Insulin versus metformin

Nineteen studies compared insulin with metformin (Ashoush 2016; Beyuo 2015; Hague 2003; Hickman 2013; Ijas 2011; Majeed 2015; Martinez Piccole 2010; Mesdaghinia 2013; Moore 2007; Niromanesh 2012; Riaz 2014; Rowan 2008; Ruholamin 2014; Saleh 2016; Spaulonci 2013; Tertti 2013; Waheed 2013; Wali 2015; Zawiejska 2016) though four of these did not contribute data (Martinez Piccole 2010; Riaz 2014; Waheed 2013; Wali 2015).

Comparisons included the following.

• Regular insulin (no other details) Majeed 2015

• Regular human insulin plus Neutral Protamine Hagedorn insulin versus metformin (Ashoush 2016; Hickman 2013; Mesdaghinia 2013)

• Neutral Protamine Hagedorn insulin versus metformin (Niromanesh 2012; Spaulonci 2013)

• Insulin lispro plus Neutral Protamine Hagedorn insulin versus metformin (Ijas 2011)

• Insulin lispro/aspart plus Neutral Protamine Hagedorn insulin versus metformin (Tertti 2013)

• Soluble and premixed insulin (no other details) (Beyuo 2015)

• Actrapid and Mixtard versus metformin (Saleh 2016)

Nine studies did not report details on the insulin type used (Hague 2003; Martinez Piccole 2010; Moore 2007; Riaz 2014; Rowan 2008; Ruholamin 2014; Waheed 2013; Wali 2015; Zawiejska 2016).

Hickman 2013 included both women with GDM and pre‐existing diabetes. The data could not be separated however as 50% of the women had GDM these data have been included. Authors have been contacted for further information.

Seven studies are ongoing (NCT01756105; IRCT2014010116025N1; CTRI/2013/10/004055; SLCTR/2011/009; CTRI/2011/08/001956; CTRI/2014/08/004835; NCT00681460). In the next update of this review we will check for publications from these studies and add data if available.

Insulin versus glibenclamide

Eleven studies compared insulin with glibenclamide (Anjalakshi 2007;Behrashi 2016; Bertini 2005; Lain 2009; Langer 2000; Mirzamoradi 2015; Mukhopadhyay 2012; Ogunyemi 2007; Pavithra 2016; Silva 2007; Zangeneh 2014).

Comparisons included the following.

• Long‐acting and short‐acting insulin (no details) versus glibenclamide (Lain 2009)

• Regular human insulin versus glibenclamide (Langer 2000)

• Regular human insulin plus Neutral Protamine Hagedorn insulin versus glibenclamide (Behrashi 2016; Bertini 2005; Mirzamoradi 2015; Silva 2007; Zangeneh 2014).

Four studies did not report any details on the insulin type that was used (Anjalakshi 2007; Mukhopadhyay 2012; Ogunyemi 2007; Pavithra 2016).

There are two ongoing studies (IRCT2013102315045N2; NCT01731431). In the next update of this review we will check for publications from these studies and add data if available.

Insulin versus acarbose

Two studies compared insulin with acarbose (Bertini 2005; De Veciana 2002).

Comparisons included the following.

• Regular human insulin plus Neutral Protamine Hagedorn insulin versus acarbose (Bertini 2005)

• Insulin lispro plus Neutral Protamine Hagedorn insulin versus acarbose (De Veciana 2002)

Combined metformin and glibenclamide

Three studies compared insulin with a combination of metformin and glibenclamide (Ardilouze 2014; Hutchinson 2008 ; Mohamed 2014). The data for Hutchinson 2008 cannot be separated out for GDM and type 2 diabetes and have not been included. Authors have been contacted.

Comparisons included the following.

• Insulin aspart or lispro plus Neutral Protamine Hagedorn insulin versus combined metformin and glibenclamide (Ardilouze 2014)

• Intermediate‐ and short‐acting insulin (no details) versus combined metformin and glibenclamide (Mohamed 2014). This study recruited women with both GDM and type 2 diabetes and the data cannot be separated. However, 65 of the women recruited had a diagnosis of GDM and 19 with type 2 diabetes. As the majority of women have GDM (77%) we have included the data in the analyses.

One study is ongoing (NCT02080377). In the next update of this review we will check for publications from these studies and add data if available.

Insulin type A versus insulin type B

Ten studies compared one preparation of insulin with another preparation of insulin (Balaji 2005; Balaji 2012; Di Cianni 2007;Herrera 2015; Jovanovic 1999; Mecacci 2003; Ismail 2007; Pettitt 2007; Poyhonen‐Alho 2002; Prasad 2008).

One ongoing study was identified (NCT01662921) that compares insulin glulisine versus insulin lispro. We will seek data for this study in subsequent updates of this review.

Comparisons included the following.

• Regular human insulin versus insulin aspart (Balaji 2005; Balaji 2012; Di Cianni 2007)

• Regular human insulin versus insulin lispro (Di Cianni 2007; Jovanovic 1999; Mecacci 2003)

• Regular human insulin versus Neutral Protamine Hagedorn insulin (Ismail 2007; Poyhonen‐Alho 2002)

• Regular human insulin plus Neutral Protamine Hagedorn insulin versus insulin aspart (Prasad 2008)

• Neutral Protamine Hagedorn insulin plus regular human insulin versus Neutral Protamine Hagedorn insulin plus insulin aspart (Pettitt 2007)

• Insulin detemir plus and insulin aspart versus Neutral Protamine Hagedorn insulin and insulin aspart (Herrera 2015)

Insulin versus diet

Four studies compared insulin with diet (Coustan 1978; Notelovitz 1971; Persson 1985; Thompson 1990). One study did not provide details of the preparation of insulin that was used (Notelovitz 1971). Two studies compared insulin (plus diet) with standard antenatal care (O'Sullivan 1975a; O'Sullivan 1975b).

Coustan 1978 used regular human insulin plus Neutral Protamine Hagedorn insulin; Persson 1985 compared fast/intermediate‐acting insulin (no details) with diet versus diet alone and Thompson 1990 compared regular human insulin plus Neutral Protamine Hagedorn insulin and diet with diet alone.

Insulin versus exercise

One study compared insulin with exercise (Bung 1993), the type of insulin used was not specified.

Regimens of insulin

Two studies compared different regimens of insulin (Castorino 2011; Nachum 1999).

One ongoing study was identified (NCT01613807) which compares three injections of Humalog® 50/50TM daily with Humalog® plus Humalin N® administered as six injections daily. In subsequent updates of this review we will seek data from this study, where possible.

Comparisons included the following.

• 50/50 combination of insulin lispro plus Neutral Protamine Hagedorn insulin given as either three pre‐meal doses or as six distinct injections (Castorino 2011)

• A twice‐daily regimen of human regular insulin plus human intermediate insulin with a four times daily regimen of regular human insulin (Nachum 1999)

Maternal primary outcomes

• Hypertensive disorders of pregnancy (any definition) was reported in 13 studies (Balaji 2012; Hague 2003; Ijas 2011; Langer 2000; Mirzamoradi 2015; Nachum 1999; Niromanesh 2012; Pavithra 2016; Rowan 2008; Saleh 2016; Spaulonci 2013; Tertti 2013; Zangeneh 2014).

• Caesarean section was reported in 25 studies (Ardilouze 2014; Ashoush 2016; Balaji 2012; Bertini 2005; Bung 1993; Castorino 2011; Coustan 1978; Hague 2003; Ijas 2011; Jovanovic 1999; Langer 2000; Mecacci 2003; Mirzamoradi 2015; Moore 2007; Nachum 1999; Niromanesh 2012; Ogunyemi 2007; Pavithra 2016; Ruholamin 2014; Saleh 2016; Silva 2007; Spaulonci 2013; Tertti 2013; Thompson 1990; Zangeneh 2014).

• Development of type 2 diabetes was reported in four studies (Coustan 1978; O'Sullivan 1975a; Rowan 2008; Tertti 2013).

Neonatal primary outcomes

• Perinatal (fetal and neonatal death) mortality was reported in 15 studies (Bertini 2005; Ijas 2011; Lain 2009; Langer 2000; Mesdaghinia 2013; Mohamed 2014; Mukhopadhyay 2012; Nachum 1999; Niromanesh 2012; O'Sullivan 1975a; O'Sullivan 1975b; Persson 1985; Silva 2007; Tertti 2013; Thompson 1990). Later infant mortality was not reported in any studies.

• Large‐for‐gestational age (LGA) was reported in 19 studies (Balaji 2012; Bertini 2005; Castorino 2011; Hickman 2013; Ijas 2011; Jovanovic 1999; Lain 2009; Langer 2000; Mecacci 2003; Mesdaghinia 2013; Mukhopadhyay 2012; Nachum 1999; Niromanesh 2012; Persson 1985; Rowan 2008; Saleh 2016; Silva 2007; Spaulonci 2013; Tertti 2013).

• Death or serious morbidity composite was reported in three studies (Hickman 2013; Nachum 1999; Rowan 2008).

• Neurosensory disability in later childhood was reported in one trial (Ijas 2011).

Maternal secondary outcomes

• Use of additional pharmacotherapy was reported in 23 studies (Ardilouze 2014; Ashoush 2016; Bertini 2005; Beyuo 2015; Di Cianni 2007; De Veciana 2002; Hickman 2013; Ijas 2011; Lain 2009; Langer 2000; Mecacci 2003; Mirzamoradi 2015; Moore 2007; Niromanesh 2012; Ogunyemi 2007; Pavithra 2016; Persson 1985; Poyhonen‐Alho 2002; Rowan 2008; Ruholamin 2014; Silva 2007; Spaulonci 2013; Tertti 2013).

• Maternal hypoglycaemia was reported in 16 studies (Anjalakshi 2007; Ashoush 2016; Balaji 2012; Bertini 2005; Di Cianni 2007; Hickman 2013; Ismail 2007; Langer 2000; Moore 2007; Mukhopadhyay 2012; Nachum 1999; Ogunyemi 2007; Pettitt 2007; Silva 2007; Thompson 1990; Zangeneh 2014).

• Glycaemic control during/end of treatment (fasting) was reported in 25 studies (Ardilouze 2014; Ashoush 2016; Balaji 2005; Balaji 2012; Behrashi 2016; Bertini 2005; Beyuo 2015; Castorino 2011; De Veciana 2002; Herrera 2015; Lain 2009; Langer 2000; Mecacci 2003; Mirzamoradi 2015; Moore 2007; Mukhopadhyay 2012; Niromanesh 2012; Ogunyemi 2007; Pavithra 2016; Rowan 2008; Saleh 2016; Silva 2007; Spaulonci 2013;Thompson 1990; Zawiejska 2016).

