost certainly a continuum of risk, rather than there being two distinct groups of normal and abnormal.

• The key risk factor might be maternal overweight leading to glucose intolerance.

• Diseases should be defined by the harm they do. The early definitions of GDM were based on levels which predicted later diabetes in the mother. Later ones incorporated fetal risk. However, that was often based on ‘macrosomia’, which was arbitrarily based on a birthweight of 4000 g (about 8 lbs 11 oz) or 4500 g. Basing it on weight does not distinguish between large healthy babies, and those with the unhealthy insulin-driven overgrowth of adipose tissue. And cut-offs of 4000 and 4500 are chosen for neatness, and are not based on physiology or pathology.

• One issue was whether to have universal screening, or selected screening based on risk factors. A policy of universal screening did not appear justified, so the approach might be to screen women with factors known to increase the risk, such as age, ethnicity and obesity. However, risk factors have low sensitivity and specificity.

• Another problem was which measure of blood glucose (BG) to use. The leading competitors amongst possible tests included fasting plasma glucose (FPG) and the 50-g glucose challenge test (GCT). However the former would miss post-prandial hyperglycaemia, which as Fraser (1995)7 and Jovanovic (2002)8 have argued, may be enough to cause overgrowth of some fetal tissues. The literature did not give a clear answer as to which of FPG or the 50-g challenge test was better. The cost per case found was similar.

• The optimum thresholds for positive screening tests were uncertain. The threshold for the FPG might have to be as low as 4.7 mmol/l in order to provide sufficient sensitivity. The most commonly cited thresholds for the GCT were 7.2 mmol/l and 7.8 mmol/l. These seemed too low and the report favoured a cut-off of 8.2 mmol/l for the GCT.

• Treatment options included diet and exercise, and insulin. However it was noted that trials of oral agents such as metformin were under way.

• Screening for GDM failed to meet some of the NSC criteria.

The case for screening was assessed against the NSC criteria as follows. The full details are given in the previous HTA report. 3 (Note that the criteria numbers are as used in 2001 and are different from those in the current list, which has been expanded to cover genetic screening.)

What has changed?

Firstly, ACHOIS has now reported,18 and has shown that treatment of glucose intolerance (i.e. not just GDM) improves outcomes. The Maternal–Fetal Medicine Units Network trial of treatment of ‘mild’ gestational diabetes has also now reported. 20 A number of trials of oral agents have been published. A few new trials on non-pharmacological treatment have been published. It is thus timely to review the full range of treatment options, and we do this in Chapter 2.

Secondly, the main results of HAPO have also been published,19 and other studies of screening and the relationship between glucose levels and outcomes have been published since the last review. There is now a need for an update of the previous review in order to inform deliberations at NSC. The HAPO study and other results are described below.

Thirdly, there have been marked trends in two of the risk factors for GDM, maternal age and weight.

Data from England and Wales,21 (Table 1) shows that from 1998 to 2008 the number of live births to mothers in the 35–39 age group increased by 47%, and in the 40-and-over age group by 95%. This is contrasted with much smaller increases in the younger age groups, i.e. the under 25 age group has increased by 12%, the 25–29 age group remained virtually unchanged and the 30–34 age group increased by 2%. Also, it can be seen that, as a percentage of the total births for each year, the proportion to women 35 years and over has increased from 14.5% in 1998 to 20.1% in 2008.

Also, Scottish data show that the proportion of pregnancies to mothers under 25 has dropped by about half, whereas the proportion to mothers aged 30–34 had trebled (from about 4.5% in the 1970s to about 16% in recent years. 22 For every year after the age of 25, the relative risk (RR) of developing GDM rises by 4% [Solomon et al. reported by Yogev et al. (2009)23].

Data on weight from the Health Survey for England24 show that the proportion of women aged 25–34 years with a body mass index (BMI) over 30 has risen from 11% in 1993 to 19% in 2007. For women aged 35–44 years, the proportion rose from 17% in 1993 to 24% in 2007. Data for single years 1993–2007 show that the proportion overweight has been fairly static around 32%, but the proportion obese has risen from 16% to 24%. By 2003, 55% of women aged 34–44 and 47% aged 25–34 years were overweight or obese. This is reflected in maternal obesity.

The Middlesbrough study25 showed that the percentage of mothers who were obese has risen steadily, from about 10% in 1990 to 16% in 2004.

Hence women are getting heavier, and having children later. Both these factors will increase the proportion with hyperglycaemia in pregnancy. A US study by Getahun et al. (2008)26 showed that the prevalence of GDM increased from 1.9% in 1989–90 to 4.2% in 2003–4.

The term ‘gestational diabetes’ may have outlived its usefulness, which is why we have used the term ‘hyperglycaemia in pregnancy’ (HGP) for this review. GDM means different things to different people, and most of what was called GDM was really IGT in pregnancy. The HAPO study should help with a revision of definitions.

The underlying hypothesis in gestational diabetes was set out by Pedersen in 1954. 27 In brief, maternal hyperglycaemia leads to fetal hyperglycaemia (because glucose crosses the placenta into the baby’s bloodstream) which stimulates fetal insulin release and hyperinsulinaemia. This causes excessive growth of certain tissues before birth, and after birth the withdrawal of maternal glucose leaves the baby at risk of hypoglycaemia because of its own over-production of insulin.

Key questions for this review

The need for an update of the previous review3 for NSC arises for several reasons:

• Overweight and obesity are increasing in the population, which will lead to an increase in the frequency of GDM and HGP.

• Maternal age at birth is also increasing, which again will raise the incidence of HGP (though some of this will be due to weight gain with age).

• We have better evidence on treatment from the ACHOIS18 and MFMUN20 trials.

• We now have the HAPO results. 19

The key questions are:

1. After HAPO and similar studies, what level of HGP do we need to detect? This depends on the level at which we intervene to reduce adverse consequences. At the high end of PG, there will be definite benefits to the baby and the benefits will outweigh the harms and inconveniences. But at the lower end of the hyperglycaemia distribution (which could be just above the upper limit of normal) the harms and inconveniences may outweigh the benefits, if any. What we need to do is decide where along the continuum of glucose, the ‘switching point’ is. Below this point intervention would result in a net disbenefit, above it intervention would produce overall benefit.

2. Having decided the level at which to intervene, which screening test should be used?

3. Should there be universal PG testing, or selection by risk factors so that only a proportion of women proceed to blood testing?

4. Are oral glucose lowering drugs effective and safe? If so, what should the treatment pathway be? Lifestyle, then oral agents, then insulin, as required to control PG? If oral drugs are clinically as effective as insulin in most cases, their availability will make treatment cheaper and easier and more acceptable to women. There have also been, since the NICE reviews, new studies of lifestyle measures.

5. What are the research needs now? Should the HTA programme seek to commission any more primary research? NICE suggested that ‘a multi-centre trial is required to test existing screening techniques in the UK’.

The continuum problem

As mentioned above, the HAPO study19 found linear associations between glucose levels and adverse outcomes. That makes deciding on cut-offs for intervention difficult. The clinical approach might be to intervene at a level which seemed ‘clinically reasonable’ perhaps based on the RR – for example, choosing the level at which adverse events increased by 50%. That would be simply a concealed arbitrary clinical decision.

An economic approach would be to model the cost per QALY of intervening at different thresholds, using the affordability threshold use by NICE of up to £30,000 per QALY. However, although the NICE threshold has become the norm in the UK, NICE does not always abide by it, and the figure of £30,000 in effect represents a concealed arbitrary NICE decision.

One option would be to choose a threshold cost per QALY well below the £30,000 NICE one, and by further modelling (outwith the scope of this review, but which could be done by those who did the modelling for NICE) identify a number of screening and treatment strategies which gave a cost per QALY of under, say, £20,000. Note that many of the options identified by the NICE modellers were well under that threshold.

Having identified such screening strategies, the effect on numbers to be screened could then be estimated, and the practicalities of screening and treating such numbers could be considered. This approach would in effect be a pragmatic mixture of cost-effectiveness analysis and clinical judgement.

The alternative would be to postpone a decision until a randomised trial, with cost-effectiveness analysis, of intervention at different thresholds had been done.

The recent trials

Despite the recognition that hyperglycaemia in pregnancy is associated with both maternal and neonatal adverse outcomes, there has been uncertainty about the benefits of treating this. However, two studies undertaken during recent years have contributed to our understanding of the benefits associated with intervention.

Treatment of hyperglycaemia in pregnancy: oral drugs versus insulin

The principal aims of treating hyperglycaemia in pregnancy are to reduce mortality and morbidity among women and their children. 2 Treatment has typically involved:51–56

• dietary modification ± exercise and

• pharmacological treatment for women unable to maintain their glucose levels within an acceptable range despite dietary modification ± exercise.

It has been claimed that most women with hyperglycaemia in pregnancy can maintain adequate glycaemic control without pharmacological therapy. 2,57 However, others have stated that up to 60% of women will require pharmacological therapy. 53,56 For those women requiring pharmacological treatment, insulin has traditionally been used. 51,52,54–56,58–61

Oral anti-diabetic agents were used during the 1970s and 1980s to treat T2DM and hyperglycaemia in pregnancy. 58 Following concerns regarding increased perinatal mortality and neonatal hypoglycaemia, they fell out of favour. 59

However, insulin therapy is not without its drawbacks:

• need for injections which may be unpopular with women,56 which may lead to problems with adherence53

• risk of hypoglycaemia56

• risk of excessive weight gain56

• cost of insulin and the equipment necessary to administer it. 54

Furthermore, as hyperglycaemia in pregnancy is associated with insulin resistance and reduced insulin secretion, oral agents may have a role. 56 Therefore, the use of oral anti-diabetic agents has been investigated and debated during recent years.

Metformin reduces gluconeogenesis and improves glucose uptake56 by increasing the number, and sensitivity, of glucose receptors to insulin. 53 It is not associated with weight gain or hypoglycaemia. 56 However, metformin is known to cross the placenta. 58,59,62

The sulphonylurea drug, glyburide/glybenclamide, acts by stimulating maternal insulin secretion. 53 It is claimed that, compared with insulin, it is less likely to cause hypoglycaemia. 63 It is uncertain if glyburide crosses the placenta. Some researchers suggest that it does not, whereas others have reported minimal transfer. 58,59,62 The key point is that, if it does, the level in the fetal circulation is very low. This applies only to glibenclamide and not to other sulphonylureas.

Debate has focused on the effectiveness and safety of oral agents compared with insulin. Jovanovic (2007)64 states that maternal PPG levels are a more significant contributor to fetal hyperinsulinaemia and macrosomia than average glucose levels. She claims that both metformin and sulfonylurea drugs do not control PPG levels as effectively as insulin, and concludes: ‘Oral hypoglycaemic agents may have enough data to prove that they do not harm the fetus; however, they clearly do not achieve the post-prandial glucose response needed to normalise birthweight’. 64

However, Coustan (2007)58 has concluded: ‘Given the available data, glyburide appears to be the best candidate insulin secretagogue for use during pregnancy, since it crosses the placenta little or not at all and benefits the mother directly and the foetus indirectly’. Discussing metformin, he writes: ‘… data suggest that significant amounts of metformin can cross the placenta … Because it is unknown whether metformin is therapeutic or deleterious to the foetus, it would seem prudent to obtain further data (perhaps from animal models) before metformin becomes commonly prescribed during pregnancy. At the very least, patients should be counselled about the unknown risks and benefits for the foetus’.

Coetzee (2007),62 although supportive of the use of oral hypoglycaemic agents in pregnancy, agrees that caution should be exercised with respect to metformin, stating that long-term follow-up studies of infants born to mothers who have taken the drug are required given that metformin is known to cross the placenta.

Homko and Reece (2006)60 conducted a non-systematic review of hypoglycaemic agents during pregnancy. With respect to metformin, they reported:

• There is no evidence of teratogenicity.

• One study which reported that women treated with metformin had a higher prevalence of pre-eclampsia and a high perinatal mortality as compared with women treated with insulin therapy.

• Data regarding the long-term implications of metformin use during pregnancy for offspring are limited.

With respect to sulphonylurea drugs, Homko and Reece (2006)59 concluded: ‘Based on the currently available data, it would appear that glyburide could be safely and effectively utilised in the management of women with GDM’.

Langer (2006),65 in a non-systematic review concluded that: ‘Glyburide is a cost-effective, patient friendly, and potentially compliance-enhancing therapy that produces perinatal outcomes in GDM pregnancies comparable to traditional insulin therapy. For GDM patients who require pharmacologic therapy, glyburide is the drug of choice and only patients who fail to achieve glycaemic control should begin insulin therapy’.

Guidance recently published by NICE allows the use of both glibenclamide and metformin in the management of women who develop hyperglycaemia in pregnancy, although it had to be written before the metformin trials had reported, and when only two glibenclamide trials were published. The GDG expressed caution about the use of glibenclamide, as reported in Chapter 1.

Review of evidence on oral drugs

The objective of this section is to assess the risks and benefits of oral glucose lowering drugs compared with insulin in the treatment of hyperglycaemia in pregnancy.

Results

Three relevant systematic reviews were identified. 2,54,55 One of these54 focused on perinatal outcomes associated with maternal glibenclamide therapy, whereas the other two considered a range of maternal and neonatal outcomes associated with maternal receipt of glibenclamide and other oral agents.

As shown in Table 10, only a minority of studies relevant to our objectives were included within each of the previous reviews. Therefore we undertook a new review of primary studies.

