Ethics, consent and permissions

The trial was performed in accordance with the Declaration of Helsinki. The present study is a planned secondary analysis of the NFFD trial, and was included in the consent and ethical approval of the trial. The Norwegian Regional Committee for Medical Research Ethics South-East-C approved the trial and modifications (REK reference 2009/429), including specific approval for the storage and analysis of frozen serum in Research Biobank Notification no. 2594. Signed, informed consent was obtained from all participants.

Randomisation and blinding

After receiving signed consent forms and confirming that blood tests and questionnaires were completed, a research nurse assigned participants consecutively to the intervention or control arm of the study utilizing a computer-generated list with 1:1 allocation ratio and blocks of 20. All examinations, blood test evaluations and scoring of questionnaires were performed by assessors blinded to group allocation.

Intervention

Details of NFFD’s dietary and physical activity components and the rationale behind them are previously published [17, 19]. The dietary component was based on ten recommendations designed to increase awareness of food choices, with advice to increase intake of water, vegetables and fruit and reduce snack food consumption. There was no calorie restriction or specific limitation of fats or carbohydrates. Counselling was performed twice, by phone, with a four to six week interval. Counsellors were either experienced clinical dieticians or graduate students in public health, trained and supervised by the NFFD team. The physical activity component consisted of access to twice-weekly exercise classes at a local gym facility, led by physical therapists or students in sports science, trained and quality-controlled by the NFFD team. Attendance was recorded. Participants were encouraged to engage in 30 min of moderate-intensity physical activity on three additional days per week. Lifestyle recommendations were reinforced with booklets, access to a NFFD internet site, and invitation to one cooking class and one evening meeting with information on the NFFD trial and the value of regular exercise and healthy diet in pregnancy.

Participants in the control group received routine prenatal care following Norwegian standard: eight prenatal appointments, including one second-trimester ultrasound examination, with additional care as needed, provided free of charge. Routine care includes a booklet with advice on prenatal nutrition, physical activity and recommendations for weight gain based on current Institute of Medicine (IOM) guidelines (normal-weight: 11.5–16 kg, overweight: 7–11.5 kg, obese: 5–9 kg) [20].

Measurements

The primary aims of the NFFD trial were to examine if intervention resulted in differences in GWG, birth weight of term infants, the proportion of term infants >4000 g, maternal fat percent at 36 gestational-weeks, and the incidence of operative deliveries. Maternal glucose levels at 30 gestational-weeks was a primary endpoint, while the proportion of women with elevated 2-h glucose tolerance tests and measurement of hormones related to glucose metabolism were secondary endpoints of the trial. Assessment of the subgroup of overweight/obese women was specified in the trial protocol.

Pre-pregnancy weight was self-reported. Participants were weighed at their healthcare clinic at inclusion, and at Sørlandet Hospital at 30 gestational-weeks (Tanita BC 418, Tokyo, Japan). Feasibility study participants reported their height; later participants were measured using a stadiometer (Seca Leicester, Hamburg, Germany). Pre-pregnancy BMI was calculated based on self-reported pre-pregnancy weight and measured height when available. Participants were weighed on admission to the delivery ward. If missing, last weight in the prenatal record was recorded with corresponding date. GWG at term was calculated for women delivering at ≥37 gestational-weeks with weight available within 2 weeks of admission.

Participants completed questionnaires at trial inclusion and at gestational-week 36, either electronically or in print. No questionnaires were completed at gestational-week 30. Diet was assessed by 43 food-frequency questions, analyzed using a pre-determined score (range 0–10, with higher score denoting healthier eating behavior). The score is previously described in detail, and has demonstrated acceptable test-retest reliability [19]. Physical activity was assessed with the International Physical Activity Questionnaire (IPAQ) short version, scored using IPAQ analysis algorithms. The IPAQ is validated in a Scandinavian population [21].

Prior to randomisation, fasting blood tests were assessed for evidence of pre-existing diabetes (defined as glucose ≥7.0 mmol/l) [22]. No participants were excluded on this basis. At gestational-week 30, plasma glucose was measured after overnight fast and again at 2-h after 75 g glucose load. All tests were performed at Sørlandet Hospital using a Cobas 6000 c501 chemistry analyzer (Roche Diagnostics). Glucose levels ≥7.0 mmol/l at fasting and/or ≥7.8 mmol/l at 2-h were classified as elevated, based on contemporary national [23] and WHO 2006 criteria [22], and participants and their primary care physicians were informed. Glucose at 2-h was missing for 12 participants (9 intervention, 3 control), primarily due to vomiting. Fasting serum samples were frozen and stored at −80 °C. Frozen samples were analyzed at Aker Hormone laboratory using a Modular E170 analyzer (Roche), batched to decrease interassay variation. Insulin was analyzed using non-competitive electrochemoluminescence immunoassay (Roche Diagnostics), with coefficient of variance of 4%. Leptin was analyzed using competitive radioimmunoassay (Millipore), with coefficient of variance of 7%. HOMA-IR was calculated as: (insulin(mU/l) x fasting glucose(mmol/l))/22.5. Leptin, insulin and HOMA-IR were missing for eight participants (3 intervention, 5 control), due to errors in freezing or transport. All missing values were considered missing completely at random. Three insulin and HOMA-IR values (1 intervention, 2 control) were excluded from analysis as outliers.