• Glycaemic control during/end of treatment (postprandial) was reported in 23 studies (Anjalakshi 2007; Ardilouze 2014; Ashoush 2016; Balaji 2005; Balaji 2012; Bertini 2005; Beyuo 2015; Di Cianni 2007; De Veciana 2002; Herrera 2015; Lain 2009; Langer 2000; Mecacci 2003; Mirzamoradi 2015; Moore 2007; Mukhopadhyay 2012; Niromanesh 2012; Ogunyemi 2007; Pavithra 2016; Rowan 2008; Saleh 2016; Silva 2007; Spaulonci 2013).

• Glycaemic control during/end of treatment (HbA1c) was reported in 15 studies (Anjalakshi 2007; Balaji 2012; Castorino 2011; Jovanovic 1999; Langer 2000; Mecacci 2003; Mesdaghinia 2013; Mohamed 2014; Mukhopadhyay 2012; Nachum 1999; Niromanesh 2012; Persson 1985; Rowan 2008; Tertti 2013; Zawiejska 2016).

• Weight gain in pregnancy was reported in 14 studies (Ashoush 2016; Balaji 2012; Bertini 2005; Coustan 1978; Herrera 2015; Ijas 2011; Langer 2000; Majeed 2015; Nachum 1999; Niromanesh 2012; Rowan 2008; Silva 2007; Spaulonci 2013; Tertti 2013).

• Induction of labour was reported in three studies (Hague 2003; Ijas 2011; Tertti 2013).

• Postpartum haemorrhage was reported in two studies (Hickman 2013; Moore 2007).

• Breastfeeding at discharge, six weeks postpartum, six months or longer was reported in two studies (Ijas 2011; Rowan 2008).

• Maternal mortality was reported in one study (Ismail 2007).

• Body mass index (BMI) was reported in one study (Rowan 2008).

• Impaired glucose tolerance was reported in three studies (Mirzamoradi 2015; Rowan 2008; Tertti 2013).

• Relevant biomarker changes associated with the intervention was reported in one study (Zawiejska 2016).

Neonatal secondary outcomes

• Stillbirth was reported in three studies (Hickman 2013; Langer 2000; Tertti 2013).

• Neonatal death was reported in three studies (Lain 2009; Langer 2000; O'Sullivan 1975a).

• Macrosomia was reported in 29 studies (Ashoush 2016; Balaji 2012; Behrashi 2016; Bertini 2005; Coustan 1978; Di Cianni 2007; Hague 2003; Hickman 2013; Ijas 2011; Ismail 2007; Lain 2009; Langer 2000; Mesdaghinia 2013; Mirzamoradi 2015; Moore 2007; Mukhopadhyay 2012; Nachum 1999; Niromanesh 2012; O'Sullivan 1975a; Pavithra 2016; Pettitt 2007; Poyhonen‐Alho 2002; Prasad 2008; Ruholamin 2014; Silva 2007; Spaulonci 2013; Tertti 2013; Thompson 1990; Zangeneh 2014).

• Small‐for‐gestational age (SGA) was reported in 13 studies (Bertini 2005; Coustan 1978; Lain 2009; Mecacci 2003; Mesdaghinia 2013; Nachum 1999; Niromanesh 2012; Persson 1985; Rowan 2008; Ruholamin 2014; Saleh 2016; Spaulonci 2013; Tertti 2013).

• Birth trauma was reported in 16 studies (Bertini 2005; Coustan 1978; Hickman 2013; Ijas 2011; Lain 2009; Mesdaghinia 2013; Moore 2007; Nachum 1999; Niromanesh 2012; Poyhonen‐Alho 2002; Rowan 2008; Ruholamin 2014; Saleh 2016; Silva 2007; Tertti 2013; Thompson 1990).

• Gestational age at birth was reported in 24 studies (Ardilouze 2014; Balaji 2012; Behrashi 2016; Bertini 2005; Bung 1993; Coustan 1978; De Veciana 2002; Hague 2003; Ijas 2011; Jovanovic 1999; Lain 2009; Langer 2000; Mirzamoradi 2015; Mohamed 2014; Moore 2007; Mukhopadhyay 2012; Nachum 1999; Niromanesh 2012; Ogunyemi 2007; Rowan 2008; Saleh 2016; Spaulonci 2013; Tertti 2013; Thompson 1990).

• Preterm birth was reported in 15 studies (Ashoush 2016; Balaji 2012; Ijas 2011; Lain 2009; Majeed 2015; Mesdaghinia 2013; Niromanesh 2012; O'Sullivan 1975a; Pavithra 2016; Prasad 2008; Poyhonen‐Alho 2002; Rowan 2008; Saleh 2016; Spaulonci 2013; Tertti 2013).

• Congenital abnormality was reported in 17 studies (Ashoush 2016; Behrashi 2016; De Veciana 2002; Hickman 2013; Jovanovic 1999; Langer 2000; Mesdaghinia 2013; Mirzamoradi 2015; Mohamed 2014; Mukhopadhyay 2012; Niromanesh 2012; Ogunyemi 2007; Pettitt 2007; Rowan 2008; Saleh 2016; Tertti 2013; Zangeneh 2014).

• Five‐minute Apgar score less than seven was reported in five studies (Mesdaghinia 2013; Nachum 1999; Rowan 2008; Ruholamin 2014; Saleh 2016).

• Birthweight was reported in 33 studies (Anjalakshi 2007; Ardilouze 2014; Ashoush 2016; Balaji 2005; Balaji 2012; Behrashi 2016; Bertini 2005; Bung 1993; Castorino 2011; De Veciana 2002; Hague 2003; Ijas 2011; Ismail 2007; Jovanovic 1999; Lain 2009; Langer 2000; Mecacci 2003; Mesdaghinia 2013; Mirzamoradi 2015; Mohamed 2014; Moore 2007; Mukhopadhyay 2012; Nachum 1999; Niromanesh 2012; Ogunyemi 2007; Pettitt 2007; Poyhonen‐Alho 2002; Rowan 2008; Ruholamin 2014; Silva 2007; Spaulonci 2013; Tertti 2013; Thompson 1990).

• Head circumference at birth was reported in three studies (Lain 2009; Niromanesh 2012; Rowan 2008).

• Length at birth was reported in seven studies (Balaji 2012; Bung 1993; Jovanovic 1999; Lain 2009; Niromanesh 2012; Pettitt 2007; Rowan 2008).

• Ponderal index at birth was reported in five studies (Balaji 2012; Lain 2009; Mecacci 2003; Rowan 2008;Thompson 1990).

• Adiposity was reported using a number of measures including skin fold thickness and percentage fat mass in two studies (Lain 2009; Rowan 2008).

• Neonatal hypoglycaemia was reported in 31 studies (Ardilouze 2014; Ashoush 2016; Behrashi 2016; Bertini 2005; Coustan 1978; Hickman 2013; Ijas 2011; Jovanovic 1999; Lain 2009; Langer 2000; Majeed 2015; Mesdaghinia 2013; Mirzamoradi 2015; Mohamed 2014; Moore 2007; Mukhopadhyay 2012; Nachum 1999; Niromanesh 2012; Ogunyemi 2007; Pavithra 2016; Persson 1985; Poyhonen‐Alho 2002; Prasad 2008; Rowan 2008; Ruholamin 2014; Saleh 2016; Silva 2007; Spaulonci 2013; Tertti 2013; Thompson 1990; Zangeneh 2014).

• Respiratory distress syndrome was reported in 13 studies (Balaji 2012; Behrashi 2016; Bung 1993; Mesdaghinia 2013; Moore 2007; Mukhopadhyay 2012; Nachum 1999; Niromanesh 2012; Pavithra 2016; Rowan 2008; Ruholamin 2014; Saleh 2016; Spaulonci 2013).

• Neonatal jaundice (hyperbilirubinaemia) was reported in 21 studies (Behrashi 2016; Bung 1993; Hague 2003; Ijas 2011; Langer 2000; Mesdaghinia 2013; Mirzamoradi 2015; Mohamed 2014; Moore 2007; Mukhopadhyay 2012; Nachum 1999; Niromanesh 2012; Pavithra 2016; Poyhonen‐Alho 2002; Prasad 2008; Ruholamin 2014; Saleh 2016; Spaulonci 2013; Tertti 2013; Thompson 1990; Zangeneh 2014).

• Hypocalcaemia was reported in seven studies (Behrashi 2016; Jovanovic 1999; Langer 2000; Mirzamoradi 2015; Pavithra 2016; Thompson 1990; Zangeneh 2014).

• Polycythaemia was reported in five studies (Langer 2000; Mirzamoradi 2015; Nachum 1999; Persson 1985; Zangeneh 2014).

• Relevant biomarker changes associated with the intervention including cord blood c peptide and cord insulin were reported in four studies (Anjalakshi 2007; Lain 2009; Langer 2000; Persson 1985).

• Childhood ‐ weight was reported in two studies (Ijas 2011; Rowan 2008).

• Childhood ‐ height was reported in two studies (Ijas 2011; Rowan 2008).

• Childhood adiposity including ponderal index and total fat mass were reported in two studies (Ijas 2011; Rowan 2008).

• Childhood blood pressure was reported in one study (Rowan 2008).

• The number of antenatal visits or admissions was reported in one study (Langer 2000).

• Admission to neonatal intensive care unit/nursery was reported in 18 studies (Ashoush 2016; Behrashi 2016; Bertini 2005; Ijas 2011; Lain 2009; Langer 2000; Majeed 2015; Mesdaghinia 2013; Mirzamoradi 2015; Mohamed 2014; Moore 2007; Mukhopadhyay 2012; Niromanesh 2012; Rowan 2008; Ruholamin 2014; Saleh 2016; Tertti 2013; Zangeneh 2014).

• Duration of stay in neonatal intensive care unit was reported in three studies (Mirzamoradi 2015; Mohamed 2014; Tertti 2013).