We identified a total of 27 primary studies from the three systematic reviews and additional searches (papers published in full and meeting abstracts), as shown in Table 10. There were four RCTs66–69 and 11 observational studies52,61,70–78 comparing glibenclamide with insulin. Seven observational studies63,79–84 examined (predictors of) glibenclamide failure/success. One of the RCTs67 also included a group receiving acarbose. There were three RCTs85–88 and three observational studies56,70,89 comparing metformin with insulin.

The trials in Table 10 will hereafter be referred to by the name of the first author and year of publication in the figures and text of the results section.

Maternal/obstetric outcomes: primary outcomes

Pre-eclampsia/hypertensive complications

Glibenclamide versus insulin

One RCT by Langer (2005)68 reported no difference in pre-eclampsia after treatment with glibenclamide versus insulin.

Four observational ones gave differing results (Figure 1). One cohort study (Jacobson 200575) reported significantly higher pre-eclampsia rates with glibenclamide than with insulin (12% vs 6%), but no differences were found in the other observational studies (Ramos 200761 which provided data as in Figure 1, and Duncan 200571 and Patterson 200877 which only reported no differences). Taking both RCTs and observational studies together, there were between 6% and 12% of women with pre-eclampsia (where reported).

FIGURE 1.

Pre-eclampsia: glibenclamide versus insulin.

Acarbose versus insulin

Pre-eclampsia rates were not reported for acarbose versus insulin.

Metformin versus insulin

Two RCTs (Hague 200385 and Rowan 200888) reported no difference in pre-eclampsia after treatment with metformin versus insulin (Figure 2).

FIGURE 2.

Pre-eclampsia: metformin versus insulin.

Nor did two cohort studies (Tertii 200856 and Balani 200889) find any difference in pre-eclampsia rates between metformin and insulin.

Overall, there were between 5.5% and 19% of women with pre-eclampsia. Balani (2008),89 Tertti (2008),56 and Rowan (2008)88 found no significant difference in pregnancy-induced hypertension rates between metformin and insulin groups.

Caesarean delivery

Glibenclamide versus insulin

Three RCTs reported no differences in caesarean delivery rates for women treated with glibenclamide versus insulin (Figure 3). The RR was 0.91 (95% CI –0.71 to 1.16).

FIGURE 3.

Caesarean delivery: glibenclamide versus insulin.

Five observational studies (Figure 3 and Patterson 200877) published as full papers reported no significant differences between treatment groups. However, Duncan (2005),71 published as an abstract, reported a significant difference in caesarean sections in favour of glibenclamide for non-morbidly obese women but not for morbidly obese women. The RR was 1.04 (95% CI 0.84 to 1.28) for cohort studies published in full, and 0.53 (95% CI 0.36 to 0.77, p = 0.001) for cohort studies published as abstracts. However, when combining the observational cohort studies published in full with those published as abstracts the RR was not significant, i.e. 0.82 (95% CI 0.59 to 1.14).

Overall, between 14% and 56% of women had caesarean deliveries.

Five per cent of women in the glibenclamide group and 6% of women in the insulin group had an operative vaginal delivery (no significant difference); 7% of deliveries in the glibenclamide group and 9% in the insulin group were assisted vaginal deliveries.

Acarbose versus insulin

Bertini (2005)67 did not report any significant difference in caesarean delivery rates between women receiving acarbose or insulin (53% and 44%), RR 1.18 (95% CI 0.65 to 2.16) (Figure 4).

FIGURE 4.

Caesarean delivery: acarbose versus insulin.

Metformin versus insulin

Three RCTs (Hague 2003,85 Moore 200786 and Rowan 200888) reported on caesarean delivery rates for women treated with metformin versus insulin (Figure 5). Significant heterogeneity was noted. However, the largest study (Rowan 200888) did not demonstrate any appreciable or significant difference between women treated with insulin and those treated with metformin. Between 21% and 63% of women had caesarean deliveries. The overall RR for the RCTs was 1.41 (95% CI 0.77 to 2.58) (but with significant heterogeneity). Rowan (2008)88 reported that 15% of the metformin group and 17% of the insulin group had an emergency caesarean delivery (no significant difference).

FIGURE 5.

Caesarean delivery: metformin versus insulin.

Two observational studies (Tertti 200856 and Balani 200889) reported no significant difference.

Glycaemic control during pregnancy

Glibenclamide versus insulin

For mean BG, 2-hour PPG and HbA_1c, none of the studies found a significant difference between the glibenclamide groups and the insulin groups.

With respect to FBG one cohort study found significantly lower values among women receiving glibenclamide compared with insulin (mean difference –0.41 mmol/l in the study by Jacobson 200575) (Figure 6). However, the RCT by Langer (2000)68 did not find any significant difference among women receiving glibenclamide or insulin. FBG values in the different studies after treatment were between 5.01 and 5.44 mmol/l, 2-hour post-prandial values were between 5.16 and 6.59 mmol/l, mean BG values were between 5.78 and 5.83 mmol/l and HbA_1c was between 5.3% and 5.5%.

FIGURE 6.

Maternal glycaemic control: glibenclamide versus insulin: (a) maternal FPG (mmol/l); (b) maternal 2-hour PG (mmol/l); (c) maternal mean BG; (d) maternal HbA_1c (%).

Acarbose versus insulin

Maternal glycaemic control was not reported for acarbose versus insulin.

Metformin versus insulin

Two RCTs reported on maternal glycaemic control with metformin compared with insulin. Neither of the trials by Moore (2007)86 or Rowan (2008)88 found any significant difference in maternal FBG between metformin and insulin (FBG values between 5.09 and 5.37 mmol/l).

There was no significant difference in 2-hour post-prandial values in Moore (2007). 86 However, Rowan (2008)88 found significantly lower 2-hour postprandial glucose values in the metformin group 1 week after randomisation and overall (mean difference –0.20 mmol/l for both measurements, p = 0.01) but not during the last 2 weeks before delivery. Mean 2-hour post-prandial values were between 5.9 and 6.69 mmol/l in the two studies. Rowan (2008)88 reported no significant difference between treatment groups with respect to HbA_1c values.

Hypoglycaemia

Glibenclamide versus insulin

Three RCTs reported on maternal hypoglycaemia (Figure 7).

FIGURE 7.

Maternal hypoglycaemia: glibenclamide versus insulin.

The trials by Anjalakshi (2007)66 and Bertini (2005)67 reported no maternal hypoglycaemia in either the glibenclamide or the insulin groups (hypoglycaemia requiring hospital admission in Bertini 2005;67 not defined in Anjalakshi 200766). However, Langer (2000)68 found significantly less hypoglycaemia (BG < 2.2 mmol/l) in the glibenclamide group than in the insulin group [20% vs 2%, RR 0.10 (95% CI 0.04 to 0.27), p = 0.03]; none of the women reported severe symptoms.

Two observational studies reported different outcomes. Jacobson (2005)75 found slightly but significantly more hypoglycaemia (values < 3.3 mmol/l) in the glibenclamide group than in the insulin group [0.20% vs 0.08%, RR 2.40 (95% CI 1.41 to 4.07), p = 0.001]. Yogev (2004),78 however, found significantly less asymptomatic hypoglycaemia (BG ≤ 4.0 mmol/l) with glibenclamide than with insulin (in 28% vs 63% of women with 242 vs 46 episodes, p = 0.04); no symptomatic hypoglycaemic episodes were reported.

Acarbose versus insulin

Bertini (2005)67 found no maternal hypoglycaemia requiring hospitalisation for acarbose versus insulin.

Metformin versus insulin

Only the RCT by Moore (2007)86 reported maternal hypoglycaemia after treatment with metformin versus insulin. No cases of maternal hypoglycaemia were seen (hypoglycaemia not clearly defined).

Neonatal/child outcomes: primary outcomes

Hypoglycaemia

Glibenclamide versus insulin

Three RCTs (as shown in Figure 8) examined neonatal hypoglycaemia (see Appendix 4 for the various definitions). Bertini (2005)67 found significantly more neonatal hypoglycaemia with glibenclamide than with insulin (33–34% with glibenclamide vs 4–14% with insulin).

FIGURE 8.

Neonatal hypoglycaemia: glibenclamide versus insulin.

Ogungyemi (2007)69 reported significantly lower neonatal lowest glucose levels with glibenclamide than with insulin (2.65 ± 1.0 mmol/l vs 3.20 ± 1.0 mmol/l, p = 0.028).

In the RCTs overall, there was significantly more neonatal hypoglycaemia (BG < 2.2 mmol/l) with glibenclamide than with insulin [14% vs 7%, RR 2.07 (95% CI 1.04 to 4.11), p = 0.04] (no significant heterogeneity).

Five observational studies also reported neonatal hypoglycaemia. Significance was not quite reached when summarising only the cohort studies (for BG between < 1.4 and 2.6 mmol/l). Rates of neonatal hypoglycaemia over all the studies ranged between 8% and 39%.

Acarbose versus insulin

Bertini (2005)67 reported no significant difference in neonatal hypoglycaemia between acarbose and insulin (one case in each group) (Figure 9).

FIGURE 9.

Neonatal hypoglycaemia: acarbose versus insulin

Metformin versus insulin

Two RCTs studies reported neonatal hypoglycaemia (with metformin versus insulin) (Figure 10). Neither found a significant difference in neonatal hypoglycaemia (below 2.2–2.6 mmol/l) between metformin and insulin. However, Rowan (2008)88 found significantly higher rates of severe neonatal hypoglycaemia (BG < 1.6 mmol/l) with insulin than with metformin (8.1% vs 3.3%, p = 0.008).

FIGURE 10.

Neonatal hypoglycaemia: metformin versus insulin.

One of two cohort studies, Tertti (2008),56 found significantly higher rates of hypoglycaemia (BG < 2.6 mmol/l) with insulin than metformin (58% vs 33%, p = 0.03). Overall values for the two cohort studies were slightly in favour of metformin [RR 0.60 (95% CI 0.38 to 0.95), p = 0.03].

Birthweight

Glibenclamide versus insulin

Four RCTs (including Goodman 2008,74 an abstract without numeric data) examined birthweight after treatment with glibenclamide or insulin (Figure 11).

FIGURE 11.

Birthweight: glibenclamide versus insulin.

Neither the RCTs nor the cohort studies found any significant difference between the groups although the meta-analysis of RCT data demonstrated a borderline increased risk of higher birthweight among infants born to mothers that had received glibenclamide. The WMD was 89.63 g (95% CI –1.48 to 180.75).

Five cohort studies (including one, Goodman 200874 which gave no numerical data) reported birthweight. The WMD was –45.49 g (95% CI –218.36 to 127.37) for observational studies, and –27.6 g (95% CI –115.49 to 60.29) for abstracts (no significant heterogeneity). Goodman (2008)74 reported no difference.

Acarbose versus insulin

No difference in birthweight was seen in the trial by Bertini (2005)67 for after acarbose and insulin treatment (Figure 12). Mean birthweight was between 3151 and 3243 g.

FIGURE 12.

Birthweight: acarbose versus insulin.

Metformin versus insulin

Three RCTs examined birthweight after treatment with metformin or insulin (see Figure 13 and Balani 200889). Neither the RCTs nor the cohort studies found any significant difference between the groups. The WMD was –34.28 g (95% CI –112.80 to 44.24) for the RCTs.

FIGURE 13.

Birthweight: metformin versus insulin.

The study by Rowan (2008)88 also reported a range of other measurements of infant size and found no significant differences between metformin and insulin groups.

There were two cohort studies. Tertii (2008)56 found no difference: –2.00 g (95% CI –254.34 to 258.34) for the observational study. Balani (2008)89 reported no significant difference in an abstract without numerical data.

Macrosomia (birthweight ≥ 4000 g)

Glibenclamide versus insulin

Two RCTs and six cohort studies examined the effect of glibenclamide versus insulin on macrosomia rates (Figure 14).

FIGURE 14.

Macrosomia: glibenclamide versus insulin.

Only the study by Goodman (2008),74 published as an abstract, and using a 4500 g definition, found significantly more neonates weighing more than 4500 g in the insulin than in the glibenclamide group (2% vs 10.3%, p = 0.004). None of the other studies found a significant difference between the treatment groups. The RR was 2.39 (95% CI 0.50 to 11.35) for RCTs, 1.06 (95% CI 0.81 to 1.39) for observational studies, and 0.42 (95% CI 0.20 to 0.90, p = 0.03) for abstracts. Overall, macrosomia was observed in between 0% and 36% of neonates. Other studies: Langer (2006)76 and Patterson (2008)77 found no difference in macrosomia. Duncan (2005)71 reported more in the insulin group – about 23% versus none at all in the glibenclamide group.

Acarbose versus insulin

No macrosomia was seen in the trial by Bertini (2005)67 when comparing acarbose with insulin.

Metformin versus insulin

Two RCTs examined the effect of metformin versus insulin on macrosomia rates (Figure 15). Neither found a significant difference between the treatment groups. Overall, macrosomia was observed in between 9% and 22% of neonates.

FIGURE 15.

Macrosomia: metformin versus insulin.

Two cohort studies, Tertii (2008)56 and Balani (2008),89 also found no difference.

A further observational study was identified that compared oral agents with insulin. It was published only as an abstract with sparse data, but the number of patients was large, and we report it for completeness. Cheng (2006)94 reported on 11,463 women diagnosed with gestational diabetes and enrolled in the California Diabetes and Pregnancy Program in a retrospective cohort study. Of these women, 18.6% were treated with oral agents and 81.4% were treated with insulin. After controlling for potential confounders (unspecified), significantly more macrosomia (birthweight > 4000 g, OR 1.28, 95% CI 1.01 to 1.63) was observed in the oral therapy group. Data on glycaemic control were not provided.