Sample size

We predicted a 20% prevalence of newborns with birth-weight > 4000 g in the control group based on 2005 statistics from the Norwegian birth registry [24], and determined empirically that a reduction to 10% in the intervention group would be clinically relevant. We calculated that 198 women were required in each study arm to demonstrate statistical significance with a power of 80%. We also expected a 10% incidence of GDM (based on 2-h glucose ≥7.8 mmol/l) [22, 23] in the control group, and determined that a reduction to 3% in the intervention group would be clinically significant. We calculated that we would have 80% power to detect a statistically significant difference between groups with 200 participants in each arm. To allow for participant drop-out and premature deliveries and to allow for analysis of subgroups, we planned to randomize 600 participants.

Statistics

Unadjusted comparison of intervention and control groups was performed using student t-test or chi-square test as appropriate. Difference between the randomized groups for continuous or binary variables was assessed using multiple linear or logistic regression models adjusted for age, education, income level and smoking at inclusion, pre-pregnancy BMI category and gestational age at measurement. Variables included in the adjusted analysis were chosen based on clinical relevance (pre-pregnancy BMI category and smoking) and/or measured differences between intervention and control group (gestational age at measurement) and/or measured differences between included and missing participants (age, education and income). Effect modification between randomized groups and patient characteristics on continuous outcomes was assessed by an interaction term in the multiple linear regression models. For binary outcomes, effect modification was assessed by the Breslow-Day test of homogeneity of odds ratios. No further adjustment for BMI category was performed when analysis was stratified according to pre-pregnancy BMI. P-values <0.05 were considered statistically significant. All tests were two-sided. We used SPSS for Windows version 21.0 for all statistical analyses.

Main findings

Overall, there was little beneficial effect of the NFFD lifestyle intervention on participant glucose levels, although there was a small but significant reduction of insulin and leptin levels. The intervention appeared to have divergent effect on glucose metabolism dependent on participants’ pre-pregnancy BMI status. For normal-weight women, the intervention had a weak positive effect on glucose metabolism, as evidenced by a trend (p < 0.1) toward reduced insulin and insulin-resistance and significantly lower leptin values, although there was no change in mean glucose levels or the proportion exceeding thresholds for GDM diagnosis. For overweight and obese women this picture was different, with a trend towards higher fasting glucose, but without any change in the other metabolic parameters. The effect of these trends on the prevalence of GDM varied depending on criteria used. When IADPSG thresholds were employed an increase in GDM was observed, whereas when using older WHO criteria there was no difference between groups.

Interpretation

There is little information available to date on the effect of prenatal combined lifestyle interventions on glucose, insulin and leptin levels, as most trials report the effect of intervention on GDM incidence rather than biochemical parameters. The effect of NFFD intervention on the glucose metabolism of normal-weight women is consistent with the findings of Vinter et al., who reported significantly lower insulin and HOMA-IR levels at gestational-week 28–30 following lifestyle intervention, but no significant differences in glucose levels or the incidence of GDM, albeit in an exclusively overweight/obese population [25]. Among non-pregnant individuals, exercise is well documented to improve glycemic control through improved insulin sensitivity [26]. It is plausible that a combination of exercise and diet can lessen insulin resistance, without being of sufficient intensity and/or duration to change plasma glucose levels. In the NFFD intervention group, women attended a median of 10 exercise classes (9 for overweight/obese participants) over a mean of 14 weeks between inclusion and testing, while the intended attendance was twice per week. Although we lack information about total physical activity level during this period of pregnancy, it is reasonable to suppose that greater compliance might have resulted in greater intervention effect. However, varying compliance is a hallmark of clinical practice, such that the present results likely reflect the effect of providing the NFFD intervention in a general population.

The temporal sequence of changes in the biochemical and clinical parameters following lifestyle intervention in pregnancy are not well known. In the current study, the reduction of leptin found in the total NFFD intervention group may indicate that adipokines are sensitive to interventions affecting energy metabolism. Leptin is essential in energy regulation and glucose metabolism [27, 28], and is secreted by both maternal adipocytes and placental trophoblasts during pregnancy [15]. Others have found that lower mid-pregnancy leptin levels are associated with reduced insulin resistance [28]. For the child, there is evidence that maternal mid-pregnancy leptin may be an indicator of fetal growth, with lower levels associated with reduced birth weight adjusted for gestational age [29]. Adipokines such as leptin may therefore be particularly sensitive to interventions affecting energy metabolism and may precede changes in glucose levels or clinical endpoints.