1.1 and 1.2 Hypertensive disorders of pregnancy (including pre‐eclampsia, pregnancy‐induced hypertension, eclampsia)

Ten studies reported on hypertensive disorders of pregnancy; three compared insulin with glibenclamide (Langer 2000; Mirzamoradi 2015; Zangeneh 2014) and seven studies compared insulin with metformin (Hague 2003; Ijas 2011; Niromanesh 2012; Rowan 2008; Saleh 2016; Spaulonci 2013; Tertti 2013). Hypertensive disorders of pregnancy were reported as pre‐eclampsia or hypertensive disorders of pregnancy (not defined). No studies reported eclampsia as an outcome.

There was no clear evidence of a difference for the risk of pre‐eclampsia between women who had been treated with insulin and those who had been treated with an oral anti‐diabetic pharmacological therapy (risk ratio (RR) 1.14, 95% CI 0.86 to 1.52; 10 studies, 2060 women; moderate‐quality evidence) (Analysis 1.1). The quality of the evidence was downgraded for risk of bias as eight of the 10 studies were not blinded.

Treatment with insulin was associated with an increased risk for hypertensive disorders of pregnancy (not defined) compared with oral anti‐diabetic pharmacological therapy (RR 1.89, 95% CI 1.14 to 3.12; four studies, 1214 women; moderate‐quality evidence,Analysis 1.2). Three studies (Niromanesh 2012; Rowan 2008; Tertti 2013) had used metformin as the anti‐diabetic pharmacological therapy and one study used glibenclamide (Pavithra 2016).

1.3 Caesarean section

Caesarean section was reported in 17 studies; seven compared insulin with glibenclamide (Bertini 2005; Langer 2000; Mirzamoradi 2015; Ogunyemi 2007; Pavithra 2016; Silva 2007; Zangeneh 2014); nine compared insulin with metformin (Ashoush 2016; Hague 2003; Ijas 2011; Moore 2007; Niromanesh 2012; Ruholamin 2014; Saleh 2016; Spaulonci 2013; Tertti 2013), and one study each compared insulin with acarbose (Bertini 2005) and a combination on metformin and glibenclamide (Ardilouze 2014).

There was no clear evidence of a difference for the risk of birth by caesarean section between women who had been treated with insulin and those who had been treated with an oral anti‐diabetic pharmacological therapy (RR 1.03, 95% CI 0.93 to 1.14; 1988 women; 17 studies; moderate‐quality evidence) (Analysis 1.3). The quality of the evidence was downgraded for risk of bias as 15 of the 17 studies were unblinded. Visual inspection of the funnel plot associated with this outcome does not suggest any evidence of publication bias (Figure 2).

1.4 Development of type 2 diabetes

Development of type 2 diabetes was reported in two studies both comparing insulin and metformin (Rowan 2008; Tertti 2013). Rowan 2008 reported the outcome at the six to eight weeks postpartum oral glucose tolerance test and Tertti 2013 at up to one year postpartum. There was no evidence of a difference between groups (RR 1.39, 95% CI 0.80 to 2.44; two studies, 754 women; moderate‐quality evidence) (Analysis 1.4). The quality of the evidence was downgraded for risk of bias as both studies were open‐label.

1.5 Perinatal (fetal and neonatal death) and later infant mortality

Perinatal death was reported in 11 comparisons in 10 studies; five studies compared insulin with glibenclamide (Bertini 2005; Lain 2009; Langer 2000; Mukhopadhyay 2012; Silva 2007); four studies compared insulin with metformin (Ijas 2011; Mesdaghinia 2013; Niromanesh 2012; Tertti 2013), and one study each compared insulin with acarbose (Bertini 2005) and a combination of metformin and glibenclamide (Mohamed 2014).

There was no clear evidence of a difference for the risk of perinatal death between infants whose mothers had been treated with insulin and those treated with oral anti‐diabetic pharmacological therapies (RR 0.85; 95% CI 0.29 to 2.49; 10 studies, 1463 infants; low‐quality evidence). The quality of evidence was downgraded for risk of bias and imprecision. Six studies reported that there were no events of perinatal group in either the insulin or the oral anti‐diabetic drug groups (Bertini 2005; Ijas 2011; Mesdaghinia 2013; Niromanesh 2012; Silva 2007; Tertti 2013). The studies were not powered to detect differences in perinatal mortality and event rates are very low 3/686 for the insulin‐treated group and 3/693 for the oral anti‐diabetic agent group (Analysis 1.5).

No data were reported for later infant mortality.

1.6 Large‐for‐gestational age (LGA)

LGA was reported in 14 studies of three comparisons; five studies compared insulin with glibenclamide (Bertini 2005; Lain 2009; Langer 2000; Mukhopadhyay 2012; Silva 2007); eight studies compared insulin with metformin (Hickman 2013; Ijas 2011; Mesdaghinia 2013; Niromanesh 2012; Rowan 2008; Saleh 2016; Spaulonci 2013; Tertti 2013) and one study (Bertini 2005) compared insulin with acarbose.

There was no clear evidence of a difference between infants whose mothers had been treated with insulin and those treated with oral anti‐diabetic pharmacological therapies for the risk of being born LGA (average RR 1.01, 95% CI 0.76 to 1.35; 2352 infants; 13 studies; I² = 35%; moderate‐quality evidence;Analysis 1.6). There was substantial heterogeneity present in this outcome so these result should be interpreted with caution (Heterogeneity: Tau² = 0.08; Chi² = 19.88 (P = 0.10)). Quality of evidence was downgraded for risk of bias and publication bias. The majority of studies were unblinded and visual inspection of the funnel plot (Figure 3) suggests asymmetry and therefore publication bias with those studies with larger effect sizes more likely to be published.

1.7 Death or serious morbidity composite (variously defined by trials, e.g. perinatal or infant death, shoulder dystocia, bone fracture or nerve palsy)

Two studies comparing insulin with metformin reported a composite outcome of serious infant morbidity (Hickman 2013; Rowan 2008). There was no evidence of a clear difference between those whose mothers had been treated with insulin and those treated with oral anti‐diabetic pharmacological therapies (metformin) for the risk of a composite of serious infant morbidity (RR 1.03, 95% CI 0.84 to 1.26; two studies, 760 infants; moderate‐quality evidence;Analysis 1.7). The quality of evidence was downgraded for risk of bias as both studies were not blinded.

1.8 Neurosensory disability in later childhood

At 18 months of age, there was no evidence of a clear difference in the risk of any mild developmental delay between infants whose mothers had been treated with insulin and those who had been treated with oral anti‐diabetic agents (RR 1.07, 95% CI 0.33 to 3.44; one study, 93 children; low‐quality evidence) (Ijas 2011). At 18 months of age, there was no evidence of a difference in the risk of hearing impairment (RR 0.31; 95% CI 0.01 to 7.49; one study, 93 children; low‐quality evidence) or visual impairment (RR 0.31, 95% CI 0.03 to 2.90; one study, 93 children; low‐quality evidence) between infants whose mothers had been treated with insulin and those who had been treated with oral anti‐diabetic agents (Ijas 2011) (Analysis 1.8).

1.9 Use of additional pharmacotherapy

Where reported, for those women who were treated with oral anti‐diabetic pharmacological therapy a total of 22% required supplementary insulin (RR 0.03, 95% CI 0.02 to 0.06; 19 studies; 2761 women). Seven studies reported using glibenclamide as the oral anti‐diabetic (Bertini 2005; Lain 2009; Langer 2000; Mirzamoradi 2015; Ogunyemi 2007;Pavithra 2016; Silva 2007). In these studies 6.7% (29/433) of women required supplementary insulin. Ten studies reported using metformin as the oral anti‐diabetic (Ashoush 2016; Beyuo 2015; Hickman 2013; Ijas 2011; Moore 2007; Niromanesh 2012; Rowan 2008; Ruholamin 2014; Spaulonci 2013; Tertti 2013). For these studies, 30% (250/837) of women required supplementary insulin. These data are skewed by the Rowan 2008 study, which is the largest in this subgroup. In this study almost half of the women (46%, 168/363) received supplementary insulin. Acarbose was used in two studies (Bertini 2005; De Veciana 2002), for which 17% (11/64) women received supplementary insulin. It should be noted that women in the De Veciana 2002 study who were randomised to acarbose commenced insulin because they could not tolerate increasing doses of acarbose rather than an inability to maintain glycaemic control. One study reported comparing insulin with a combination on metformin and glibenclamide (Ardilouze 2014), 37% of the oral anti‐diabetic agent group required additional insulin (Analysis 1.9).

1.10 Maternal hypoglycaemia

There was no evidence of a difference for the risk of maternal hypoglycaemia between women who had been treated with insulin and those who had been treated with an oral anti‐diabetic pharmacological therapy (average RR 3.01, 95% CI 0.74 to 12.27; 10 studies, 998 women; I² = 84%) (Analysis 1.10). Five studies reported no events of maternal hypoglycaemia in either the insulin nor the oral anti‐diabetic pharmacological therapy group (Anjalakshi 2007; Bertini 2005; Moore 2007; Mukhopadhyay 2012; Silva 2007). Heterogeneity may be due to lack of blinding reported in the studies reporting this outcome or to definitions of hypoglycaemia which are poorly report in the individual studies (Heterogeneity: Tau² = 1.91; Chi² = 24.87 (P = < 0.0001)).

1.11 to 1.13 Glycaemic control during/end of treatment

There was no evidence of a clear difference in fasting blood glucose (average standardised mean difference (SMD) 0.05, 95% CI ‐0.09 to 0.19; 19 studies, 2812 women; I² = 66%; Tau² = 0.06; Chi² = 56.67 (P <0.0001)) (Analysis 1.11), postprandial blood glucose (average SMD 0.12, 95% CI ‐0.05 to 0.29; 18 studies 2508 women; I² = 73%; Tau² = 0.09; Chi² = 67.85 (P < 0.00001)) (Analysis 1.12) or HbA1c (average SMD 0.01, 95% CI ‐0.12 to 0.15; nine studies, 1963 women; I² = 45%; Tau² = 0.02; Chi² = 14.42 (P <0.07)) (Analysis 1.13) between women who had been treated with insulin and those treated with oral anti‐diabetic agents. Heterogeneity was substantial for all of these outcomes. Median data for fasting blood glucose concentration, one‐hour postprandial glucose concentration and HbA1c were reported by Hickman 2013 (Table 8). Waheed 2013 and Riaz 2014 report data as the number of women who had values for glycaemic control within the treatment target (Table 8).