Large for gestational age (> 90th percentile)

Glibenclamide versus insulin

Two RCTs (one the small Bertini67 study; the other the large Langer76 one) examined the effect of glibenclamide versus insulin on rates of LGA babies (Figure 16). The RR was 1.95 (95% CI 0.29 to 13.09).

FIGURE 16.

LGA: glibenclamide versus insulin.

Five cohort studies reported on LGA. None of the studies found a significant difference between the treatment groups, with RR 1.10 (95% CI 0.84 to 1.45) for observational studies, and 1.00 (95% CI 0.53 to 1.88) for the single abstract. Langer (2006)76 reported no difference.

Acarbose versus insulin

No significant difference in rates of LGA neonates was seen in the trial by Bertini (2005)67 when comparing acarbose with insulin (11% with acarbose and 4% with insulin, see Figure 17).

FIGURE 17.

LGA: acarbose versus insulin.

Metformin versus insulin

One RCT and one cohort study examined the effect of metformin versus insulin on rates of LGA neonates (Figure 18). Neither of the studies found a significant difference between the treatment groups. Between 15% and 19% of neonates were LGA.

FIGURE 18.

LGA: metformin versus insulin.

Perinatal mortality

Most studies were underpowered for providing a reliable estimate of perinatal mortality, with almost all reporting no or one perinatal death. Given the small number of studies, meta-analysis cannot be any more helpful. Only the RCT by Langer (2000)68 reported two perinatal deaths per comparison group (Figures 19 and 20). Bertini (2005)67 found no perinatal mortality when comparing acarbose with insulin groups. There was no obvious difference between comparison groups.

FIGURE 19.

Perinatal mortality: glibenclamide versus insulin.

FIGURE 20.

Perinatal mortality: metformin versus insulin.

Birth trauma (e.g. shoulder dystocia, bone fracture, nerve palsy)

Glibenclamide versus insulin

The RCT by Bertini (2005)67 observed no birth trauma in either group.

Neither of the observational studies by Jacobson (2005)75 and Ramos (2007)61 found a significant difference in birth trauma between groups. One cohort study published as an abstract (Gilson 200273) reported no difference in shoulder dystocia between the glibenclamide and the insulin groups (Figure 21).

FIGURE 21.

Birth trauma/shoulder dystocia: glibenclamide versus insulin.

Acarbose versus insulin

Bertini (2005)67 observed no birth trauma when comparing acarbose with insulin groups.

Metformin versus insulin

One RCT (Rowan 200888) reported on birth trauma after treatment with metformin or insulin (Figure 22). No significant difference in birth trauma rates was observed between the comparison groups.

FIGURE 22.

(a) Birth trauma; (b) shoulder dystocia: metformin versus insulin.

Nor did the one cohort study (Tertii 200856) find any difference.

Two RCTs reported no significant difference in shoulder dystocia between the metformin and the insulin groups. Overall, there were between 0% and 3% of shoulder dystocias in the different studies.

Secondary outcomes

Information regarding the secondary outcomes (see Review of evidence on oral drugs, Inclusion criteria) are reported briefly here. A more detailed version is available on request.

Maternal/obstetric outcomes: secondary outcomes

Induction of labour

Glibenclamide versus insulin

Two cohort studies (Holt 200852 and Duncan 200571) reported on rates of induction of labour for glibenclamide versus insulin, with only Holt (2008)52 providing numerical data. There was no significant difference in induction of labour between the groups.

Metformin versus insulin

Three RCTs (Hague 2003,85 Moore 2007,53 Rowan 200888) found no significant difference between the treatment groups.

Maternal weight gain

A few studies reported maternal weight gain.

Glibenclamide versus insulin

Two RCTs reported no difference in maternal weight gain for women treated with glibenclamide versus insulin (Figure 23).

FIGURE 23.

Maternal weight gain: glibenclamide versus insulin.

One of the two observational studies (Jacobson 200575) reported significantly more weight gain with glibenclamide than with insulin [mean difference 1.15 kg (95% CI 0.14 to 2.16)] p = 0.03. The other (Ramos 200761) showed no difference.

Acarbose versus insulin

Bertini (2005)67 reported no significant difference in maternal weight gain during pregnancy for acarbose versus insulin.

Metformin versus insulin

One RCT, Rowan (2008),88 reported significantly more weight gain from enrolment to 36–37 weeks of gestation in the insulin group than in the metformin group [i.e. 2 kg (SD 3.3) vs 0.4 kg (SD 2.9), p = 0.001].

Similarly, one of the two observational studies, Balani (2008),89 reported significantly more weight gain from enrolment until delivery in the insulin than in the metformin group (1.4 kg vs 0.3 kg, p = 0.01). The second observational study, Tertti (2008),56 reported no significant difference in weight gain after the diagnosis of GDM of women receiving metformin or insulin, i.e. 3.0 (SD 3.6) versus 3.5 (SD 5.2) respectively. The mean number of gestational weeks at diagnosis was 24.5 weeks.

Post-partum glucose tolerance

Metformin versus insulin

One trial (Rowan 200888) of metformin versus insulin reported on post-partum glucose tolerance. A 75-mg OGTT was carried out at between 6 and 8 weeks post partum. There was no significant difference in any measures between the metformin and the insulin groups.

An observational study, Balani (2008),89 found no significant difference between metformin and insulin groups in abnormalities in the post-natal glucose tolerance test.

Acceptability of treatment

Only three studies (Chmait 2004,79 Holt 200852 and Rowan 200888) gave some data on patient preference. None of the studies provided a detailed analysis of maternal adverse events considered to be treatment-related but five studies (Holt 2008,52 Jacobson 2005,75 Kahn 2006,63 Rochon 2006,83 Rowan 200888) gave information on discontinuation/change of oral treatment because of adverse events. Details are shown in Table 14.

TABLE 14 - Acceptability of treatment

Study	Treatment discontinued because of adverse effects	Preference
Glibenclamide
Chmait 200479		Women preferred glibenclamide (none chose insulin) – no information on post-treatment satisfaction
Holt 200852	One woman changed from glibenclamide to insulin because of hypoglycaemia and two for other side effects (not specified) In Nasruddin (2009)92 three changed because of unpredictable hypoglycaemia and four for other unspecified reasons	Women treated with glibenclamide felt that their treatment was both convenient and satisfactory The doctors in the clinic felt that there was less flexibility with glibenclamide because of its long action and therefore it was harder to control the hyperglycaemia
Jacobson 200575	~ 19 women (8%) stopped glibenclamide (either switching to insulin or continuing without treatment) for reasons primarily attributed to hypoglycaemia
Kahn 200663	Two women were switched from glibenclamide to insulin because of recurrent hypoglycaemia (BG < 3.3 mmol/l despite dietary manipulation)
Rochon 200683	One woman was switched from glibenclamide to insulin because of symptomatic hypoglycaemia and one because of patient preference
Metformin
Balani 200889	11 women stopped taking metformin (intolerance in four, refusal to continue in seven)
Coetzee 198670	Two women on metformin stopped treatment because of gastrointestinal side effects; lactic acidosis was not seen
Rowan 200888	Seven women (1.9%) on metformin stopped treatment because of gastrointestinal side effects 32 women (8.8%) on metformin had their dose reduced because of gastrointestinal side effects (but 31 maintained a dose of at least 1000 mg/day) Seven women in the metformin group and three women in the insulin group stopped treatment because of withdrawing consent No significant difference in maternal serious adverse events (infection, surgery, pelvic arthropathy) No significant difference in neonatal infection requiring hospitalisation	Questionnaire data indicated that women preferred metformin to insulin Significantly more medication adherence in the insulin than in the metformin group (p < 0.001)

None of the studies reported data on the maternal outcomes of anxiety, depression or health status.

Child/neonatal outcomes: secondary outcomes

Hyperbilirubinaemia/need for phototherapy

Glibenclamide versus insulin

Two cohort studies (Jacobson 200575 and Ramos 200761) and one RCT (Langer 200068) examined the rate of hyperbilirubinaemia in general. As shown in Figure 24, none of the studies showed any significant difference between the glibenclamide and the insulin groups with respect to hyperbilirubinaemia.

FIGURE 24.

(a) Hyperbilirubinaemia/need for; (b) phototherapy: glibenclamide versus insulin.

Four cohort studies (Coetzee 1986,70 Holt 2008,52 Jacobson 200575 and Ramos 200761) examined the rates of hyperbilirubinaemia requiring phototherapy for glibenclamide versus insulin (see Figure 24). Two (Holt 200852 and Jacobson 200575) showed significantly higher rates among infants born to mothers who received glibenclamide rather than insulin. When all four observational studies reporting this outcome were analysed together, there was no significant difference between the treatment groups. However, there was significant heterogeneity as a consequence of one study (Coetzee 198670) reporting in the opposite direction to the others.

Metformin versus insulin

Two RCTs (Hague 200385 and Moore 200753) and two cohort studies (Tertti 200856 and Balani 200889) examined the rate of hyperbilirubinaemia in general. Only the observational study published as an abstract (Balani 200889) found significantly more hyperbilirubinaemia with insulin than with metformin (Figure 25).

FIGURE 25.

(a) Hyperbilirubinaemia/need for; (b) phototherapy: metformin versus insulin.

Two RCTs (Hague 200385 andRowan 200888) and two cohort studies (Coetzee 198670 and Tertti 200856) examined the rates of hyperbilirubinaemia requiring phototherapy for metformin versus insulin. Figure 25 shows that there was no significant difference between the two groups in any study.

Small for gestational age (< 10th percentile)

Glibenclamide versus insulin

One RCT reported an RR of 0.22 (95% CI 0.01 to 4.45) for SGA (Figure 26).

FIGURE 26.

SGA: glibenclamide versus insulin.

None of three cohort studies found a significant difference between the treatment groups [RR 0.78 (95% CI 0.42 to 1.45), no significant heterogeneity]. Between 0% and 13% of neonates were SGA.

Acarbose versus insulin

No significant difference in rates of SGA neonates was seen in the trial by Bertini (2005)67 when comparing acarbose with insulin (0% with acarbose and 7% with insulin) – see Figure 27.

FIGURE 27.

SGA: acarbose versus insulin.

Metformin versus insulin

One RCT and two cohort studies examined the effect of metformin versus insulin on rates of SGA neonates (Figure 28). None of the studies found a significant difference between the treatment groups. The RR was 0.74 (95% CI 0.45 to 1.19) for the RCT and 1.39 (95% CI 0.56 to 3.50) for the observational studies (no significant heterogeneity). Between 2% and 19% of neonates were SGA.

FIGURE 28.

SGA: metformin versus insulin.

Neonatal ICU admission

Glibenclamide versus insulin

One RCT found no difference in NICU admission (RR 0.87, 95% CI 0.41 to 1.83).

Of six cohort studies reporting on NICU admission after treatment with glibenclamide or insulin (Figure 29 and below) only one, Jacobson (2005),75 found significantly more NICU admission in the insulin group than in the glibenclamide group (24% vs 15%, p = 0.008); however, NICU length of stay was significantly longer in the glibenclamide group (8.0 ± 10.1 days vs 4.3 ± 9.6 days with insulin, p = 0.002). None of the other studies found a significant difference, and there was no significant difference overall [0.75 (95% CI 0.53 to 1.05)]. The rate of NICU admission was between 6% and 25%. Other studies: Fines (2003),72 Goodman (2008)74 and Langer (2006),76 found no significant differences.

FIGURE 29.

NICU admission: glibenclamide versus insulin.

Acarbose versus insulin

Bertini (2005)67 found no NICU admissions when comparing acarbose with insulin groups.

Metformin versus insulin

Two RCTs reported no difference in NICU admissions or length of stay (Figure 30).

FIGURE 30.

0 NICU admission: metformin versus insulin.

One cohort study by Balani (2008),89 published as an abstract, found significantly more admission to the NICU in the insulin group than in the metformin group (19% vs 5%, p = 0.01). The other, Tertti (2008),56 found no significant difference in treatment days spent at the NICU although the CIs only just overlapped with no difference (RR 0.68, 95% CI 0.45 to 1.02).

Congenital malformations

The presence of congenital malformation might suggest pre-existing diabetes rather than GDM, since the key period for teratogenesis is in the first trimester. However, we give the figures below for reassurance.

Glibenclamide versus insulin

Two RCTs and three cohort studies reported congenital malformations or anomalies after treatment with glibenclamide or insulin (Figure 31). There were no significant differences in congenital malformations or anomalies between the comparison groups (rates of malformations/anomalies 0% to 10%). In the study by Ramos (2007),61 congenital abnormalities for infants in the glibenclamide group included patent ductus arteriosus, ventricular septal defect, atrial septal defect, inguinal hernias, intestinal atresia and spine anomaly. No details were reported in the other studies.

FIGURE 31.

Congenital malformations/anomalies: glibenclamide versus insulin.

Metformin versus insulin

One RCT and three cohort studies reported congenital malformations or anomalies after treatment with metformin or insulin (see Figure 32 and below). There were no significant differences in congenital malformations or anomalies between the comparison groups (rates of malformations/anomalies 0% to 10%). Other studies: Balani (2008)89 showed no significant difference in rate of congenital malformations.

FIGURE 32.

Congenital malformations/anomalies: metformin versus insulin

Respiratory distress

Glibenclamide versus insulin

The Langer (2000)68 RCT reported no difference between groups, and the Bertini (2005)67 trial reported that there were no reports of respiratory distress in the glibenclamide group (insulin not reported).

Langer (2000)68 reported lung complications in 8% of infants in the glibenclamide group and 6% of infants in the insulin group (no significant difference).