The divergent effect of lifestyle intervention on glucose metabolism based on pre-pregnancy BMI has, to our knowledge, not previously been reported. However, earlier trials have shown that women with higher pre-pregnancy BMI demonstrate resistance to intervention effect. Polley et al. reported that behavioural intervention reduced excessive GWG among normal-weight women, while overweight and obese women had a trend in the opposite direction [30]. Hui et al. [31] and Phelan et al. [32] both reported that a lifestyle intervention performed in a mixed population only reduced GWG among normal-weight women, and Phelan also reported a significant treatment-by-weight interaction for gestational hypertension [32]. The BMI-modified effects of lifestyle intervention have several possible explanations, which may be synergistic. Larger women may differ from normal-weight women in their understanding of and compliance with intervention. Additionally, overweight and obese women may enter trials with a metabolic state that is less sensitive to intervention than that of normal-weight women.

For overweight/obese women participating in the NFFD trial, assignment to exercise classes may have inadvertently discouraged further leisure-time physical activity, particularly among sedentary women. Exercise routines were designed to adjust to varied fitness levels, possibly allowing larger women to limit their exertion. Larger women may also have been intimidated by classes where normal-weight women were in the majority, perhaps explaining why overweight/obese women had lower attendance than normal-weight participants. In addition, NFFD dietary recommendations were not specifically designed to reduce GDM risk and contained no advice on restriction of calories, carbohydrates or fat.

Our finding of an increased proportion of elevated glucose levels among intervention participants compared to controls was unexpected, and its significance is unclear. Reassuringly, we found no increase in large newborns among intervention participants, an outcome that is closely associated with elevated maternal glucose. Several meta-analyses have concluded that combined lifestyle interventions in pregnancy have no effect on risk of GDM, with approximately half of the included trials demonstrating a non-significant increase in risk of GDM using varied criteria [12, 13]. The recently published RADIEL study is one of only two trials, to our knowledge, to report a significant reduction in the incidence of GDM following a combined lifestyle intervention [11, 33]. While results from individual trials must be assessed with caution, comparison may provide some insight. In contrast to the NFFD trial, RADIEL participants were included pre-gestation or in early pregnancy, which may be of critical importance. There is evidence that disposition for GDM is determined prior to pregnancy, with subclinical metabolic dysfunction before conception [14, 34]. RADIEL participants were also presumably highly motivated, as they were included in the trial based on their high-risk status. In contrast, including overweight/obese women with a normal-weight population, as was done in the NFFD trial, may have undermined the potentially greater importance of lifestyle changes for this more high-risk group.

Acknowledging that the effect of intervention may vary significantly among groups and individuals is important in planning future studies. Also important, in the current analysis, the effect of intervention on GDM risk was dependent on the thresholds used. This finding illustrates the difficulty of assessing trials that employ varying criteria for GDM diagnosis, and suggests that systematic review of individual patient data (IPD analysis) may be more suitable than standard meta-analysis for exploring the effect of prenatal interventions on glucose metabolism and gestational diabetes risk.

Strengths and limitations

The major strengths of the NFFD trial are its randomized, controlled design and the large size of the population studied, with relatively few missing values. Measured weight at inclusion and at the time of testing make it possible to accurately assess GWG and its association with metabolic findings. A major limitation of the current analysis is that, although examination of the subgroup of overweight/obese women was detailed in the trial protocol, <30% of participants were overweight/obese and the trial was not adequately powered to detect changes in smaller subgroups. Intervention effect of equivalent size may therefore be more easily detectable in the large subgroup of normal-weight women, as in the analysis of leptin. Another limitation is that due to individual randomization, women living in close proximity and attending the same clinic were often in different trial groups; it is possible that control participants were influenced by both intervention participants and clinic personnel who were informed of the purpose of the trial. While cluster randomization of clinics would have reduced such “contamination”, it would have introduced within-clinic correlations such as familial/genetic distribution, and likely required larger sample size in order to demonstrate intervention effect [35]. Due to practical and financial constraints, insulin resistance was assessed using HOMA-IR, which has shown significant correlation in pregnancy with the gold standard of the euglycemic insulin clamp [36], although an index incorporating multiple insulin measurements during glucose-testing might more accurately reflect skeletal muscle insulin resistance [37, 38]. This may be relevant in an intervention including exercise, which is proposed to lower insulin resistance at least in part through up-regulation of skeletal muscle glucose transporter protein GLUT4 [39]. Information regarding lifestyle at the time of glucose-testing is not available, limiting our assessment of the impact of diet and physical activity on biochemical results. In addition, lack of information on participants’ ethnic background and family history, both of which can affect glucose metabolism, may contribute to residual confounding. Also important, NFFD trial participants were predominantly white, European and highly educated, which may limit the external validity of results.