1.14 Weight gain in pregnancy

Insulin was associated with an increase in gestational weight gain compared with oral anti‐diabetic pharmacological therapy (average mean difference (MD) 1.06 kg, 95% CI 0.63 to 1.48; 10 studies, 2336 women; I² = 65%, Tau² = 0.23) . The subgroup interaction test based on type of oral anti‐diabetic pharmacological therapy was not significant suggesting no differential effect between glibenclamide, metformin, or acarbose (I² = 0%, Chi² = 0.69, P = 0.71) (Analysis 1.14). Median data for gestational weight gain were reported by Hickman 2013 (Table 8). None of the studies detailed wither the gestational weight gain was within, above or below acceptable standards, we suggest caution when interpreting these data.

1.15 Induction of labour

Insulin may possibly increase the risk of induction of labour compared with oral anti‐diabetic pharmacological therapy although the evidence was not clear (average RR 1.30, 95%CI 0.96 to 1.75; three studies, 348 women; I² = 32%; Tau² = 0.02; moderate quality of evidence;Analysis 1.15). Quality of evidence was downgraded for risk of bias as all three studies were not blinded and there was insufficient detail to be able to judge allocation concealment and randomisation. All three studies reporting this outcome (Hague 2003; Ijas 2011; Tertti 2013) used metformin as the oral anti‐diabetic agent.

1.16 Postpartum haemorrhage

There was no evidence of a difference for the risk of postpartum haemorrhage between women who had been treated with insulin and those who had been treated with an oral anti‐diabetic pharmacological therapy (RR 0.34, 95% CI 0.01 to 8.13; two studies, 91 women; Analysis 1.16). One of the two studies reported no events in the insulin nor in the oral anti‐diabetic group (Hickman 2013). There is evidence of imprecision with wide CIs, small sample size and low event rates.

1.17 Breastfeeding

Breastfeeding was poorly reported. There was no evidence of a difference for breastfeeding at six to eight weeks postpartum between women who had been treated with insulin and those who had been treated with an oral anti‐diabetic pharmacological therapy (RR 1.03, 95% CI 0.86 to 1.23; two studies, 411 women) (Analysis 1.17). Both studies (Ijas 2011; Rowan 2008) used metformin as the oral anti‐diabetic.

1.18 Relevant biomarker changes associated with the intervention

There was no clear evidence of a difference between insulin and oral anti‐diabetic pharmacological therapy (metformin) for any of the reported biomarkers associated with the intervention: HOMA‐IR (MD 0.00, 95% CI ‐2.92 to 2.92; one study, 78 women), total cholesterol (MD 0.10, 95% CI ‐0.49 to 0.69; one study, 78 women), high‐density lipoprotein (HDL) cholesterol (MD 0.10, 95% CI ‐0.13 to 0.33; one study, 78 women) or triglycerides (MD ‐0.30, 95% CI ‐0.68 to 0.08; one study, 78 women).

Views of the intervention (data not shown)

Women in the Rowan 2008 trial were asked about the acceptability of the intervention one week postpartum. Of those treated with metformin, 76% said they would choose metformin if needed in a subsequent pregnancy whereas only 27% of women would choose insulin again. Women in the metformin group thought that taking medication was the easiest part of treatment compared with women taking insulin (59% versus 35%).

Adherence to the intervention (data not shown)

Sixty‐nine per cent of women taking metformin in the Rowan 2008 trial reported that they never or rarely forgot to take their medication compared with 81% of women taking insulin.

No data were reported for placental abruption, postpartum infection; perineal trauma/tearing; maternal mortality, sense of well‐being or quality of life or behavioural changes associated with the intervention.

1.19 Body mass index (BMI)

There was no evidence of a difference between women who had been treated with insulin and those treated with oral anti‐diabetic pharmacological therapy for BMI at six weeks postpartum (MD ‐0.20 kg/m², 95% CI ‐1.29 to 0.89; one study, 733 women) (Analysis 1.19).

1.20 Postnatal weight retention or return to pre‐pregnancy weight

Maternal weight (kg) was reported from the Tertti 2013 study at six to eight weeks' postpartum and at one‐year postpartum. There was no clear evidence of difference in postnatal weight between women treated with insulin and those treated with oral anti‐diabetic pharmacological therapy (metformin) at six to eight weeks postpartum (MD ‐1.60 kg, 95% CI ‐6.34 to 3.14; one study, 167 women; low‐quality evidence) or one year postpartum (MD ‐3.70 kg, 95% CI ‐8.50 to 1.10; one study, 176 women; low‐quality evidence) (Analysis 1.20).

1.21 Impaired glucose tolerance

Three studies reported on impaired glucose tolerance at follow‐up (Mirzamoradi 2015; Rowan 2008; Tertti 2013). At six weeks postpartum there was no clear evidence of a difference for impaired glucose tolerance between women who had been treated with insulin and those with oral anti‐diabetic agents (RR 1.16, 95% CI 0.80 to 1.68; three studies, 841 women). Nor was there evidence of a difference between groups at one‐year postpartum reported in one study (Tertti 2013) (RR 0.84, 95% CI 0.56 to 1.26; one study, 179 women). See Analysis 1.21.

No data were reported for postnatal depression or cardiovascular health.

1.22 Stillbirth

There was no evidence of a clear difference between infants whose mothers had been treated with insulin and those treated with oral anti‐diabetic pharmacological therapy for the risk of stillbirth (RR 0.60, 95% CI 0.08 to 4.52; three studies, 653 infants) (Analysis 1.22).

1.23 Neonatal death

There was no evidence of a clear difference between infants whose mothers had been treated with insulin and those treated with oral anti‐diabetic pharmacological therapy for the risk of neonatal death (RR 0.99, 95% CI 0.06 to 15.72; two studies, 503 infants) (Analysis 1.23).

1.24 Macrosomia

There was no overall evidence of a clear difference between infants whose mothers had been treated with insulin and those treated with oral anti‐diabetic pharmacological therapy for the risk of macrosomia (average RR 1.17, 95% CI 0.77 to 1.78; 19 studies, 2305 infants; I² = 42%, Tau² = 0.27; Chi² = 27.47 (P = 0.04)). The subgroup difference was not clear suggesting there was no differential effect between the type of oral anti‐diabetic agent used. The Bertini 2005 study of 33 infants reported no events for macrosomia in either the insulin or the acarbose groups. See Analysis 1.24.

1.25 Small‐for‐gestational age (SGA)

There was no evidence of a clear difference between insulin and oral anti‐diabetics for the risk of being born SGA (RR 1.16, 95% CI 0.79 to 1.69; nine studies, 1812 infants; Analysis 1.25). One study (Mesdaghinia 2013) reported no events in either the insulin or the oral anti‐diabetic group.

1.26 Birth trauma

Birth trauma (not defined) was reported in five studies. There were no events reported in four studies (Bertini 2005; Lain 2009; Saleh 2016; Silva 2007). One study reported no differences for the risk of birth trauma (not specified) for insulin compared with metformin (Rowan 2008) (RR 1.04, 95% CI 0.53 to 2.03; one study, 733 infants) (Analysis 1.26).

There was no evidence of a clear difference for the risk of shoulder dystocia between insulin and oral antidiabetics groups (RR 1.44, 95% CI 0.62 to 3.34; eight studies, 968 infants; Analysis 1.26), the risk of bone fracture (RR 4.71, 95% CI 0.23 to 95.53; two studies, 196 infants) or nerve palsy (RR 5.05, 95% CI 0.24 to 103.90; two studies, 320 infants; Analysis 1.26).

1.27 Gestational age at birth

There was no evidence of a clear difference between insulin and oral anti‐diabetics for gestational age at birth (average MD weeks ‐0.01, 95% CI ‐0.20 to 0.18; 18 studies, 2834 infants; I² = 59%, Tau² = 0.09; Chi² = 43.71 (P = 0.0006); Analysis 1.27). Median data for gestational age at birth were reported by Hickman 2013 (Table 9).

1.28 Preterm birth (less than 37 weeks’ gestation)

For preterm birth less than 37 weeks' gestation, there was no evidence of a clear difference between insulin and oral anti‐diabetics (average RR 1.10, 95% CI 0.64 to 1.88; 11 studies, 2417 infants; I² = 65%, Tau² = 0.45; Chi² = 28.57 (P = 0.001); Analysis 1.28). No data were presented for preterm birth less than 32 weeks' gestation.

1.29 Congenital abnormality (not pre‐specified)

There was no evidence of a clear difference between insulin and oral anti‐diabetics for the risk of congenital abnormality (RR 1.35, 95% CI 0.88 to 2.08; 15 studies, 2671 infants; Analysis 1.29). Three studies reported no events in either intervention group (De Veciana 2002; Hickman 2013; Zangeneh 2014).

1.30 Five‐minute Apgar less than seven

There was no evidence of a clear difference between insulin and oral anti‐diabetics for a five‐minute Apgar score less than seven (RR 0.49, 95% CI 0.09 to 2.64; four studies, 1170 infants; Analysis 1.30) . There were no events in two of the studies (Mesdaghinia 2013; Ruholamin 2014).

1.31 Birthweight

Birthweight did not differ overall between infants whose mothers had been treated with insulin and those treated with oral anti‐diabetic agents (MD ‐20.14 g, 95% CI ‐83.58 to 43.29; 22 studies, 3183 infants; I² = 67%, Tau² = 14034.8; Chi² = 67.47 (P < 0.00001)). The subgroup interaction test including glibenclamide, metformin and acarbose was not clear suggesting there was no differential effect between the oral anti‐diabetic agents used (Analysis 1.31). Data for median birthweight were reported by Hickman 2013 (Table 9).

1.32 Head circumference at birth (cm)

There was no evidence of a difference between insulin and oral anti‐diabetics for head circumference at birth (average MD 0.14 cm, 95% CI ‐0.26 to 0.53; three studies, 975 infants; I² = 65%, Tau² = 0.08; Chi² = 5.71 (P = 0.06); Analysis 1.32).

1.33 Length at birth (cm)

There was no evidence of a difference between insulin and oral anti‐diabetics for length at birth (average MD 0.11 cm, 95% CI ‐0.50 to 0.71; three studies, 975 infants I² = 61%, Tau² = 0.17; Chi² = 5.10 (P = 0.08); Analysis 1.33).