Four cohort studies reported on respiratory distress (Holt 200852) and oxygen/assisted ventilation (Jacobson 200575 and Ramos 200761) (Figure 33). None of the studies reported a significant difference between groups (rates 2–9%)

FIGURE 33.

Respiratory distress/respiratory support: glibenclamide versus insulin.

Metformin versus insulin

Two RCTs reported no differences in respiratory distress after treatment with metformin or insulin. One cohort study reported the same (Figure 34).

FIGURE 34.

Respiratory distress: metformin versus insulin.

Apgar scores

Glibenclamide versus insulin

Apgar scores at 1 and 5 minutes were reported by one RCT and four cohort studies (see Figure 35 and below). Holt (2008)52 found a significantly higher 1-minute Apgar score in the insulin group than in the glibenclamide group (8.2 vs 7.3, p = 0.05). The RCT by Bertini (2005)67 found a significantly higher 5-minute Apgar score (9.4 vs 9.0, p = 0.03) in the insulin group than in the glibenclamide group.

FIGURE 35.

Apgar scores: glibenclamide versus insulin. (a) 1-minute Apgar score; (b) 5-minute Apgar score.

The Fines (2003),72 Goodman (2008)74 and Paterson (2008)77 studies reported no significant differences in any Apgar scores.

Acarbose versus insulin

Bertini (2005)67 did not find any significant differences in Apgar scores at 1 or 5 minutes between the acarbose and the insulin groups. Apgar scores at 1 minute were between 8.1 and 8.4, and Apgar scores at 5 minutes were between 9.3 and 9.4 (Figure 36).

FIGURE 36.

Apgar scores: acarbose versus insulin. (a) 1-minute Apgar score; (b) 5-minute Apgar score.

Metformin versus insulin

In the RCT by Rowan (2008),88 three neonates in the metformin group (0.8%) and one (0.3%) in the insulin group had 5-minute Apgar scores below 7; all these infants had Apgar scores of 6.

Apgar scores at 5 minutes were reported by one RCT and one cohort study (Figure 37). None of the studies found a significant difference between metformin and insulin. Mean Apgar scores at 5 minutes ranged from 8.6 to 9.0.

FIGURE 37.

5-minute Apgar score: metformin versus insulin

Timing of delivery

Glibenclamide versus insulin

The RCT by Bertini (2005)67 reported that there were no reports of prematurity in the glibenclamide group (insulin not reported), and two cohort studies (Figure 38) found no significant difference between comparison groups.

FIGURE 38.

Preterm delivery: glibenclamide versus insulin.

Three RCTs and five cohort studies reported on gestational age at delivery (see Figure 39, Duncan 200571 and Goodman 200874) reported no significant differences between the glibenclamide and the insulin groups (WMD –0.06 weeks, 95% CI –0.26 to 0.15) (no significant heterogeneity). Gestational age at delivery ranged from 37.6 to 39 weeks.

FIGURE 39.

Gestational age at delivery.

Acarbose versus insulin

Rates of preterm deliveries for acarbose versus insulin were not reported. The age of gestation at delivery was 38.2 ± 1.2 weeks in the acarbose group and 38.5 ± 1.2 weeks in the insulin group (no significant difference).

Metformin versus insulin

One RCT and two cohort studies reported on premature deliveries after treatment with metformin or insulin (Figure 40). Rowan (2008)88 found a significantly higher rate of preterm delivery with metformin than with insulin (12.1% vs 7.6%, p = 0.04); in both groups, about half the preterm births were iatrogenic (indicated) and half were spontaneous.

FIGURE 40.

Preterm delivery: metformin versus insulin.

The cohort study by Balani (2008),89 published only as an abstract, found significantly more preterm deliveries in the insulin group than in the metformin group (0% vs 11%, p = 0.01). No significant difference in premature deliveries between groups was observed in the study by Tertti (2008). 56

Three RCTs reported on gestational age at delivery (Figure 41). A meta-analysis of the RCTs showed a significant difference (p = 0.03) of –0.21 weeks (95% CI –0.40 to –0.02) lower gestational age with metformin compared with the insulin group.

FIGURE 41.

Gestational age at delivery: metformin versus insulin.

One observational study (Tertii 200856) showed a non-significant difference of 3 weeks’ older gestational delivery with metformin.

Other outcomes/data

Appendix 9 contains information on additional outcomes reported within studies included within this review. In addition, Appendix 9 also contains information from studies which did not meet the inclusion criteria for our review but nevertheless are of interest with respect to the treatment of hyperglycaemia in pregnancy.

Glibenclamide versus metformin

The small RCT by Moore (2005),95 reported as an abstract, studied 46 women with gestational diabetes treated either with metformin (n = 22) or glibenclamide (n = 24) (Table 15).

TABLE 15 - Results of the Moore (2008)96 RCT of metformin versus glibenclamide

NS, not significant.

Outcome	Metformin	n	Glibenclamide	n	p
Maternal
Treatment failure	41.2%	63	19.6%	61	0.01
Severe pre-eclampsia	n = 2	63	n = 3	61	NS
Caesarean delivery (non-elective)	n = 11	63	n = 2	61	0.02
FBG (mmol/l ± SD)	5.1 ± 0.7	22	5.0 ± 0.4	24	NS
2-hour PPG breakfast (mmol/l ± SD)	5.6 ± 0.8	22	5.4 ± 0.4	24	NS
2-hour PPG lunch (mmol/l ± SD)	5.9 ± 0.8	22	6.6 ± 0.8	24	NS
2-hour PPG dinner (mmol/l ± SD)	6.1 ± 0.8	22	6.1 ± 0.4	24	NS
Maternal hypoglycaemia (BG < 3.3 mmol/l)	n = 2	63	n = 1	61	NS
Neonatal
Neonatal hypoglycaemia	n = 1	63	0	61	NS
Hyperbilirubinaemia		22		24	NS
Macrosomia	n = 1	63	n = 4	61	NS
NICU admission	n = 4	63	n = 1	61	NS
Shoulder dystocia	0	63	n = 1	61	NS
Respiratory distress		22		24	NS
5-minute Apgar score < 7	0	63	0	61	NS

No significant differences between the groups were seen in glycaemic control or outcomes. Another abstract by Moore (2008)96 appears to present an extension of the RCT previously reported, now with 63 patients on metformin and 61 on glibenclamide (see Table 10). Outcomes remain similar, apart from there being more non-elective caesarean deliveries with metformin than with glibenclamide [17% (three breech presentation, eight non-reassuring fetal status) vs 3% (one failure to progress, one non-reassuring fetal status), p = 0.02].

Oral medication failure

The issue here is whether it is possible to predict failure on oral agents, in which case they would not be used, but patients would go straight to insulin once diet and physical activity failed.

Appendix 10 contains details of studies considering failure of oral medication.

Failure to achieve adequate glycaemic control on glibenclamide

Between 4% and 24% of women receiving glibenclamide were considered to fail treatment and were switched to insulin. The RCT by Langer (2000)68 reported that 4% of patients in the glibenclamide group were switched to insulin because of glibenclamide failure. However, 18% of women in the glibenclamide group had self-monitored BG values that were not in the desired range. The cohort study by Langer (2006)81 suggested that glibenclamide failure was also related to glibenclamide dose: with a dose of less than 10 mg/dl, 60% achieved mean BG targets and 76% achieved 2-hour PPG targets, while when including patients receiving a dose of more than 10 mg/dl, 85% achieved target glycaemic control (p = 0.00002 for dose effect).

Most studies did not report whether glycaemic control was adequate with insulin. However, data from some studies suggest that control with insulin was not necessarily more reliable. Langer (2000)68 report that 12% of women in the insulin group had self-monitored BG values that were not in the desired range. When studying glycaemic values of a subsample of 122 women on glibenclamide and 137 women on insulin in their cohort study, Jacobson (2005)75 found that mean fasting and/or post-prandial values were within goal for 86% in the glibenclamide group but only for 63% in the insulin group (p = 0.001). Gilson (2002)73 reported similar numbers of patients not achieving adequate control with glibenclamide or insulin. While Holt (2008)52 reported that all of the women switched to insulin because of glibenclamide failure achieved adequate control, Conway (2004)80 reported that 67% of the patients switched to insulin also failed to achieve adequate control with insulin, as did 55% of glibenclamide failures in the study by Kahn (2006). 63

Failure to achieve adequate glycaemic control on metformin

Of the metformin studies, Moore (2007)53 reported no cases of metformin failure, while Tertti (2008)56 reported a failure rate of 18%. In the RCT by Rowan (2008)88 46.3% of women on metformin required supplemental insulin, and in the study by Balani (2008)89 11% of participants on metformin required additional insulin. In the RCT by Moore (2008),86 comparing metformin with glibenclamide, significantly more women failed metformin therapy than glibenclamide therapy (41.2% vs 19.6%, p = 0.01). None of the studies reported whether adequate control was achieved with insulin.

Factors associated with failure to achieve adequate glycaemic control on oral medication

Only five studies (Holt 2008,52 Kahn 2006,63 Chmait 2004,79 Conway 200480 and Rochon 2006,83 all cohort studies) reported some comparison data for glibenclamide success versus glibenclamide failure groups, and only the RCT by Rowan (2008)88 compared characteristics of women with adequate control on metformin alone and women requiring supplemental insulin. The factors analysed most frequently were maternal age, parity, maternal BMI, history of gestational diabetes, family history of diabetes mellitus, gestational age at diagnosis of gestational diabetes, and results of the OGTT. Maternal and neonatal outcomes following treatment for those studies (Kahn 2006,63 Chmait 2004,79 Conway 200480 and Rochon 200683) only comparing glibenclamide-only groups with those switching to insulin are reported above.

Maternal age

Six studies (Chmait 2004,79 Conway 2004,80 Kahn 2006,63 Rochon 2006,83 Holt 2008,52 Langer 200681) reported maternal age for glibenclamide failure and glibenclamide success groups. Kahn (2006)63 found that mothers in the glibenclamide failure group were significantly older than mothers in the success group (mean difference 5 years, 95% CI 2.43 to 7.57). Langer (2006)81 reported that older age was a predictor of failure but gave little detail. The other four studies found no difference. The age range in the failure groups was 31–34 years, and in the success groups 29–32 years.

Parity

Four studies (Chmait 2004,79 Kahn 2006,63 Conway 2004,80 Rochon 200683) reported on parity or number of multiparous women for glibenclamide failure versus glibenclamide success. There was a tendency for higher parity and a larger proportion of multiparous women in the failure group; however, this only reached significance in the study by Kahn (2006)63 (parity 2 ± 1.7 in the failure group vs 1 ± 1 in the success group, p = 0.03; similar result for gravidity). Parity was between 2 and 2.2 in the failure groups and 1 and 1.7 in the success groups. Between 86% and 92% in the failure groups and 70% and 86% in the success groups were multiparous.

In the RCT by Rowan (2008),88 significantly more women requiring supplementary insulin were multiparous than women with adequate control on metformin only (76% vs 61.5%, p = 0.003).

Maternal BMI

Maternal BMI was reported in three studies comparing glibenclamide failure with glibenclamide success. One of those studies found a significantly higher BMI in the failure group than in the success group (33.2 ± 5.4 vs 28.8 ± 5.8 kg/m², p = 0.02). BMI was between 32 and 33.2 kg/m² in the failure groups and between 28.8 and 31.5 kg/m² in the success groups. Langer (2006)81 found no significant difference in obesity between glibenclamide failure and success groups.

Rowan (2008)88 reported significantly higher BMI values (in early pregnancy) for those requiring additional insulin than those with adequate control on metformin alone (33.6 ± 8.6 vs 31.1 ± 7.8 kg/m², p = 0.01). BMI values at enrolment were also significantly higher in those requiring supplemental insulin (p = 0.001).

History of gestational diabetes

None of five studies (Chamit 2004,79 Conway 2004,80 Kahn 2006,63 Rochon 2006,83 Parrish 200882) found a significant difference in rates of history of gestational diabetes between the failure and the success groups (RR 1.42, 95% CI 0.85 to 2.37 for full publications). Rates of history of previous gestational diabetes ranged between 14% and 58% in the failure groups and 13% and 41% in the success groups.

The RCT by Rowan (2008)88 reported significantly higher rates of history of gestational diabetes for those requiring additional insulin than for those with adequate control on metformin alone (33.3% vs 19.5%, p = 0.009).

Family history of diabetes

Four studies (Chmait 2004,79 Conway 2004,80 Rochon 2006,83 Parrish 200882) reported on family history of diabetes mellitus in women failing glibenclamide versus those with adequate glycaemic control with glibenclamide. Three of the studies found no significant difference in rates of family history of diabetes between the failure and the success groups (in their original analyses). In the study by Rochon (2006),83 marginally more women with a family history of diabetes were seen in the success group than in the failure group (52% vs 14%, p = 0.076 in the original analysis). Rates of family history of diabetes were between 14% and 93%.

The RCT by Rowan (2008)88 reported significantly higher rates of women with a maternal family history of diabetes among those requiring additional insulin compared with those with adequate control on metformin alone (52.4% vs 38.0%, p = 0.006).

Gestational age at diagnosis

Four observational studies (Figure 42) found that women failing glibenclamide were diagnosed at a significantly earlier gestational age (14–22.7 weeks vs 25–28 weeks, p = 0.003) and overall (WMD –3.94 weeks, 95% CI –6.42 to –1.47, p = 0.002). An abstract by Parrish (2008)82 showed a large difference in gestational age at diagnosis between oral success and oral failure groups (14 weeks vs 25 weeks, p = 0.00001).