1.34 Ponderal index at birth

Ponderal index did not differ overall between infants whose mothers had been treated with insulin and those treated with oral anti‐diabetics (average MD 0.03 kg/m³, 95% CI ‐0.13 to 0.19; two studies, 815 infants; I² = 78%; Tau² = 0.01; Chi² = 4.46 (P = 0.03); very low‐quality evidence). The subgroup interaction test indicated a differential effect between these studies, one of which compared insulin with glibenclamide (Lain 2009) and showed no difference in ponderal index and one study that compared insulin with metformin (Rowan 2008), which showed a higher ponderal index for infants whose mothers had been treated with insulin (Analysis 1.34). Evidence is based on only two studies and caution should be taken when interpreting the data.

1.35 to 1.38 Adiposity at birth

Skinfold thickness measurements at birth (mm) ‐ There was no clear difference between intervention groups for triceps (MD ‐0.07 mm, 95% CI ‐0.25 to 0.10; two studies, 815 infants; Analysis 1.35), subscapular (average MD ‐0.13 mm, 95% CI ‐0.50 to 0.24; two studies, 815 infants; I² = 52%, Tau² = 0.04; Chi² = 2.09, (P = 0.15); Analysis 1.36) or skinfold sum (MD ‐0.80 mm, 95% CI ‐2.33 to 0.73, one study, 82 infants; very low‐quality evidence).

Percentage (%)fat mass ‐ Percentage fat mass did not clearly differ between intervention groups (MD ‐1.60%, 95% CI ‐3.77 to 0.57; one study, n = 82 infants; moderate‐quality evidence;Analysis 1.38).

1.39 Neonatal hypoglycaemia

Neonatal hypoglycaemia did not differ overall between infants whose mothers had been treated with insulin and those treated with oral anti‐diabetic agents (RR 1.14, 95% CI 0.85 to 1.52; 24 studies; 3892 infants; I² = 47%; Tau² = 0.18; Chi² = 43.73 (P = 0.006); low‐quality evidence;Analysis 1.39). The subgroup interaction test suggests a differential effect between the oral anti‐diabetic agents used (Chi² = 11.11, df = 3, P = 0.01, I² = 73%; Analysis 1.39). Insulin was associated with a decreased risk of neonatal hypoglycaemia compared with glibenclamide (RR 0.70, 95% CI 041 to ;1.19; 10 studies, 1283 infants) and an increased risk for neonatal hypoglycaemia when compared with metformin (RR 1.58, 95% CI 1.16 to 2.16; 12 studies, 2424 infants). There was no clear difference between groups for insulin compared with acarbose (RR 0.44, 95% CI 0.02 to 10.16; one study, 33 infants) or insulin compared with combined metformin and glibenclamide (RR 0.76, 95% CI 0.49 to 1.18; two studies, 152 infants; Analysis 1.39) although these latter two comparisons are based on small studies of participants and low event rates.

1.40 Respiratory distress syndrome

There was no evidence of a clear difference between intervention groups for the risk of respiratory distress syndrome (RR 1.29, 95% CI 0.83 to 1.99; 10 studies, 1894 infants; Analysis 1.40).

1.41 Neonatal jaundice (hyperbilirubinaemia)

There was no clear evidence of a difference between intervention groups for the risk of neonatal jaundice (RR 0.99, 95% CI 0.83 to 1.19; 16 studies, 2183 infants; Analysis 1.41) . The subgroup interaction test was not significant suggesting no differential effect between the oral antidiabetic pharmacological therapies.

1.42 Hypocalcaemia

For this outcome there was no evidence of a clear difference between exposure to insulin or oral anti‐diabetic pharmacological therapy (glibenclamide) (RR 1.95, 95% CI 0.49 to 7.78; five studies, 939 infants; Analysis 1.42) . Caution is advised in interpreting the results due to wide CIs and low event rates (5/487 insulin group; 2/452 oral antidiabetic group).

1.43 Polycythaemia

For the outcome of polycythaemia, there was no evidence of a clear difference between exposure to insulin or oral anti‐diabetic pharmacological therapy (glibenclamide) (RR 1.16, 95% CI 0.38 to 3.57; three studies, 590 infants) (Analysis 1.43). Caution is advised in interpreting the results due to wide CIs and low event rates (6/308 insulin group; 5/282 oral antidiabetic group).

1.44 to 1.45 Relevant biomarker changes associated with the intervention

There was no evidence of a clear difference between exposure to insulin or oral anti‐diabetic pharmacological therapy (glibenclamide) for cord blood C‐peptide concentration (MD ‐0.20 ng/mL, 95% CI ‐0.82 to 0.42; one study, 59 infants, Analysis 1.44) or cord blood insulin concentration (SMD 0.03, 95% CI ‐0.15 to 0.21; three studies, 486 infants; Analysis 1.45). Data were reported for median values for cord‐C peptide by Hague 2003, Hickman 2013. The data were not included in the meta‐analysis (Table 9).

No data were reported in the included studies for perinatal death; z scores for birthweight, head circumference or length.

1.46 Childhood weight

At six months of age there was no evidence of a clear difference in weight between infants whose mothers had been treated with insulin and those who had been treated with oral anti‐diabetic agents (MD ‐0.35 kg, 95% CI ‐0.75 to 0.05; one study, 93 children). At 12 months maternal treatment with insulin was associated with reduced weight compared with infants whose mothers had been treated with oral anti‐diabetic pharmacological therapy (MD ‐0.62 kg, 95% CI ‐1.18 to ‐0.06; one study, 93 children). At 18 months to two years there continued to be reduced weight associated with maternal treatment with insulin compared with infants whose mothers had been treated with oral anti‐diabetic agents (MD ‐0.44 kg, 95% CI ‐0.83 to ‐0.05, two studies, 411 children; Analysis 1.46).

1.47 Infant/Childhood height

There was no evidence of a clear difference in height between infants whose mothers had been treated with insulin and those who had been treated with oral anti‐diabetic pharmacological therapy at six months of age (MD ‐0.70 cm, 95% CI ‐2.05 to 0.65, one study, 93 infants), 12 months (MD ‐1.30 cm, 95% CI 2.60 to 0.00; one study, 93 children) or at 18 months to two years (average MD ‐0.65 cm, 95% CI ‐2.61 to 1.31; two studies; 411 children; I² = 80%, Tau² = 1.61; Chi² = 5.10 (P = 0.02); Analysis 1.47).

1.48 to 1.49 Infant/childhood adiposity

There was no evidence of a clear difference in ponderal index between infants whose mothers had been treated with insulin and those who had been treated with oral anti‐diabetic pharmacological therapy at six months (MD ‐1.00 kg/m³, 95% CI ‐3.43 to 1.43; one study, 93 children); 12 months (MD ‐0.20 kg/m³, 95% CI ‐1.12 to 0.72; one study, 93 children) or at 18 months to two years (MD ‐0.10 kg/m³, 95% CI ‐0.88 to 0.68; one study, 93 children) (Analysis 1.48). There was no evidence of a clear difference in childhood total fat mass (%) at up to two years of age between infants whose mothers had been treated with insulin and those who had been treated with oral anti‐diabetic pharmacological therapies (MD 0.50%, 95% CI ‐0.49 to 1.49; one study, 318 children;low‐quality evidence;Analysis 1.49).

1.50 Childhood blood pressure (at two years)

The children of mothers randomised in the Rowan 2008 study were followed up at two years of age (median age 29 months, range 22 to 38 months). Of the original 751 women randomised, 733 (98%) completed the study. At two years 577 (78%) were eligible to be seen of which 323 (56%) were seen at follow‐up. Blood pressure readings were obtained from 83/154 (54%) of children whose mother had been treated in the metformin arm and 87/164 (53%) who had been treated in the insulin arm. There was no evidence of a difference for systolic blood pressure (MD ‐2.24 mmHg, 95% CI ‐5.02 to 0.54; one study, 170 children) or diastolic blood pressure (MD ‐0.50 mmHg, 95% CI ‐16.75 to 15.75; one study, 170 children; Analysis 1.50).

No data were reported for z scores for height or weight, BMI, educational attainment, dyslipidaemia/metabolic syndrome or diabetes (type 1, type 2, or impaired glucose tolerance).

1.51 Number of antenatal visits or admissions

There was no evidence of differences in the number of clinic visits between women treated with insulin and those treated with oral anti‐diabetic pharmacological therapies (RR 1.00, 95% CI ‐0.08 to 2.08; one study, 404 women; Analysis 1.51).

1.52 Admission to neonatal intensive care unit/nursery

Maternal treatment with insulin was associated with an increased risk of the infant being admitted to the neonatal intensive care unit or special care baby unit compared with the infants of women treated with oral anti‐diabetic pharmacological therapy (RR 1.38, 95% CI 1.19 to 1.59; 18 studies, 3441 infants; Analysis 1.52).

1.53 Duration of stay in neonatal intensive care unit or special care baby unit

There was no evidence of a difference in the duration of stay between the groups (average MD ‐0.20 days, 95% CI ‐1.79 to 1.39; three studies, 401 infants; I² = 72%; Tau² = 1.19; Chi² = 7.18 (P = 0.03); Analysis 1.53). Hague 2003 reported data for median duration of stay in the special care nursery and duration of hospital stay for the infant. The data were not in a suitable format to be included in a meta‐analysis (Table 9).

Costs associated with the intervention

Costs of treatment were reported by one study (Ogunyemi 2007). They provide data on the cost of drugs per month which was US$20 for insulin and US$7 for glibenclamide. No other details are provided. Ashoush 2016 reported costs of 174.9 Egyptian pounds for insulin treatment and 89.66 Egyptian pounds for metformin only, or for combined metformin and insulin the cost was 159.48 Egyptian pounds. This did not include the cost of syringes for insulin treatment.

No data were reported for the following health service use outcomes: number of hospital or health professional visits (including midwife, obstetrician, physician, dietician, diabetic nurse); length of antenatal stay, length of postnatal stay (maternal), length of postnatal stay (baby), cost of maternal care, cost of offspring care, costs to families associated with the management provided, cost of dietary monitoring (e.g. diet journals, dietician, nurse visits, etc), costs to families ‐ change of diet, extra antenatal visits, extra use of healthcare services (consultations, blood glucose monitoring, length and number of antenatal visits) or women’s view of treatment advice.

2.1 Hypertensive disorders of pregnancy (including pre‐eclampsia, pregnancy‐induced hypertension, eclampsia

There were no events of pre‐eclampsia reported in one study comparing regular human insulin with insulin aspart in 320 women (Analysis 2.1) (Balaji 2012).

2.2 Caesarean section

There was no evidence of a clear difference in the caesarean section rate between women treated with regular human insulin compared with another insulin preparation (RR 1.00, 95% CI 0.91 to 1.09, three studies, 410 women). There was no evidence of a clear difference between regular human insulin and insulin aspart (RR 1.02, 95% CI 0.94 to 1.10; one study, 320 women) or insulin lispro (RR 0.81, 95% CI 0.42 to 1.56; two studies, 90 women). See Analysis 2.2.

Development of type 2 diabetes

No studies included in this review comparing one preparation of insulin with another reported on development of type 2 diabetes for the woman.

2.3 Large‐for‐gestational age (LGA)

There was no evidence of a cleat difference in the risk of being born LGA between women treated with human insulin compared with another insulin preparation (RR 1.21, 95% CI 0.58 to 2.55; three studies, 411 infants). There was no evidence of a clear difference between regular human insulin and insulin aspart (RR 1.14, 95% CI 0.50 to 2.61; one study, 320 infants) or insulin lispro (RR 1.56, 95% CI 0.29 to 8.55; two studies, 91 infants). See Analysis 2.3.

No studies included in this review comparing one preparation of insulin with another reported on perinatal death (fetal and neonatal death) and later infant mortality, death or serious neonatal morbidity composite or neurosensory disability.

2.4 Use of additional pharmacotherapy

There was no evidence of a clear difference between regular human insulin and other types of insulin for the use of additional pharmacotherapy (average RR 1.11, 95% CI 0.72 to 1.70; three studies, 168 women; I² = 56%; Tau² = 0.08; Chi² = 4.50 (P = 0.11)). There was no evidence of a clear difference for the use of additional pharmacotherapy between regular human insulin and insulin aspart (RR 1.33, 95% CI 0.83 to 2.14; one study, 47 women) insulin lispro (RR 1.38, 95% CI 0.90 to 2.09; two studies, 98 women) or Neutral Protamine Hagedorn insulin (RR 0.65, 95% CI 0.36 to 1.19; one study, 23 women). See Analysis 2.4.

2.5 Maternal hypoglycaemia

There was no evidence of a clear difference between regular human insulin and other types of insulin for the risk of maternal hypoglycaemia (RR 0.96, 95% CI 0.64 to 1.44; four studies, 504 women). There was no evidence of a clear difference for the risk of maternal hypoglycaemia between regular human insulin and insulin aspart (RR 0.90, 95% CI 0.59 to 1.35; three studies, 394 women), or Neutral Protamine Hagedorn insulin (RR 5.16, 95% CI 0.26 to 103.25; one study, 61 women). See Analysis 2.5. There were no events of maternal hypoglycaemia reported in the study by Di Cianni 2007 that compared regular human insulin with insulin lispro and with insulin aspart.

2.6 to 2.8 Glycaemic control during/after treatment

There was no evidence of a clear difference between human insulin and other insulin‐treated women for maternal HbA1c at the end of treatment (average MD 0.04%, 95% CI ‐0.06 to 0.14; three studies, 411 women; I² = 46%; Tau² = 0.00; Chi² = 3.74 (P = 0.15); Analysis 2.6). We explored the heterogeneity by looking at the type of insulin used. The subgroup interaction test suggested there was not a differential effect between regular human insulin compared with either insulin lispro or insulin aspart. The Di Cianni 2007 study reported no differences for HbA1c between regular human insulin compared with insulin aspart or insulin lispro but did not tabulate the data.

There was no evidence of a clear difference between regular human insulin and other insulin‐treated women for fasting plasma glucose (MD 0.15 mg/dL, 95% CI ‐2.02 to 2.32; four studies, 466 women) Analysis 2.7). The subgroup interaction test was not significant, suggesting there was not a differential effect between regular human insulin compared with either insulin lispro or insulin aspart. Herrera 2015 presented data for fasting plasma glucose, but combined GDM and type 2 diabetes which cannot be separated. This is a conference abstract only and we will wait for full publication to see if data are separated and will contact the authors. The Di Cianni 2007 study reported no differences for fasting plasma glucose between regular human insulin compared with insulin aspart or insulin lispro but did not tabulate the data.

There was no evidence of a clear difference between regular human insulin and other insulin preparations for postprandial glucose concentration (average SMD 0.08, 95% CI ‐0.25 to 0.42; five studies, 562 women; I² = 61%, Tau² = 0.10; Chi² = 12.89 (P = 0.02); Analysis 2.8). Subgroup analysis suggested no differential effect between regular human insulin and insulin aspart, insulin lispro or insulin detemir. Herrera 2015 presented data for postprandial plasma glucose but combined GDM and type 2 diabetes which cannot be separated. This is a conference abstract only and we will wait for full publication to see if data are separated and will contact authors.

Median data for HbA1c values were reported by Ismail 2007 (Table 8).

2.9 Weight gain in pregnancy

There was no evidence of a clear difference between regular human insulin and other insulin‐treated (insulin aspart, Neutral Protamine Hagedorn insulin) women for weight gain in pregnancy (MD 0.46 kg, 95% CI ‐0.15 to 1.07; two studies, 407 women; Analysis 2.9). The Di Cianni 2007 study reported no differences for weight gain between regular human insulin compared with insulin aspart or insulin lispro but did not tabulate the data. Median data for gestational weight gain were reported by Mecacci 2003 (Table 8).

2.10 Maternal mortality

There were no events of maternal mortality in the one study, of 61 women, that reported this outcome (Analysis 2.10) (Ismail 2007)

Views of the intervention

Balaji 2005 reported that women were more comfortable with insulin aspart as it was administered just before a meal. No other data were reported.

No data were reported for adherence to the intervention; induction of labour; placental abruption, postpartum haemorrhage; postpartum infection; perineal trauma/tearing; breastfeeding at discharge, six weeks postpartum, six months or longer, quality of life, behavioural changes associated with the intervention or relevant biomarker changes associated with the intervention.

2.11 Stillbirth

Fetal death was reported in four studies including 508 infants (Balaji 2012; Ismail 2007; Pettitt 2007; Prasad 2008). One death due to umbilical cord strangulation was reported by the Pettitt 2007 study (human insulin versus insulin aspart). The other studies reported no deaths in either intervention or control groups (RR 0.36, 95% CI 0.02 to 8.06; four studies, 508 infants; Analysis 2.11).

2.12 Macrosomia

There was no evidence of an overall difference for the risk of being born macrosomic between infants whose mothers had been treated with regular human insulin and those treated with another insulin analogue (average RR 0.99, 95% CI 0.40 to 2.46, six studies, 627 infants; Analysis 2.12). There was evidence of a subgroup difference (Chi² = 5.83, df = 2, P = 0.05, I² = 65.7%). Regular human insulin was associated with reduced risk of macrosomia compared with Neutral Protamine Hagedorn insulin (RR 0.11, 95% CI 0.01 to 0.79; two studies, 84 infants), while there was no evidence of a clear difference between regular human insulin and insulin aspart (RR 1.47, 95% CI 0.71 to 3.05; four studies, 494 infants) or insulin lispro (RR 1.03, 95% CI 0.21 to 5.05; one study, 49 infants).

2.13 Small‐for‐gestational age (SGA)

There was no evidence of a clear difference between regular human insulin and insulin lispro in a single study reporting data for 49 infants (RR 1.04, 95% CI 0.07 to 15.73; one study, 49 infants, Analysis 2.13).

2.14 Birth trauma

There was no evidence of a clear difference between human insulin and Neutral Protamine Hagedorn insulin for the risk of nerve palsy in a single small study reporting data for 23 infants (RR 0.36, 95% CI 0.02 to 8.04; one study, 23 infants, Analysis 2.14). No data were reported for bone fracture or shoulder dystocia or birth trauma (not defined).

2.15 Gestational age at birth

Two studies reported data for gestational age at birth (Balaji 2012; Jovanovic 1999). Balaji 2012 compared regular human insulin with insulin aspart and Jovanovic 1999 compared regular human insulin with insulin lispro. The data were not combined in a meta‐analysis as heterogeneity was I² = 92%, Tau² = 0.21 (data not shown). Balaji 2012 found that regular human insulin was associated with a lower gestational age at birth compared with insulin aspart (MD ‐0.67 weeks, 95% CI ‐1.01 to ‐0.33; one study, 320 infants; Analysis 2.15), there was no evidence of a clear difference reported between regular human insulin and insulin lispro in the Jovanovic 1999 study (MD 0.00 weeks, 95% CI ‐0.16 to 0.16; one study, 41 infants; Analysis 2.15). Median data for gestational age at birth were reported by Mecacci 2003 (Table 9).

2.16 Preterm birth (less than 37 weeks’ gestation)

Three studies reported data for preterm birth (less than 37 weeks' gestation). Two studies compared regular human insulin with insulin aspart (Balaji 2012; Prasad 2008) and one study compared regular human insulin with Neutral Protamine Hagedorn insulin (Poyhonen‐Alho 2002). There was no evidence of a clear difference between regular human insulin and another insulin analogue (RR 2.29, 95% CI 0.52 to 10.05; three studies, 443 infants; Analysis 2.16). Caution is advised in interpreting these data as there are wide CIs and low event rates (5/218 for human insulin; 2/225 for other insulin). No data were reported for preterm birth < 32 weeks' gestation.

2.17 Congenital anomaly (not pre‐specified)

Two studies including 69 infants reported data for congenital anomaly. There was no clear evidence of a difference between regular human insulin and other forms of insulin (RR 3.21, 95% CI 0.14 to 72.55; two studies, 69 infants). Pettitt 2007 compared regular human insulin with insulin aspart and Jovanovic 1999 compared regular human insulin with insulin lispro. Pettitt 2007 reported one event in the regular human insulin group. Jovanovic 1999 reported no events in either the regular human insulin or insulin lispro group (Analysis 2.17).

2.18 Birthweight

Overall, there was no evidence of a clear difference between regular human insulin and other insulin analogues for birthweight (average MD ‐0.04 g, 95% CI ‐0.17 to 0.08; seven studies, 531 infants; I² = 62%, Tau² = 0.01; Chi² = 15.68 (P = 0.02); Analysis 2.18). Three studies compared regular human insulin with insulin aspart (Balaji 2005; Balaji 2012; Pettitt 2007) (MD ‐0.01 g, 95% CI ‐0.11 to 0.09; three studies, 357 infants); two studies compared regular human insulin with insulin lispro (Jovanovic 1999; Mecacci 2003) (MD 0.04 g, 95% CI ‐0.07 to 0.14; two studies, 90 infants) and two studies compared regular human insulin with Neutral Protamine Hagedorn insulin (Ismail 2007; Poyhonen‐Alho 2002) (MD ‐0.44 g; 95% CI ‐1.20 to 0.32; two studies, 84 infants, I² = 86%; Tau² = 0.26; Chi² = 7.28 (P = 0.007)). Di Cianni 2007 reported an increased birthweight in regular human insulin group compared with lispro and aspart groups but provided no data.