FIGURE 42.

Gestational age (weeks) at GDM diagnosis: oral failure versus oral success.

Women were diagnosed at between 14 and 22.7 weeks of gestation in the failure group and between 20 and 28 weeks of gestation in the success group.

Glucose levels

Five studies (Figure 43) examined FBG on the OGTT as a predictor of glibenclamide failure. The study by Kahn (2006)63 found a significant difference, with women in the failure group having significantly higher FBG values (6.22 vs 5.55 mmol/l, p = 0.045). The study by Langer (2006)81 published in abstract form only also found FPG to be a predictor of glibenclamide failure but provided no further data. Overall, FBG values were significantly higher in the failure group than in the success group (WMD 0.49 mmol/l, 95% CI 0.19 to 0.79) (no significant heterogeneity). FBG values were between 5.8 and 6.4 mmol/l in the failure group and between 5.2 and 5.7 mmol/l in the success group.

FIGURE 43.

FBG (mmol/l) in OGTT as predictor of glibenclamide failure.

Rowan (2008)88 also found significantly higher FBG values on the OGTT for the women requiring additional insulin compared with those with adequate control on metformin alone (6.1 vs 5.3 mmol/l, p < 0.001). Similarly, enrolment FBG and HbA_1c were significantly higher in those subsequently requiring supplementary insulin (p = 0.001).

Three studies reported on 1-hour, 2-hour and 3-hour OGTT values in relation to glibenclamide failure or success (Figure 44). In their original analyses, Chmait (2004)79 found no significant difference in the OGTT values between glibenclamide failure and success groups at any time point; Conway (2004)80 found significantly higher values in the failure group than in the success group at all time points (p = 0.001), and Rochon (2006)83 found significantly higher values at 1 hour in the failure group, no difference between groups for the 2-hour value, and significantly lower 3-hour values for the failure than the success group. Overall, 1-hour values were significantly higher in the failure group than in the success group (WMD 1.20 mmol/l, 95% CI 0.56 to 1.84, p = 0.0002), as were 2-hour values (WMD 1.11 mmol/l, 95% CI 0.44 to 1.77, p = 0.001); there was no significant difference for 3-hour values. Values at 1 hour ranged from 11.4 to 12.8 mmol/l in the failure group and 11.0 to 11.4 mmol/l in the success group; values at 2 hours ranged from 10.5 to 11.3 mmol/l in the failure group and 9.4 to 9.9 mmol/l in the success group; values at 3 hours ranged between 6.33 and 9.8 mmol/l.

FIGURE 44.

OGTT results: oral failure versus oral success. (a) 1-hour BG (mmol/l); (b) 2-hour BG (mmol/l); (c) 3-hour BG (mmol/l).

Additionally, Holt (2008)52 reported that there was no difference in baseline glycaemia between the women in their study failing glibenclamide and those who did not (no values given).

Other studies: Langer (2006)81 also reported that FPG was a predictor of glibenclamide failure, p = 0.007.

In addition to the results presented in Figure 44, Chmait (2004)79 found that there were significantly more women in the glibenclamide failure group who had had an infant with macrosomia in a previous delivery (54% vs 22%, p = 0.02). No significant difference in ethnicity was found. During treatment, both FBG and 1-hour post-prandial value were significantly lower in the glibenclamide success than in the failure group (FBG: 4.9 ± 0.6 mmol/l vs 6.3 ± 0.9 mmol/l, p = 0.001; 1-hour PPG: 6.9 ± 0.7 mmol/l vs 8.0 ± 1.1 mmol/l, p = 0.001). Results of the receiver-operator curve (ROC) analysis gave the following predictors for glibenclamide success: (1) dietary therapy failure after 30 weeks’ gestation, and (2) dietary therapy failure at less than 30 weeks’ gestation with mean FBG ≤ 6.1 mmol/l (110 mg/dl) and mean 1-hour PPG ≤  7.8 mmol/l (140 mg/dl), with a sensitivity of 98% and a specificity of 65%.

Conway (2004)80 found no clear cut-off levels in fasting glucose levels on the 3-hour OGTT for predicting glibenclamide failure. At FBG values of 5.3 mmol/l or more, 92% of women converting to insulin were detected, but with a false-positive rate of 70%. At 5.8 and 6.1 mmol/l, sensitivities were 67% and 50%, and false-positive rates were 40% and 29%, respectively.

Ethnicity

Rowan (2008)88 found that there was a significant difference in ethnicity between those with adequate glycaemic control on metformin alone and those requiring supplementary insulin, with significantly fewer Polynesians (11.8% vs 29.8%, p = 0.001) and significantly more Chinese/south-east Asians (16.9% vs 9.5%, p = 0.04) among those managing on metformin alone. Significantly more women in the group receiving supplementary insulin had had three or more terminations or miscarriages (33.9% vs 13.8%, p = 0.001), presumably because such a history was an indication for tight control. Langer (2006)81 also found a significant difference in ethnicity between women failing glibenclamide and women with adequate control on glibenclamide (p = 0.00001, no further details given).

Medication dose

In the study by Langer (2006),81 glibenclamide dose (more or less than 10 mg/dl) was also a significant variable in predicting glibenclamide success or failure (p = 0.00002, no further details given).

Maternal and neonatal outcomes

Most studies on glibenclamide success or failure found no significant difference in maternal or neonatal outcomes between the success and the failure groups (see above).

In contrast, Rochon (2006)83 found a significantly higher NICU admission rate in infants of mothers in the glibenclamide success group than in the failure group. In a regression analysis they found that glibenclamide was predictive of NICU admission without other potential risk factors (including maximum dose used and length of glibenclamide use) playing a role. Of the studies of glibenclamide versus insulin, Holt (2008)52 reported no apparent difference in birth outcomes between women failing glibenclamide in their study and those who did not. Similarly, Rowan (2008)88 reported no significant difference in their primary outcome composite and any of its components (neonatal hypoglycaemia, respiratory distress, phototherapy, birth trauma, Apgar score < 7 at 5 minutes, prematurity) when comparing women on metformin requiring supplementary insulin with those who did not.

Screening studies linked to outcomes

One problem with screening studies is that they often compare the performance of a screening test with the OGTT, rather than with patient outcomes.

The US Preventive Services Task Force looked specifically for studies which related results of screening test to outcomes of pregnancy, rather than to other measures of glucose such as the OGTT. Only two were found. 44 One was that by Cheng et al. (2007)39 using GCT data, already described in Chapter 1.

The other was by Dodd et al. (2007)40 from Adelaide (but excluding women who took part in ACHOIS). Women were screened with the 50-g GCT, and those who had PG of 7.8 mmol/l or more, went on to have a 75-g OGTT. Those who were GCT-negative did not have OGTTs. Using the GCT and OGTT results, women were divided into groups:

• 1-hour GCT < 7.8, 7.8–11.0, and > 11.0 mmol/l

• FPG < 5.5, 5.5–7.0, and > 7.0 mmol/l

• 2-hour OGTT < 7.8, 7.8–11.0, and > 11.0 mmol/l.

Those with FPG of 7.0 mmol/l or over, or a 2-hour OGTT level of 11 mmol/l or more, were classed as having GDM and treated. Those in the intermediate group were diagnosed as having ‘mild’ GDM, and treated with diet, BG monitoring and insulin if necessary. This complicates interpretation, but some adverse outcomes increased in line with glucose levels despite treatment.

As with the HAPO study, Dodd et al. (2007)40 found that:

‘Our study indicates a continuum of adverse pregnancy outcomes with increasing levels of glucose intolerance. Using the information presented in the ROC curves, there does not appear to be a plasma glucose concentration above which the risk of adverse outcomes increases.’

Problem: relative risks and absolute numbers

A common finding in public health is that most events happen in people at low risk, because there are more of them. This is seen in the HAPO study. 19

It could be argued that the ‘purest’ of the HAPO primary outcomes was cord blood C-peptide. Some would justifiably argue that it is just a biochemical measurement, not an outcome which matters to patients, but of the others, macrosomia is a mixture of normal and abnormal big babies, delivery by section reflects clinical practice (range 8.6–23.5%), and neonatal hypoglycaemia may be affected by recording practice (range 0.3–6.4%). However, cord blood C-peptide was presumably a strong indicator of adverse outcomes.

The C-peptide results can be expressed in different ways as shown in Table 16, using FPG as the glucose measure.

So most of the adverse outcomes are in the low risk groups, with almost 60% being in the three lowest risk groups.

Let us assume that intervention could reduce adverse outcomes by half (a conservative assumption, if we look at ACHOIS18). Some exploratory calculations are shown in Table 17. Column 2 of the table shows the proportion of adverse outcomes which would be in each HAPO category, if reduced by half. So 12% of all such outcomes would be in category 2, and so on. Column 3 shows the proportion of the outcomes avoided by category. Over half of all the gains would be in categories 2 and 3. This is because their risks may be low, but their numbers are high. Column 4 shows the numbers needed to treat to avoid one adverse outcome, which is inversely related to the risk of outcomes.

However, the number needed to treat would be much greater in the lower categories, and hence the cost per adverse event avoided would be higher, as column 5 shows – it would cost 8.3 times as much to avoid an event in category 1 as in category 7. Expressed differently, if it cost £100 to prevent an adverse event, and we had £1000, we could prevent 10 events in category 7 or one in category 1.

Note that treatment is not the only cost. It is much more cost-effective to treat women in category 7, but it is much more expensive per each case identified. Using the numbers in the C-peptide column in Table 2 (HAPO glycaemia categories), we would need to screen 107 women to find each one in category 7, compared with only 4 in category 3. However, that assumes universal screening, and the cost of finding the category 7 women would be reduced by selective screening – though not all might be found. This issue is dealt with in Chapter 4.

Implication: if we are only going to treat women at higher risk, we need to define how high that risk has to be, and consider the most efficient screening strategies for detecting those women. For example, if we were only going to treat HAPO categories 6 and 7, we need to know how many of those women would be identified by different degrees of selectivity of screening, for example if only women with BMI over 30 had glucose measured. The HAPO study has not yet published data on the age and BMI predictors of the higher glucose categories.

Choice of test

As mentioned in Chapter 1, Agarwal and Dhatt (2007)11 reviewed the variation in diagnostic criteria. They also considered choice of test. They were not impressed by the OGTT:

‘It is expensive and time-consuming. It is non-physiologic, unpleasant, not reproducible, unrelated to body weight, and its predictive value may vary with ethnic origins.’

However they noted that it was still the diagnostic test recommended by the expert panels, despite the lack of agreement on the glucose load (75 g vs 100 g – they could also have mentioned 50 g being used in Australia), as well as on cut-offs.

Early screening

Screening for GDM has traditionally been done in around the start of the third trimester, at 24–28 weeks’ gestation. The rationale is that insulin resistance is due to placental hormones (lactogen) which reach their highest level around then. Several commentators have raised concerns about leaving it so late. The American College of Obstetrics and Gynecology (ACOG) summed the problem up in its 2001 guidelines:105

‘Insulin resistance increases as pregnancy progresses, therefore, testing later in pregnancy will result in a higher yield of abnormal tests. However, the later the abnormality is diagnosed, the less time will be available for intervention.’

The ACOG guideline commented that the recommendation to screen at 24–28 weeks was an arbitrary decision.

Riskin-Mashiah et al. (2009)106 point out that the standard two-stage process of screening and diagnostic testing takes time away from the brief therapeutic window for treatment. They suggest that this may be one reason why intervention has often failed to have much effect on outcomes – it is coming too late. They cite the study by Schaefer-Graf et al. (2003)107 which reported that 20% of fetuses showed macrosomia on ultrasound at diagnosis of GDM. Riskin-Mashiah et al. (2009)106 argue that the optimum time for lifestyle intervention might be at the end of the first trimester, after morning sickness has abated and when renewed appetite may lead to excessive weight gain and an increased risk of GDM.

Riskin-Mashiah et al. (2009)106 carried out a retrospective study of 6129 sets of case notes of women delivered over a 5-year period and who had a first trimester FPG level recorded, the mean gestation at time of FPG being 9.5 weeks. 106 They divided FPG into seven bands similar to HAPO, and compared rates of development of GDM, LGA, macrosomia and caesarean section. Their results are shown in Table 18.

Hence, the frequency of later GDM is low until after band 4. However, the relationship between first trimester FPG and LGA/macrosomia is more linear, as is the frequency of caesarean section.

One weakness of this study for our purposes is that there were no post-natal data on maternal glycaemic status. Some of those with early hyperglycaemia might have had pre-gestational T2DM rather than GDM. However, the study supports the HAPO findings of a continuum of risk.

Bartha et al. (2000)108 in Spain screened 3986 women at the first antenatal visit, using a 50-g GCT and following up levels over 140 mg/dl with a 100-g OGTT. They then divided women who developed GDM into two groups, early diagnosis (mean 18 weeks gestation) and late (mean 33 weeks). In practice the late group were those whose early tests had been normal but who were diagnosed after being re-screened at 24–28 weeks. (NB the delay to diagnosis at 33 weeks supports the concern of Riskin-Mashia et al. (2009)106 about missing the key period for intervention.)

Bartha et al. (2000)108 reported that 5.9% of women had GDM, of whom 28% were in the early diagnosis group. This group had higher pre-gestational BMI than the late diagnosis group – 29 versus 25. No data on post-natal glucose levels are given, so again it is possible that some of the early diagnosis group had T2DM before pregnancy. They were more likely to require insulin (34% vs 7% for the late onset group) and to have had a history of hypertension (19% vs 6%).