2.19 Length (cm) at birth

There was no clear evidence of a difference for length at birth between maternal treatment with regular human insulin compared with other insulin analogues (insulin lispro and aspart) (MD ‐0.11 cm, 95%CI ‐0.57 to 0.34; three studies, 388 infants) (Analysis 2.19).

2.20 Ponderal index

One study compared regular human insulin with insulin aspart (MD ‐0.10 kg/m³, 95% CI ‐1.01 to 0.81; one study, 320 infants; Balaji 2012) and one study compared human insulin with insulin lispro (MD 0.03 kg/m³, 95% CI ‐0.06 to 0.12; one study, 49 infants; Mecacci 2003). There was no evidence of a clear difference overall between regular human insulin and another insulin preparation for ponderal index (MD 0.03 kg/m³, 95% CI ‐0.06 to 0.12; two studies, 369 infants; Analysis 2.20).

2.21 Neonatal hypoglycaemia

There was no clear evidence of an overall difference between regular human insulin and another insulin preparation for the risk of neonatal hypoglycaemia (RR 2.28, 95% CI 0.06 to 82.02; three studies, 165 infants; I² = 65%, Tau² = 4.37; Chi² = 2.89 (P = 0.09); Analysis 2.21). One study found no evidence of a clear difference in the risk for neonatal hypoglycaemia between infants whose mothers had been treated with regular human insulin plus Neutral Protamine Hagedorn insulin or insulin aspart (Prasad 2008) (RR 13.00, 95% CI 0.75 to 224.77; one study, 100 infants). One study that compared regular human insulin and insulin lispro (Jovanovic 1999) reported no events of neonatal hypoglycaemia in either the regular human insulin or the insulin lispro group. There was no evidence of a difference in the risk of neonatal hypoglycaemia between infants whose mothers had been treated with regular human insulin or Neutral Protamine Hagedorn insulin (RR 0.36, 95C% CI 0.02 to 8.04; one study, 23 infants). Data should be interpreted with caution due to low event rates, low sample size and wide CIs.

2.22 Respiratory distress syndrome

There was no evidence of a difference in the risk of respiratory distress syndrome between infants whose mothers had been treated with regular human insulin or insulin aspart (RR 0.52, 95% CI 0.10 to 2.79; one study, 320 infants, Analysis 2.22).

2.23 Neonatal jaundice (Hyperbilirubinaemia)

Overall there was no evidence of a difference in the risk of hyperbilirubinaemia between infants whose mothers had been treated with regular human insulin and those treated with another insulin analogue (average RR 0.48, 95% CI 0.05 to 4.93; two studies, 123 infants; I² = 57%; Tau² = 1.74; Chi² = 2.30, (P = 0.13); Analysis 2.23). There was no clear evidence of a difference between groups for regular human insulin plus Neutral Protamine Hagedorn insulin compared to insulin aspart reported by Prasad 2008 (RR 0.11, 95% CI 0.01 to 2.01; one study, 100 infants) or between regular human insulin compared with Neutral Protamine Hagedorn insulin reported by Poyhonen‐Alho 2002 (RR 1.09, 95% CI 0.28 to 4.32, one study, 23 infants). Data should be interpreted with caution due to low event rates, low sample size and wide CIs.

2.24 Hypocalcaemia

No events of hypocalcaemia were reported in infants whose mothers had been treated with regular human insulin or insulin lispro in a single small study of 42 infants (Jovanovic 1999) (Analysis 2.24) .

No data were reported for neonatal death; perinatal death; five‐ minute Apgar less than seven; head circumference; z scores for birthweight, head circumference, length; measures of adiposity; polycythaemia or relevant biomarker changes associated with the intervention cord blood measures.

Hypertensive disorders of pregnancy (including pre‐eclampsia, pregnancy‐induced hypertension, eclampsia)

Pre‐eclampsia, pregnancy‐induced hypertension or eclampsia were not reported as outcomes in any of the included studies in this review.

3.1 Caesarean section

There was no evidence of a difference in the risk of birth bycaesarean section between women treated with insulin and those treated with diet (RR 0.85, 95% CI 0.50 to 1.42, two studies, 133 women) (Analysis 3.1).

3.2 Development of type 2 diabetes

There was no evidence of a difference in the risk of developing type 2 diabetes postpartum between women treated with insulin and those treated with diet or standard antenatal care (RR 0.98, 95% CI 0.79 to 1.21; two studies, 653 women; Analysis 3.2). O'Sullivan 1975a reported follow‐up to a maximum of 15 years. Coustan 1978 reported data for follow‐up at five weeks postpartum.

3.3 Perinatal (fetal and neonatal)death and later infant mortality

There was no evidence of a difference in the risk of perinatal death between infants whose mothers had been treated with insulin and those who had been treated with diet/standard antenatal care (RR 0.74, 95% CI 0.41 to 1.33; four studies, 1137 infants; Analysis 3.3). No events of perinatal death were reported in two studies including 297 infants (Persson 1985; Thompson 1990).

3.4 Large‐for‐gestational age (LGA)

There was no difference in the risk of being born LGA between infants whose mothers had been treated with insulin or diet in one study including 202 infants (Persson 1985) (RR 0.85, 95% CI 0.41 to 1.78; one study, 202 infants, Analysis 3.4).

No data were reported for death or serious morbidity composite (variously defined by trials, e.g. perinatal or infant death, shoulder dystocia, bone fracture or nerve palsy) or neurosensory disability.

3.5 Use of additional pharmacotherapy

Use of additional pharmacotherapy was required by 14% (15/105) of women randomised to the dietary intervention groups reported in a single study including 202 women (Persson 1985). No analysis was conducted as the women in the insulin group all received insulin (data shown, see Analysis 3.5).

3.6 Maternal hypoglycaemia

There were no events of maternal hypoglycaemia reported on one study of 95 women (Thompson 1990) ( Analysis 3.6).

3.7 Glycaemic control during/end of treatment

Insulin was associated with a slight increase in the mean HbA1c at the end of treatment (6.9 mmol/mol ± 0.2; 86 women) compared with diet only treatment (6.8 mmol/mol ± 0.2; 75 women) (MD 0.10 mmol/mol, 95% CI 0.04 to 0.16; one study, 161 women, Analysis 3.7) (Persson 1985). There was no evidence of a difference in the mean fasting blood glucose concentration (MD 1.60 mg/dL, 95% CI ‐2.97 to 6.17, one study, 68 women, Analysis 3.7) or two‐hour postprandial glucose level (MD 0.30 mg/dL, 95% CI ‐5.32 to 5.92, one study, 68 women, Analysis 3.7) during treatment reported in a single study including 68 women (Thompson 1990). Persson 1985 reported maternal blood glucose in the insulin‐treated group and diet‐treated group, however the values are presented in a figure with median and 95% CI values. The data could not be included in a meta‐analysis and authors were unable to provide original data. HbA1c values at birth were estimated from a bar graph (Persson 1985). Coustan 1978 reported data for fasting glucose concentration and two‐hour postprandial glucose concentration at the end of treatment but the data are reported by the number of observations and not the number of women randomised. The data could not be included in a meta‐analysis but indicate a reduced fasting glucose concentration in the insulin groups compared with the diet group (Table 8). O'Sullivan 1975a reported data as the number of samples rather than per woman randomised and the data have therefore not been included in the meta‐analysis. Insulin was associated with reduced fasting blood glucose concentration (blood glucose concentration 80.1 mg/dL (SD 23); 295 samples) compared with routine antenatal care (blood glucose concentration 69.1 mg/dL (SD 16.9); 71 samples). At two to three hours postprandial, the authors report that insulin reduced blood glucose concentration (mean blood glucose concentration 80.1 mg/dL (SD 23); 295 samples) compared with routine antenatal care (mean blood glucose concentration 83.0 mg/dL (SD 21.2); 233 samples).

3.8 Weight gain during pregnancy

There was no evidence of a difference between insulin and dietary intervention groups for weight gain in pregnancy (MD 1.73 kg, 95% CI ‐3.31 to 6.77, one study, 38 women, Analysis 3.8) (Coustan 1978).

No data were reported in the included studies for the other maternal secondary outcomes pre‐specified in this review (use of adherence to the intervention; induction of labour; placental abruption; postpartum haemorrhage; postpartum infection; perineal trauma/tearing; breastfeeding at discharge; six weeks postpartum, six months or longer; maternal mortality; sense of well‐being and quality of life; behavioural changes associated with the intervention; views of the intervention or relevant biomarker changes associated with the intervention).

3.9 Neonatal death

There was no evidence of a difference for the risk of neonatal death between infants whose mothers had been treated with insulin and those who had been treated with standard antenatal care (RR 0.72, 95% CI 0.23 to 2.23; one study, 611 infants; Analysis 3.9).

3.10 Macrosomia

Maternal treatment with insulin was associated with a reduced risk of macrosomia in the infant compared with diet/standard antenatal care (RR 0.30, 95% CI 0.18 to 0.50; three studies, 717 infants; three studies; I² = 0%; Analysis 3.10). Of interest, these data differ from the single study reporting LGA which found no difference between groups. Two studies were undertaken in the 1970s (Coustan 1978; O'Sullivan 1975a) and one in 1990 (Thompson 1990). The insulin regimens may differ from those administered today and appear to be fixed doses of Neutral Protamine Hagedorn insulin, with or without regular human insulin rather than a weight adjusted dose which is seen in more recent studies. This may in part explain the reduced effectiveness of insulin for this outcome.

3.11 Small‐for‐gestational age (SGA)

There was no evidence of a difference between insulin‐treated and diet‐treated groups for the risk of the infant being born SGA (RR 0.35, 95% CI 0.05 to 2.40; two studies, 240 infants; Analysis 3.11).

3.12 Birth trauma (shoulder dystocia, bone fracture, nerve palsy)

There were no events of shoulder dystocia in the infants of mothers who had been treated with either insulin or the diet alone in two studies including 133 infants (Coustan 1978; Thompson 1990). There were no events of nerve palsy in either group reported from a single study including 38 infants (Coustan 1978). No data were reported for bone fracture in any of the included trials.