Most et al. (2009)109 carried out a 50-g GCT at first visit (mean gestation not given: stated to be first trimester, and to be prior to 16 weeks) and in those who were negative, again at 24–28 weeks. Of 340 women diagnosed with GDM (100-g OGTT), 29% were diagnosed after their first visit. The early diagnosis group were older (34 years vs 30 years) and heavier (BMI 28 vs 26: percentages with BMI < 25, 24% and 50%). Women with early diagnosis of GDM were more likely to require drug treatment than the late onset group (45% vs 19%). Once again, no post-natal glucose data were given, so some of the early diagnosis group may have had pre-gestational diabetes.

Seshiah et al. (2008)110 compared glycaemic indicators and outcomes in women diagnosed at different gestational ages. All had 75-g OGTTs and HbA_1c. However, it is not clear whether those diagnosed later in pregnancy had been screen-negative earlier. They were a group at higher than usual risk, as shown by their eventual cumulative prevalence of GDM of 42%, compared with the host population’s 17%. The study was based in Chennai, India. Women known to have had pre-gestational diabetes were excluded.

Seshiah et al. (2008)110 divided the women into five groups, the first four being based on gestational age at diagnosis of GDM, and the fifth being women who had normal glucose tolerance. The WHO criterion used for the diagnosis was 2-hour PG 140 mg/dl or over. Their bands and other details are shown in Table 19.

The HbA_1c was done at the same time as the OGTT. The high HbA_1c in group 1 suggests that these women had had undiagnosed T2DM before pregnancy.

The birthweight figures suggest that early diagnosis may allow more effective intervention. Sixty-two percent were diagnosed before 24 weeks of gestation, but it should be remembered that this was a high risk group. However, perhaps high risk women should be screened earlier in pregnancy. The slight rise in HbA_1c from groups 2 to 4, might suggest that the later onset group could have been hyperglycaemic for longer.

Could HbA_1c be used as a screening test? There is a clear difference between all the GDM groups, but confidence intervals are wide. Or could an early HbA_1c at time of booking be used for triage of women into four groups:

• 6.5% or over – assume diabetes

• 6.0% to 6.4% – treat as HGP

• under 5.3% – assume that HGP can be ruled out, with no further testing

• 5.4 to 5.9% – re-test in 3rd trimester?

Agarwal and Dhatt (2007)11 report that glucose screening in early pregnancy can detect most cases of GDM, but that FPG is not suitable, since in the early stages, as with T2DM, hyperglycaemia is only post prandial. Hence the 50-g GCT is recommended, at least in high risk populations.

Sacks et al. (2003)111 also investigated the value of an early (mean 10 weeks’ duration) FPG and concluded that it was an unsatisfactory test because of poor specificity.

Selective or universal screening

NICE recommended selective screening based on the ADA criteria, or on high risk ethnicity. 2 This recommendation was based on the probability of being diagnosed with GDM on the basis of the 75-g OGTT, and so belongs to the dichotomy era.

What we need to know now is the intervention threshold, in terms of HAPO category, and how good selection criteria are in terms of negative predictive value. If we used, say, both age under 25 years and BMI under 28 as exclusion criteria from selective screening, how many women in HAPO categories 5–7 would be missed? The HAPO study19 has the data but this analysis has not yet been carried out.

The 2002 HTA report3 concluded that selective screening by risk factors would miss about half of the women with GDM. A recent before and after study from Paris, Cosson et al. (2006),112 comparing selective versus universal screening, reported that GDM was diagnosed in 8.3% and 12.6% respectively. An odd feature in this study was that outcomes were as good as, and sometimes slightly better, in those women with GDM found by universal screening than in non-diabetic controls. In addition to treatment to control glucose levels, they also had enhanced monitoring of blood pressure, weight, proteinuria and fetal heart rates.

This is reminiscent of the ACHOIS trial,19 where the intervention group did better than the non-diabetic population norm, suggesting that the improvement may be due to more than just glucose control, but may, as pointed out by Masson and Lindow (2006),113 reflect provision of a more comprehensive package of care.

Conclusions

Most studies published since the previous HTA review3 have compared a number of screening tests with the presence or not of GDM (variably defined) based on the OGTT (various forms). These are not that applicable in the post-HAPO era, where the key questions are:

1. At what level of PG is intervention worthwhile? This could be expressed in terms of the HAPO categories.

2. What measure of PG should we use?

3. Can de-selection by absence of risk factors remove many women from BG testing?

FPG is looking as if it might suffice, thereby avoiding the costs and inconvenience of the OGTT. However, we still need to know what proportion of adverse events occurred in HAPO in those who were in categories 1–4 by FPG but in higher categories by post-load PG.

Question 3 could also be answered by further analysis of the HAPO data.

So this is not so much a case of saying that ‘more research is necessary’ but of recommending further analysis of data already collected, which could be done over a much shorter timescale.

Methods

Treatment of gestational diabetes: cost studies

Chen et al. (2009)115 in a relatively complicated comparative analysis of National Hospital Discharge Survey statistics in the USA, estimated the additional cost arising from pregnancy management among women with gestational diabetes as compared with women without gestational diabetes in 2007 US$. (Note that the 5 April 2007 exchange rate was US$1.97/£1, which could be seen as historically unrepresentative and unduly high. Similar concerns may apply elsewhere in this review to 5 April 2008 when the exchange rate was US$1.99/£1.) In the USA in 2007, the authors estimated that the prevalence of gestational diabetes ranged from 1.3% in those under 20 years to 8.7% in those over 36 years, with a national average prevalence of 4.5% amounting to an annual 180,000 pregnancies with gestational diabetes.

Mothers with gestational diabetes used more inpatient days than those without gestational diabetes, with ratios of days (estimated through poisson regression) for caesareans (1.195); pre-eclampsia (1.499); other hypertension complication (1.560); and, ‘other pregnancy-related event’ (1.286), all statistically significant. They used about half the days for amniotic cavity infections. Ratios of rates for emergency visits where pregnancy was within the secondary diagnosis code (1.426) were also significant. With regards to ‘other ambulatory visits’ these were typically also significant, with ratios of rates for caesareans (presumably for post-operative follow-up) (1.221); polyhydramnios (1.855); urinary tract infection (1.119); pre-eclampsia (1.454); other hypertension complication (1.495); ‘other pregnancy-related event’ (1.386); and visits where pregnancy was within the secondary diagnosis code (1.123).

Ratios of rates of inpatient days among newborns to mothers who had gestational diabetes compared with those of non-gestational diabetes were significant for hospital inpatient days for: endocrine and metabolic disorders (2.907); respiratory distress syndrome (0.701); jaundice (1.754); congenital abnormalities (0.676) and ‘other neonatal events’ (1.035). Emergency visits rates were not statistically significantly different. Ratios of rates for ‘other ambulatory visits’ were significant for: macrosomia (1.826); endocrine and metabolic disorders (3.443); birth trauma (0.620); other complications of labour (1.315); respiratory distress syndrome (0.820); jaundice (1.213); congenital abnormalities (1.127); and ‘other neonatal events’ (1.027).

Given these rates and the prevalences of gestational diabetes across the age spectrum, the authors estimated an average additional cost per mother with gestational diabetes of US$3305 [£1733] and an average additional cost per newborn to a mother with gestational diabetes of US$209 [£110]. So an increasing prevalence of gestational diabetes will place a considerable cost burden on health services.

Screening for gestational diabetes: cost-effectiveness studies full papers

Nicholson et al. (2005)123 performed a cost–utility analysis of four alternative screening strategies:

• the 1-hour 50-g GCT

• the 100-g OGTT

• the 75-g OGTT

• no screening.

The GCT was taken to have a sensitivity of 80% and a specificity of 86%, with a positive result being a value equal to or greater than 7.8 mmol/l. The 75-g OGTT was assumed to have the same sensitivity and specificity as the GCT of 80% and 86% respectively, with a positive result being a 2-hour value equal to or greater than 7.8 mmol/l. A positive result from the GCT required a confirmatory 100-g OGTT performed in the fasting state. The 100-g OGTT resulted in a positive result given: fasting 5.3 mmol/l, 1 hour 10.0 mmol/, 2 hours 8.6 mmol/l; and 3 hours 8.1 mmol/l, and was taken to have a sensitivity and specificity of 100%.

Maternal outcomes included hypertensive disease, polyhydramnios and caesarean or vaginal delivery and their associated complications. Maternal complications modelled covered operative injury, endometritis, deep vein thrombosis, severe haemorrhage and hysterectomy. Neonatal outcomes included mild hypoglycaemia, macrosomia, respiratory distress syndrome, shoulder dystocia, no/mild morbidity, moderate morbidity and severe morbidity/death. Long term health states were defined within the modelling, these having the following quality of life values assumed (Table 21).

Cost estimation was within the US setting and adopted a social perspective, applying unit costs from Medicare and the Maryland Health Care Commission database. No real details of the model structure, probabilities of the various complications or the effectiveness of treatment for gestational diabetes upon the probabilities of the various complications were presented. The modellers adopted two models, one assessing the cost-effectiveness from the maternal perspective, the other from the neonatal perspective.

Based on an assumed prevalence of gestational diabetes of 4%, the least costly strategy was the 1-hour 50-g GCT followed by a confirmatory 100-g OGTT. This applied within both the maternal model with an average cost of US$2836 [£2181] and the neonatal model with an average cost of US$77 [£59]. Given the differences in average costs between the maternal model and the neonatal model, this suggests that the costs of maternal complications were included in only the maternal model and the costs of the neonatal complications were included in only the neonatal model, though this cannot be stated with certainty.

The next cost strategy was the 100-g OGTT: within the maternal model an average cost of US$2874 [£2210] and within the neonatal model an average cost of US$98 [£75] resulted in a cost increase of US$36 [£27.69] and US$11 [£8.46], respectively. This was also more effective and yielded an estimated QALY gain of 0.001 QALYs on average in both models, to yield cost a cost-effectiveness estimate of US$32,374 [£24,901] in the maternal model and US$8252 [£6347] in the neonatal model. The 75-g OGTT was more costly than the 100-g OGTT, with cost increases of US$59 [£45.38] and US$13 [£10.00] for the maternal model and neonatal model respectively. But it was less effective than the 100-g OGTT so was dominated. A similar pattern applied to the no screening strategy, being US$159 [£122] and US$3 [£2.31] more expensive than the 100-g OGTT but also less effective. Details of methods and results are sparse, and no CIs are given for the very small QALY differences.

The ordering of effectiveness of the strategies naturally followed from the assumptions as to sensitivities and specificities. Given the assumed equal sensitivity and specificity for the 50-g GCT and the 75-g OGTT, and perfect sensitivity and specificity for the 100-g OGTT, the GCT followed by the 100-g OGTT would have the same number of false-negatives as the 75-g OGTT but no false-positives. The 100-g OGTT would have no false-negatives or false-positives.

Given this and the lack of clarity as to the model structure and input there is little confidence that can be placed in these results.

Larijani et al. (2003)124 provided an estimate of the cost of alternative screening strategies among a sample of 2416 patients in four Iranian teaching hospitals, also calculating the cost per case detected within this sample. These patients were stratified into low, intermediate and high risk categories using the ADA criteria. The screening strategies compared were:

• universal screening

• selective screening, with low risk patients not being screened.

High risk patients were tested at their first prenatal visit and if not found to have gestational diabetes were tested again between gestation weeks 24 and 28. Intermediate and low risk patients were tested once between gestation weeks 24 and 28. The data were then used retrospectively to assess the effect of selective screening, with low risk women not tested. The 1-hour 50-g GCT was used for the screen with alternative thresholds of a 7.8-mmol/l cut-off and a 7.2-mmol/l cut-off, and was costed at US$1.38 [£1.06]. Confirmation of positive results from the screen used the 100-g OGTT and a minimum of two of the standard cut-off criteria, and was costed at US$4.19 [£3.22].

The overall prevalence of gestational diabetes in the sample was estimated as 4.7% with an assumption that universal screening with a 7.2-mmol/l cut-off was 100% sensitive, and the 100-g OGTT was 100% sensitive and specific. Universal screening with a 7.8-mmol/l cut-off was estimated to be 88% sensitive. Selective screening with 7.2-mmol/l cut-off was estimated to be 86% sensitive, while selective screening with a 7.8-mmol/l cut-off was estimated to be 77% sensitive. These led to: an average cost per patient across the entire sample, i.e. not restricted to those tested within the selective strategy; an average cost per case of gestational diabetes detected; and cost per additional case detected as shown in Table 22.

On the assumption that the cost per additional case conforms to the usual criteria for cost-effectiveness, the universal strategy with a cut-off of 7.8 mmol/l is extendedly dominated by the universal strategy with a cut-off of 7.2mmol/l. However, this dominance cannot be stated with confidence for any longer term measures of effectiveness and cost, given the restricted nature of the outcome measure and the costs included being restricted to the cost of testing.

Di Cianni et al. (2002)125 undertook a retrospective study of two Italian patient samples:

• universal screening with those diagnosed with gestational diabetes being intensively treated (n = 1338)

• selective screening of high risk patients, with those diagnosed with gestational diabetes being conventionally treated (n = 4035).

As such, the paper reports a joint test of the impact of alternative screening and treatment strategies, rather than alternative screening strategies per se. Intensive treatment aimed to achieve tighter metabolic control of 5.0 mmol/l FPG and 6.7 mmol/l PPG as compared with 5.5 mmol/l FPG and 7.2 mmol/l PPG for conventional treatment.