3.13 Gestational age at birth

There was no evidence of a difference for gestational age at birth between infants whose mothers had been treated with insulin or with diet (MD ‐0.66 weeks, 95% CI ‐1.37 to 0.06; two studies, 106 infants; Analysis 3.13).

3.14 Preterm birth (less than 37 weeks' gestation; and less than 32 weeks' gestation)

There was no evidence of a difference for the risk of preterm birth (less than 37 weeks' gestation) between infants whose mothers had been treated with insulin and those who had been treated with standard antenatal care (RR 1.09, 95% CI 0.64 to 1.85; one study, 611 infants, Analysis 3.14). No data were reported for preterm birth less than 32 weeks' gestation.

3.15 Birthweight

Maternal treatment with insulin was associated with a reduced birthweight compared with maternal treatment with diet alone (MD ‐342.85 g, 95% CI ‐561.11 to ‐124.60; two studies, 106 infants; Analysis 3.15).

3.16 Ponderal Index

Maternal treatment with insulin was associated with a reduced ponderal index compared with maternal treatment with diet alone (MD ‐0.18 kg/m³, 95% CI ‐0.34 to ‐0.02; one study, 68 infants; Analysis 3.16).

3.17 Neonatal hypoglycaemia

There was no evidence of a difference between infants whose mothers had been treated with insulin and diet‐treated groups for the risk of neonatal hypoglycaemia (RR 0.88, 95% CI 0.34 to 2.24; three studies, 176 infants; I² = 40%, Tau² = 0.24, Analysis 3.17).

3.18 Hyperbilirubinaemia

There were no events of hyperbilirubinaemia in a single study including 68 infants (Thompson 1990) (Analysis 3.18).

3.19 Hypocalcaemia

There were no events of hypocalcaemia in a single study including 68 infants (Thompson 1990) (Analysis 3.19).

3.20 Polycythaemia

There was no evidence of a difference between infants whose mothers had been treated with insulin and diet‐treated groups for the risk of polycythaemia (RR 0.90, 95% CI 0.30 to 2.67; one study, 70 infants; Analysis 3.20).

3.21 Relevant biomarker changes associated with the intervention (including insulin, cord c‐peptide)

Dietary interventions were associated with a reduction in cord blood C‐peptide concentration compared with insulin‐treated groups (MD 0.03 ng/mL, 95% CI 0.02 to 0.04; one study, 202 infants; Analysis 3.21).

No data were reported for other neonatal outcomes pre‐specified in this review (stillbirth; five‐minute Apgar less than seven; head circumference; length; z score for birthweight, head circumference and z score, length and z score; skinfold thickness measurements (mm); fat mass or respiratory distress syndrome).

Caesarean section

There was no evidence of a difference between treatment with insulin and treatment with exercise for the risk of birth by caesarean section (RR 1.50, 95% CI 0.29 to 7.87; one study, 34 women) (Analysis 4.1).

No data were reported for the outcomes of pre‐eclampsia or development of type 2 diabetes.

Macrosomia

There was no evidence of a difference for the risk of macrosomia between infants whose mothers had been treated with insulin and those treated with exercise (RR 2.00, 95% CI 0.42 to 9.50; one study, 34 infants) (Analysis 4.2).

Gestational age at birth

There was no evidence of a difference for the timing of gestational age at birth between infants whose mothers had been treated with insulin and those treated with exercise (MD ‐0.80 weeks, 95% CI ‐2.05 to 0.45, one study, 34 infants) (Analysis 4.3).

Birthweight (g)

There was no evidence of a difference in birthweight between infants whose mothers had been treated with insulin and those treated with exercise (MD 103.00 g, 95% CI ‐245.40 to 451.40; one study, 34 infants) (Analysis 4.4).

Length at birth (cm)

There was no evidence of a difference in length at birth (cm) between infants whose mothers had been treated with insulin or those treated with diet (MD 1.60 cm, 95% CI ‐0.01 to 3.21; one study, 34 infants) (Analysis 4.5).

Neonatal hypoglycaemia

There was no evidence of a difference for the risk of neonatal hypoglycaemia between infants whose mothers had been treated with insulin and those treated with exercise (RR 0.50, 95% CI 0.05 to 5.01; one study, 34 infants) (Analysis 4.6).

Respiratory distress syndrome

There were no events of respiratory distress reported in one study of 34 infants (Bung 1993).

Hyperbilirubinaemia

There were no events of hyperbilirubinaemia reported in one study of 34 infants (Bung 1993).

Hypocalcaemia

There were no events of hypocalcaemia reported in one study of 34 infants (Bung 1993).

There were no data reported for stillbirth; neonatal death; SGA, birth trauma, preterm birth, five‐minute Apgar less than seven minutes, birthweight z score, head circumference and z score, length z score, ponderal index, adiposity, polycythaemia, relevant biomarker changes associated with the intervention.

Pregnancy‐induced hypertension

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of pregnancy‐induced hypertension (RR 1.11, 95% CI 0.51 to 2.42; one study, 274 women) (Analysis 5.1).

Caesarean section

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of birth by caesarean section (RR 0.99, 95% CI 0.68 to 1.44; one study, 274 women). There was no evidence of a difference between a maternal regimen using three injections daily or six injections daily for the risk of birth by caesarean section (RR 1.06, 95% CI 0.17 to 6.72; one study, 37 women) (Analysis 5.2).

No data were reported for the development of type 2 diabetes in either of the included studies.

Perinatal (fetal and neonatal death) and later infant mortality

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of perinatal death (RR 3.04, 95% CI 0.13 to 74.07; one study, 274 women) (Analysis 5.3).

Large‐for‐gestational age (LGA)

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of being born LGA (RR 1.16, 95% CI 0.79 to 1.69; one study, 274 women). There was no evidence of a difference between a maternal regimen using three injections daily or six injections daily for the risk of being born LGA (RR 0.35, 95% CI 0.04 to 3.08; one study, 37 infants) (Analysis 5.4).

Composite of serious neonatal morbidity and mortality

A maternal regimen of insulin twice daily was associated with an increased risk of death or serious morbidity composite compared with a four times daily insulin regimen (RR 1.69, 95% CI 1.08 to 2.64; one study, 274 women) (Analysis 5.5).

No data were reported for neurosensory disability.

Maternal hypoglycaemia

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of maternal hypoglycaemia (RR 1.01, 95% CI 0.06 to 16.06; one study, 274 women) (Analysis 5.6).

Glycaemic control (end of treatment)

There was no evidence of a difference for mean fasting blood glucose concentration between women using a regimen of three injections per day compared with six injections per day (MD 4.00 mg/dL, 95% CI ‐0.84 to 8.84; one study, 37 women) (Analysis 5.7). A maternal regimen of insulin four times daily was associated with a decrease in HbA1c at the end of treatment compared with a twice‐daily regimen (MD 0.30%, 95% CI 0.06 to 0.54; one study, 274 women). There was no evidence of a difference for mean HbA1c between women using a regimen of three injections per day compared with six injections per day (MD ‐0.10%, 95% CI ‐0.29 to 0.09; one study, 37 women) (Analysis 5.8).

Weight gain in pregnancy

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for weight gain in pregnancy (MD 0.70 kg, 95% CI ‐0.14 to 1.54; one study, 274 women) (Analysis 5.9). Castorino 2011 reported no evidence of a difference in gestational weight gain per week 0.2 +/‐ 0.3 in the three injection regimen and 0.3 +/‐ 0.2 in the six injection regimen.

No data were reported for other pre‐specified maternal secondary outcomes for this review (use of additional pharmacotherapy, adherence to the intervention, induction of labour, placental abruption, postpartum haemorrhage, postpartum infection, perineal trauma/tearing, breastfeeding at discharge, six weeks postpartum, six months or longer, maternal mortality, sense of well‐being and quality of life, behavioural changes associated with the intervention, views of the intervention, relevant biomarker changes associated with the intervention).

Macrosomia

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of macrosomia (RR 1.20, 95% CI 0.72 to 2.01; one study, 274 women) (Analysis 5.10).

Small‐for‐gestational age (SGA)

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of being born SGA (RR 1.78, 95% CI 0.53 to 5.93; one study, 274 women) (Analysis 5.11).

Birth trauma (not specified)

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of birth trauma (RR 1.52, 95% CI 0.26 to 8.97; one study, 274 women) (Analysis 5.12).

Gestational age at birth

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the timing of gestational age at birth (MD ‐0.30 weeks, 95% CI ‐0.72 to 0.12; one study, 274 women) (Analysis 5.13).

Five‐minute Apgar less than seven

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for a five‐minute Apgar score less than seven (RR 0.34, 95% CI 0.07 to 1.65; one study, 274 women) (Analysis 5.14).

Birthweight

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the outcome of birthweight (MD ‐1.00 g, 95% CI ‐150.49 to 148.49; one study, 274 women). There was no evidence of a difference between a maternal regimen of three injections of insulin daily compared with six injections for the outcome of birthweight (MD ‐197.00 g, 95% CI ‐495.43 to 101.43; one study, 37 infants) (Analysis 5.15).

Neonatal hypoglycaemia

A maternal regimen of insulin twice daily was associated with an increased risk of neonatal hypoglycaemia compared with a regimen of insulin four times daily (RR 8.12, 95% CI 1.03 to 64.03; one study, 274 infants) (Analysis 5.16). Data should be interpreted with caution due to large treatment effect and wide CIs suggesting imprecision.

Respiratory distress syndrome

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of respiratory distress syndrome (RR 0.34, 95% CI 0.01 to 8.23; one study, 274 infants) (Analysis 5.17).

Hyperbilirubinaemia

A maternal regimen of insulin twice‐daily insulin was associated with an increased risk of hyperbilirubinaemia compared with a four times a day regimen (RR 1.96, 95% CI 1.10 to 3.49; one study, 274 infants) (Analysis 5.18).

Polycythaemia

There was no evidence of a difference between a maternal insulin regimen twice daily or four times daily for the risk of polycythaemia (RR 0.43, 95% CI 0.11 to 1.65; one study, 274 infants) (Analysis 5.19).

There were no data reported for other secondary neonatal outcomes pre‐specified for this review (stillbirth; neonatal death; preterm birth, birthweight z score, head circumference and z score, length z score, ponderal index, adiposity, relevant biomarker changes associated with the intervention).