It should also be noted that the samples were drawn from the same hospital, being primarily differentiated by time. Those universally screened related to the period 1994 to 1997 while those selectively screened related to the period 1987 to 1992. Furthermore, the screening tests applied also differed between the two groups: those universally screened received the 50-g GCT with a cut-off of 7.5mmol/l, with positive results being confirmed with the 3-hour 100-g OGTT; those selectively screened received only the 3-hour 100-g OGTT.

For the universal screening, 367 patients (27.4%) screened positive with 84 patients (22.9% of the screen-positives) being confirmed as having gestational diabetes to yield an overall diagnosis rate of 6.3%. A random sample of 250 of the 971 screening negative were tested with the 3-hour 100-g OGTT, this yielding an additional five patients diagnosed with gestational diabetes, implying a sensitivity of 81%. {Note that the authors report an implied sensitivity of 94% on the basis of 84/(84+5) rather than 81% as would be calculated by 84/[84+5 × (971/250)]. The authors also report a specificity of 78% but it is unclear how this has been calculated.} The additional five patients diagnosed were included in the analysis to yield a total of 89 patients for analysis.

For selective screening the number of high risk patients was 600 out of the total patient sample of 4035. Ninety-three (15.5%) of these high risk patients tested positive to yield an overall prevalence of gestational diabetes of 2.3%. As a consequence, the selective testing of high risk patients appeared to miss significant numbers of patients with gestational diabetes in the non-high risk patient group.

Given the above, the characteristics of those diagnosed with gestational diabetes will have differed markedly between the two screening strategies. As a consequence, when reviewing the results of the paper, it should be borne in mind that the treatment strategies for those diagnosed with gestational diabetes also differed between the two strategies.

As would be expected, among those diagnosed with gestational diabetes through universal screening, followed by more intensive management, FPG showed a slightly better average level of 4.8 mmol/l as compared with 5.0 mmol/l among those diagnosed with gestational diabetes under selective screening, with this difference being reported as statistically significant, though clinically slight. Similarly, the percentages requiring insulin treatment, and the average insulin dose per day, were higher: 33% and 16.4 IU/day as compared with 7% and 13.6 IU/day, these differences again being reported as statistically significant. Neonatal and maternal morbidity was also typically better in those diagnosed though universal screening coupled with intensive management (Table 23).

The paper was not entirely clear in its reporting of costs, but it appears that the cost per diagnosed case under universal screening was €424 [£349] as compared with €406 [£334] for selective screening. However, there is a general lack of clarity around the costs within the paper, and these results should not be taken at face value. Given the different treatment regimes of those diagnosed with gestational diabetes and there being no consideration within the paper of the outcomes of treatment among those not diagnosed with gestational diabetes, a proportion of whom will be false-negatives, there appears to be little that can be drawn from the results of Di Canni et al. (2002). 125

Poncet et al. (2002)126 undertook a cost–consequence analysis of the four screening strategies:

• the 1-hour 50-g GCT among high risk women

• the 1-hour 50-g GCT among all women

• the 75-g OGTT among all women

• no screening

with a positive result from the GCT of a value equal to or greater than 7.2 mmol/l but requiring a confirmatory 100-g OGTT, this in turn being positive if two or more of the following applied: fasting 5.3 mmol/l, 1-hour 10.0 mmol/, 2-hour 8.6 mmol/l; and 3-hour 7.8 mmol/l. These values appear to relate to the ADA cut-offs. The 75-g OGTT was positive if either the fasting level was equal to or greater than 5.5 mmol/l or the 2-hour level was equal to or greater than 8.0 mmol/l.

The high risk group was defined as having: a close family history of diabetes; age 35 years or over; a BMI of more than 27 kg/m²; previous pregnancy complicated by diabetes, pre-eclampsia or a fetal death after 3 months gestation; or previous delivery of a child of more than 4 kg.

The outcomes measured for the cost–consequence analysis were macrosomia, prematurity, perinatal mortality and hypertension disorders. Estimates of these were drawn from a pooled analysis of 38 papers from within the literature with the following probabilities being assumed (Table 24).

The figure for perinatal mortality for undiagnosed GDM seems high. The text of the article says this was based on a collection of 38 articles, published from 1973 to 1998, and that the perinatal mortality ranged from 0.6% to 16.4%. However the table gives a range of 6.4–26.4%, of which the baseline figure used, 16.4%, is mid-point. But then in Table 4 of the Poncet et al. (2002)126 article, perinatal mortality untreated is only 1.09%.

The presence of risk factors given the presence of gestational diabetes was estimated as 63.3% as compared with 51.6% for those without gestational diabetes. The success of dietary advice to cope with gestational diabetes was estimated as 73.6%. Agreement to the 50-g GCT was anticipated to be 66.3% as compared with 85.9% for the 75-g OGTT.

Resource use was drawn from a prospective study of 120 patients of the Rhône-Alpes region. Unit costs from the French social health insurance system were applied. These costs included the costs of sick leave starting from the 24th week of gestation to discharge from maternity. Average costs per patient ranged from €5014 for an unscreened patient who did not have diabetes to €6026 for a patient screened as being diabetic with the 50-g GCT, having this confirmed with the 100-g OGTT and receiving insulin thereafter.

Within a decision tree analysis framework, the estimated costs and outcomes were as shown in Table 25 and in turn resulted in the cost-effectiveness ratios for the screening strategies compared with no screening (shown in Table 26).

The strategies of testing all women rather than just high risk women naturally identified more women with diabetes, and, given the assumptions about acceptance rates of the tests, the 75-g OGTT performed better than the 50-g GCT. But costs naturally also increased with the higher rates of acceptance, and, as the paper did not estimate the downstream costs of the complications, the most accepted and so most effective test was also the most expensive.

Given the cost–consequence analysis adopted, and despite the authors claiming that the strategy of screening high risk patients with the GCT has the best cost-effectiveness ratio, it cannot be concluded which strategy is the most cost-effective. There is also no obvious reason for the authors having excluded the strategy of screening high risk patients with the 75-g OGTT. The ranking of the strategies in terms of effectiveness was also critically dependent upon the assumed acceptance rates of 66.3% for the GCT as compared with 85.9% for the 75-g OGTT. No real detail was provided as to the source of these estimates, but it seems likely that the model would always predict a higher effectiveness but also a higher cost for the test with the higher acceptance rate.

Screening for gestational diabetes: cost-effectiveness studies abstracts

Lee et al. (2008)37 briefly described a model comparing alternative cut-off values for a positive diagnosis of gestational diabetes when using the 75-g OGTT test, drawing on data from the HAPO trial19 for cut-offs of 6.1 mmol/l, 7.0 mmol/l and 7.8 mmol/l of 2-h PG. The potential prevention of pre-eclampsia and shoulder dystocia from a diagnosis of gestational diabetes was drawn from the ACHOIS trial. 18 The model was long term, and possibly lifetime, estimating maternal and neonatal QALYs and costs from the societal perspective. Results indicated that lowering the threshold from the WHO recommended 7.8 mmol/l to 7.0 mmol/l had a cost-effectiveness of US$76,000 [£38,108] which was deemed to be not cost-effective. Results were reported as being sensitive to prevalence rates, cost of management and the effect of treatment for gestational diabetes on pre-eclampsia.

Thung et al. (2007)127 also undertook a modelling exercise comparing alternative cut-off values but for the 50-g GCT with the values of 7.2 mmol/l and 7.8 mmol/l being compared in a population with a prevalence of gestational diabetes of 3.3%, a third alternative of no screening also being considered. This found an average incremental cost of US$8.09 [£4.24] using a cut-off of 7.2 mmol/l as compared with 7.8 mmol/l but also an average patient gain of 0.00054 QALYs to yield a cost-effectiveness estimate of US$14,961 [£7845] per QALY. Perhaps surprisingly, the cost-effectiveness of the 7.8-mmol/l cut-off compared with no screening had a similar cost-effectiveness estimate of US$12,269 [£6433]. The cost-effectiveness of the 7.2-mmol/l cut-off compared with the 7.8-mmol/l cut-off rose above US$50,000 [£26,218] per QALY if the prevalence of gestational diabetes fell below 0.8%. However they assumed a high baseline perinatal mortality of 20%, and in sensitivy analysis, cost-effectiveness was lost if this fell below 6%.

Summary of published studies

In brief:

• Costs are higher for pregnancies with gestational diabetes – about £1833 according to Chen et al. (2009). 115

• Costs are lower for treatment with glibenclamide than with insulin, taking into account both drug costs and those of hypoglycaemia (Goetzl and Wilkins 2002117), by about $166 [£142].

• Costs are higher in those who need to go on to insulin (Kitzmiller et al. 1998119).

• The economic analysis of the ACHOIS study (Moss et al. 2007120) found that intervention with more intensive dietary advice, blood monitoring and insulin when required reported a cost per serious perinatal event avoided of £12,688. The (statistically not significant) impact upon perinatal mortality suggested a cost per life-year of £1376.

• Some studies (Lavin et al. 2001,121 Nicholson et al. 2005123) found that screening with the GTC and then testing screen-positives with the OGTT was less costly than going straight to universal OGTT.

The NCC report for the NICE Guideline Development Groups

A high quality cost-effectiveness analysis was provided for the GDGs by the National Collaborating Centre for Women’s and Children’s Health. This analysis is included as Appendix D of the Pregnancy in Diabetes guideline. 2 Full details are available on the NICE website. In brief, it:

• developed a single model covering both screening and treatment

• had a wide range of 21 screening options, from none at all, to various combinations of risk factor selection (ADA criteria, BMI alone, ethnicity) and glucose testing, some using a screening test such as the GCT, others going direct to a 75-g OGTT

• incorporated false-positives and false-negatives

• included all costs of screening and treatment

• gave good detail of assumptions used

• used the outcomes data from ACHOIS for benefits of treatment.

The analysis found that two screening strategies dominated:

• selection by ADA criteria followed by the 75-g OGTT (ICER £3678)

• selection by high risk ethnicity followed by the 75-g OGTT (ICER £21,739).

A few issues arise with the modelling:

• There was not at the time, evidence on the use of metformin.

• The ACHOIS group had mild gestational diabetes, and their outcome rate untreated would be less than if the whole spectrum of GDM was used (as noted in the analysis).

• The perinatal mortality rate was increased in the control arm of ACHOIS, but there were only five deaths and the difference was not statistically significant. Perinatal mortality with an average loss of 25 QALYs (normal life expectancy discounted to present values) was the main factor in the ICERs. Sensitivity analyses with decreasing numbers of death rapidly raised the ICERs above what would normally be considered affordable – one fewer death raised the ICER for the first strategy above from £678 to £27,634.

• There was no screening strategy which used risk factors only – e.g. treating all women above 30 with lifestyle measures (perhaps justifiably since only a minority would have HGP).

• The 75-g OGTT was assumed to have 100% sensitivity and specificity.

• The analysis relied on single sources for accuracy of screening and diagnostic tests.

• The cost of severe hypoglycaemia (£500) may have been too high since it appeared to assume hospital attendance.

• The risk of hypoglycaemia with metformin was assumed to be the same as with glibenclamide.

• The analysis preceded HAPO, and was therefore based on the WHO criteria for GDM, and could not examine the bands of glucose level and outcomes.

It would therefore be useful to have an updated cost-effectiveness analysis for each of the HAPO categories, with screening strategies which incorporated selection by risk factors (not published for the seven HAPO categories yet) and by single tests such as FPG. The analysis would model the effect of treatment (HAPO women being untreated). The updated analysis could add the supra-HAPO glycaemic groups (e.g. FPG over 5.8 mmol/l).

Such an analysis would provide a better basis for recommendations. Our best guide at present is probably the abstract by Lee et al. (2008). 37

Conclusions

The evidence from cost-effectiveness studies shows that it is worth screening for and treating hyperglycaemia in pregnancy at and above the ACHOIS levels. The NICE modelling shows that two-stage screening, first with selection by risk factors and then with the 75-g OGTT is also cost-effective.

However, the economics studies don’t yet help with our most difficult issue – at what level of BG does intervention become cost-effective. The study by Lee et al. (2008)37 addresses that issue, but is only available as an abstract. We recommend that the team which did the modelling for the NICE guidelines group should be asked to update their analysis.

Inevitably, modelling requires assumptions, about probabilities and utilities, and any estimate will have confidence intervals. So even the best modelling may not be able to come up with a neat solution to the continuum problem.

Changes since the previous Health Technology Assessment review in 2002

The evidence base has improved since the last HTA review in 2002,3 and since the main review for the NICE guidelines was done in 2007. 2 One of the main changes is the evidence on the continuum of risk, principally from the HAPO study19 but also supported by a number of smaller studies which examined outcomes for women on the borderlines of GDM as previously defined.

A second change is the evidence on the benefits of treatment at lower levels, from the ACHOIS study,18 and from the trial by Landon et al. (2008/9). 20,50

A third change is on mode of treatment, with good evidence that women who fail to control HGP on lifestyle measures alone can be safely and effectively treated with oral agents such as metformin or glibenclamide rather than going directly to insulin, though some women will still need insulin. There is also more evidence on the benefits of lifestyle change – calorie control, low glycaemic index foods, and physical activity.

The demonstration of the continuum by the HAPO study19 was predicted in the last HTA review,3 but we now have really strong evidence for it. That unfortunately causes problems in deciding on the level of BG at which we should intervene. There is no easy clinical threshold for intervention. It could be argued that since the first line of treatment would be diet and physical activity, it would not matter if we adopt a low threshold and perhaps give lifestyle advice to some women unnecessarily, but there is some evidence that the diagnosis itself can generate anxiety.

One way forward is through cost-effectiveness analysis, to determine the level, probably in terms of HAPO category, at which intervention is cost-effective. This is recommended below, under research needs.

Other perhaps interim options include intervention based on relative risk, i.e. intervening at the glucose level (or HAPO category) at which the risk is increased by a certain amount – say by 50%. This is rather arbitrary and might mean treating most women. For birthweight over the 90th percentile, the OR was 1.72 in HAPO category 3. It would be better to treat based on absolute risk.

Revisiting the National Screening Committee criteria

Some of the criteria which were not met in the last HTA review,3 are now met.

The condition should be an important health problem. Met. The condition has become more important because of rising prevalence and the HAPO demonstration of adverse outcomes over a much wider range of BG.

All cost-effective primary prevention interventions should have been implemented as far as possible. Debatable. Public health campaigns have not prevented the rise in general population obesity, but primary prevention has not been tried specifically in women planning pregnancy who may be more motivated for the sake of their babies.

The distribution of test values in the target population should be known and a suitable cut-off level defined and agreed. Not yet met, pending further cost-effectiveness analysis post-HAPO.

There should be an agreed policy on the further diagnostic investigation of individuals with a positive test result and on the choices available to those individuals. Partially met, but further analysis required. HAPO has shown that a single measure of BG is highly predictive, and it is possible that FPG alone could be used.

There should be an effective treatment or intervention for patients identified through early detection, with evidence of early treatment leading to better outcomes than late treatment. Met. There are effective treatments, including diet, physical activity, glibenclamide and metformin, and insulins. It is also reasonable to assume that missing the therapeutic window in early third trimester may lead to poorer outcomes. Some fetuses are macrosomic by the time of screening.

There should be agreed evidence based policies covering which individuals should be offered treatment and the appropriate treatment offered. Met. The ACHOIS trial18 has shown that intervention at lower levels is cost-effective. Trials of oral drugs have shown they are safe and effective, as well as being cheaper and preferred by patients. But there is still some doubt over the threshold for intervention.

There must be evidence from high-quality RCTs that the screening programme is effective in reducing mortality or morbidity. Not met – still no RCTs of screening versus no screening.

The benefit from the screening programme should outweigh the physical and psychological harm caused by the test, diagnostic procedures and treatment. Partly met. The balance has swung towards easier testing and easier treatment, which coupled with increasing prevalence should shift the balance towards benefit.

The opportunity cost of the programme should be economically balanced in relation to expenditure on medical care as a whole (i.e. value for money). Met for some groups following the economic analyses by the ACHOIS group and for the NICE GDGs, but still some uncertainties to be resolved.

Prevention of gestational diabetes mellitus

Could we tackle the problem at source, by reducing overweight and obesity in pregnancy? There might be two approaches. The first would be by health education campaigns aimed at persuading women to control weight before conception, and not to put on too much weight during pregnancy. The second could be opportunistic advice given to women attending for contraceptive services.

A recent systematic review found that for every one point increase in BMI, the prevalence of GDM increased by 0.9%. 128 The ORs for overweight and obese women were 1.97 and 3.01, respectively.

Recent reviews and guidelines

The German IQWiG published its review of screening for gestational diabetes in August 2009. 46 The summary in English noted that there was clear evidence of benefit of treating GDM. It noted that there was a lack of suitable screening studies, but it concluded that ‘an indication can be indirectly deduced that screening for gestational diabetes leads to a reduction in perinatal complications.’

The SIGN guideline1 notes the lack of international consensus over screening for GDM. It concludes that given present evidence, the best approach at present was two-stage screening, firstly by risk factors, and secondly by 75-g OGTT at 24–28 weeks in women selected by risk factors.The SIGN guideline considered the threshold for intervention, and noted the lack of an international consensus. It recommended that the threshold for intervention should be at a level where the RCTs showed an impact not just on birthweight, but on outcomes including shoulder dystocia and caesarean section.

The International Association of Diabetes and Pregnancy Study Groups (IADPSG) consensus panel advocated testing of all women with a 75-g OGTT at 24–28 weeks, unless they had been found to have abnormal glucose levels earlier in pregnancy. 129 The IADPSG did not take cost-effectiveness into account, but it noted that further analysis of the HAPO data might allow identification of low risk pregnancies where screening might not be required.

The International Diabetes Federation issued a ‘global guideline’ in 2009 which favoured universal testing of all women but issued a compromise recommendation of testing for GDM for all women at 26–28 weeks, ‘unless a selective process based on risk factors is deemed more appropriate’. 130

Research needs

It appears from the HAPO study19 that screening could use FPG. However we need further analysis before we can adopt that policy.

Research need 1: analysis of HAPO data to determine how many women in categories 1–4 by FPG are in categories 5 to 7 by post-load PG.

HAPO data could also be used to address the question of selective or universal screening, by comparing risk factors, and different thresholds, with each category. The hypothesis might be that women with risk factors are more likely to be in the higher categories.

There is one issue which needs to be considered if we relied on FPG for screening, which is that of later maternal T2DM. Retnakaran et al. (2009)131 found that FPG was better for predicting LGA infants, but that the post-load PGs were better for predicting post-gestational T2DM. The risk of T2DM has been thoroughly reviewed by Bellamy et al. (2009),132 with a meta-analysis showing that women who had had GDM (defined in various ways) had a relative risk of 7.4 compared with those who were normoglycaemic during pregnancy.

Research need 2: can risk factors identify a group of women whose risk of adverse outcomes is very low and who need not be screened?

The third research need concerns screening at booking clinic and how that should be done. At this stage we would be looking for pre-gestational diabetes or non-diabetic hyperglycaemia, so HbA_1c might be useful. It has been recommended by the National GDM Technical Working Party in New Zealand. 133

Research need 3: is HbA_1c a useful test at booking clinic for detecting pre-gestational diabetes, and also pre-gestational insulin resistance likely to be followed by HGP?

The above needs concern how best to screen, but an unresolved issue is the level of glucose at which intervention is worthwhile, which requires a cost-effectiveness analysis. This has been addressed by Lee et al. (2008)37 (whose results are available only as an abstract). It needs to be repeated in a UK context, by re-running the National Collaborating Centre for Women’s and Children’s Health modelling with updated assumptions and for the seven HAPO categories.

Research need 4: at which HAPO category does treatment become cost-effective, taking into account infant and maternal outcomes, and treatment with the cheaper oral agents when lifestyle measures fail, with insulin being used only when the oral drugs fail?

As mentioned above, it would be better to reduce the problem at source, by persuading women to achieve normal weight before conception. Trials of targeted health education are necessary.

Research need 5: could a health education campaign raise awareness of the problems of HGP amongst women of child-bearing age, and reduce the number becoming pregnant while overweight? Or at least reduce the BMI, and hence the risk?

The time continuum: research need 6: given the increasing age and weight of mothers-to-be, should screening start earlier? Screening is usually done at 24–28 weeks. Agarwal et al. (2007)134 described GDM as a form of T2DM which comes on over months not years. Are there studies which report the prevalence of HGP by gestational age, perhaps at 2-week intervals? Could such studies identify optimum time to screen, perhaps depending on age and BMI? Several commentators have noted that there can be delays between screening, diagnostic testing and treatment, and that these can occur during the ‘therapeutic window’ and hence result in poorer outcomes.

Research need 7: the HAPO study19 recorded head circumference. Given that the reported macrosomic babies will consist of a mixture of large healthy babies and truly macrosomic ones, the HAPO investigators could isolate the abnormal macrosomic ones by comparing weight with head circumference. This might allow a more refined analysis of macrosomia by HAPO category.

Research need 8: Reece et al. (2009)135 report that in the USA, the prevalence of pre-gestational diabetes in pregnancy has increased, but that of GDM has not. Conversely, Massicotte et al. (2009)136 reported that the prevalence of GDM in Canada had tripled over the last 10 years. It would be of interest to monitor trends in the UK.

One issue, raised by one of the peer reviewers for this report, was whether GDM should be defined on the basis of glucose at all. It may seem odd to define any diabetic condition on the basis of anything other than glucose, but the referee’s point was presumably that we are looking at a metabolic condition with abnormalities in physiological variables other than glucose, and that some of these other variables, such as lipids, might have a stronger relationship with adverse outcomes than glucose. Hadden and McLaughlin in a 2009 review note that normal pregnancy is hyperlipidaemic, and that birthweight is positively correlated with both plasma triglycerides and free fatty acid concentrations. 137 Is what we call GDM, part of a set of ‘metabolic syndromes of pregnancy’, with some women displaying changes more in glucose homeostasis, but others more in lipids, and some in both?

Ongoing research

• Two Cochrane protocols have been published and reviews are presumably under way:

  • Alternative strategies for diagnosing gestational diabetes mellitus to improve maternal and infant health. 138

  • Screening for gestational diabetes to improve maternal and child health. 139

The first is considering different screening tests. The protocol of the second suggests that it is doing the same, but the protocol for the first says that the second one is about whether screening should be selective or universal.

• A Finnish cluster-randomised controlled trial (ISRCTN33885819), described as due to complete at the end of 2008, used a lifestyle intervention (intensive dietary advice, counselling on physical activity, frequent contacts, and a group session) to reduce the prevalence of GDM and of large babies in women at risk.

• A Stanford University study is comparing the 50-gm glucose tolerance test, followed by a blood test 1 hour later, with 7-day continuous glucose monitoring (NCT00850135). The aim is to determine which test correlates better with adverse pregnancy outcomes as well as which one more accurately identifies patients at risk for adverse pregnancy outcomes.

• A University of North Carolina trial (NCT00835861), described as a pilot, is randomising women, with either pre-gestational T2DM or GDM diagnosed before 20 weeks, to metformin or insulin. The hypothesis is that metformin will provide glycaemic control that is equivalent to insulin in these women.

• A Swedish study (NCT00625781) is randomising pregnant women with IGT (FPG < 7 mmol/l and 2-hour PG between 10.0 and 12.2 mmol/l) to either insulin or diet. The purpose is to evaluate whether treatment of women close to normoglycaemia reduces children’s birthweight and neonatal morbidity.

• A study in Texas (NCT00744965) is randomising women with ‘mild gestational diabetes’ (PG after 50-g GCT ≥ 140 mg/dl but FPG of ≤ 105 mg/dl) to glibenclamide or diet alone. The hypothesis being tested is that in women with mild GDM, use of glibenclamide in addition to diet and nutritional counselling lowers mean infant birthweight by 200 g as compared with diet and nutritional counselling alone.

• In São Paulo, a recently completed trial (NCT00815828) has tested the effect of a programme of resistance exercises in women with gestational diabetes. Primary outcome measures is the comparison of the frequency of women who use insulin in the group who participate in the exercise programme with the group that don’t do the exercises.

• A Swedish trial currently under way is comparing a low glycaemic index carbohydrate diet versus no dietary intervention in pregnant women whose first baby was macrosomic. The aim is to prevent recurrence of a large baby (ISRCTN54392969).

• A randomised controlled trial in Tennessee comparing a combined metformin and glibenclamide tablet versus insulin in pregnant women with gestational or T2DM was described as due for completion at the end of 2008 (NCT00371306). The hypothesis is that patients will have similar or improved BG control on the combined oral agents as compared with control on insulin.

• An NIDDK (National Institute of Diabetes and Digestive and Kidney Diseases) trial to investigate the effects of a motivationally tailored, individually targeted 12-week physical activity intervention on the risk of GDM in women at high risk of the condition currently recruiting women (NCT00728377). The hypothesis being tested is that an exercise intervention is an effective tool for preventing GDM among women with a history of GDM.

• A Danish study is recruiting pregnant women with BMI over 30, who will be randomised to lifestyle intervention or a control group. The lifestyle intervention will include both diet (individual dietary counselling) and exercise (weekly aerobic classes) (NCT00530439). The primary outcome measures will be caesarean section, GDM, hypertension/pre-eclampsia, LGA and admission to NICU.

Conclusions

About the Aberdeen HTA group

The Aberdeen Health Technology Assessment Group is part of the Institute of Applied Health Sciences (IAHS), which is part of the College of Medicine and Life Sciences of the University of Aberdeen. The Institute of Applied Health Sciences is made up of discrete but methodologically related research groups. The HTA Group is drawn mainly from the Health Services Research Unit, Public Health, and the Health Economics Research Unit. In the last RAE (Research Assessment Exercise), Aberdeen was first equal with the University of York in the health services research unit of assessment.

The HTA Group carries out independent health technology assessments [Technology Assessment Reports (TARs)] for the NIHR HTA programme, which commissions TARs for NICE and other bodies, such as the NSC. It also carries out evidence reviews to support the NICE Single Technology Appraisal programme.

Particular interests include evaluation of non-pharmacological technologies, screening and diabetes. Previous TARs from Aberdeen include:

• Inhaled insulin for diabetes (for NICE)140

• Continuous subcutaneous insulin infusion (NICE)141

• Screening for type 2 diabetes (NSC and Department of Health)142

• Newer drugs for type 2 diabetes (NICE)143

• Self-monitoring of blood glucose in type 2 diabetes (Department of Health). 144

We also do Cochrane reviews on diabetic topics.

Contribution of authors

Norman Waugh drafted Chapters 1, 3 and 5. Christine Clar, Pamela Royle and Rob Henderson drafted Chapter 2. Ewen Cummins drafted Chapter 4. Pamela Royle carried out literature searches. Donald Pearson, Robert Lindsay and David Hadden provided expert clinical advice and commented on drafts. Pamela Royle and Norman Waugh edited the final document.

Disclaimers

The views expressed in this publication are those of the authors and not necessarily those of the HTA programme or the Department of Health.