Effects of antenatal diet and physical activity on maternal and fetal outcomes: individual patient data meta-analysis and health economic evaluation

Criteria for including studies in individual participant data

We included studies that addressed the components of the structured question presented in Table 1.

Randomised trials, with or without clustering, that evaluated the effects of diet- and physical activity-based interventions in pregnancy on maternal, fetal and neonatal outcomes were eligible for inclusion. We included studies on normal, overweight and obese pregnant women. Interventions that addressed mainly diet or mainly physical activity and interventions adopting a mixed approach that combined the two, with or without behavioural modification techniques, were eligible. The control arms included women without any intervention or with routine antenatal care, as defined by local health-care practices. The primary outcomes were maternal weight gain in pregnancy, and composite maternal and composite fetal/neonatal events complications. Studies should have assessed both maternal weight gain and clinical outcomes.

Maternal weight gain was defined as the difference between the weights recorded (kg or lb) at first clinic visit and last weight measured before birth. If weight at first clinic visit was not available, we used pre-pregnancy weight.

The maternal composite outcome included gestational diabetes mellitus (GDM), pre-eclampsia (PE) or pregnancy-induced hypertension (PIH), preterm delivery and Caesarean section. The fetal and neonatal composite outcome comprised intrauterine death (IUD), small for gestational age (SGA), large for gestational age (LGA) and admission to the neonatal intensive care unit (NICU). The components of the composite outcome were identified by a two-round Delphi survey. The final scores of the components are provided in Appendix 1. The details of the development of the composite outcomes are published elsewhere. 19

We excluded studies published before 1990, animal studies and those that evaluated the effects of intervention only on non-clinical outcomes (behaviour change and consumption of particular food groups) or aimed to increase weight gain in pregnancy.

Criteria for including participants in individual participant data

We excluded underweight women (BMI of < 18.5 kg/m²) and women with multiple pregnancies.

Study selection

We undertook a two-stage study selection process. In the first stage, the abstracts of all citations were evaluated for their eligibility. In the second stage, we studied the identified studies in detail before their inclusion. Two independent reviewers (ER and EM) evaluated all papers. In case of disagreement, an opinion of the third reviewer (ST) was sought. We applied the eligibility criteria provided above for inclusion of studies.

Data items

We considered all recorded variables for inclusion when appropriate, including those not reported in the published studies. Data were extracted on the study and data set levels. At the study level, we collected information regarding study settings, intervention type, components, format and provider. At the participant level we requested information on individual characteristics including BMI, age, parity, ethnicity, socioeconomic status, pre-existing medical conditions, adherence to intervention and outcome data. The list of final variables collected during the project is available in Appendix 3.

Definition and standardisation of variables

Participant characteristics and other measurements were recorded in various different formats within the individual data sets. We chose the meta-data set format by including the variables that were most commonly reported. The standardisation process followed a predefined procedure (Figure 1).

FIGURE 1.

Flow diagram of standardisation of variables within the IPD sets. Interim DS refers to a temporary work-in-progress data set that will not be stored onto the database. DS, data set.

Data quality (individual participant data integrity)

Data sets included in the i-WIP analysis were expected to be clean on receipt from the original trial team. We performed range checks on the variables used during the analysis and produced summary tables. We focused on checking the randomisation ratio, baseline characteristics and the method of analysis in the IPD data set with the published information. Any major discrepancies were discussed with the trial team.

Handling of trials without individual participant data availability

We explored the potential for publication bias, and the possible impact of unavailable non-IPD data, in accordance with recent guidelines. 22 For each analysis containing 10 or more studies, the potential for publication bias was investigated through contour-enhanced funnel plots, and appropriate statistical tests for ‘small-study effects’ (i.e. the tendency for small studies to provide more positive findings than large ones).

For all studies in which IPD were not available, we extracted suitable aggregate data from the study publications. When possible, we then incorporated these aggregate data into the second stage of the two-step meta-analysis framework, to combine the IPD trials with the aggregate data from other trials for the outcome of GWG. This allowed us to examine whether or not conclusions (on summary results and potential publication bias) were changed by the inclusion of additional non-IPD trials. 14,23 If the inclusion of studies that did not provide IPD seemed to have an important statistical or clinical impact, we compared the characteristics of the studies with IPD with those without to see if there were any key differences (such as in their quality, follow-up length and statistical methods). This was achievable only when examining the overall treatment effect, as aggregate data for subgroup effects were rarely provided by the non-IPD studies. For individual maternal and fetal/neonatal complications, we compared meta-analyses findings of only aggregate published data with IPD.

Primary analyses of studies providing individual participant data

For each outcome (i.e. GWG, composite maternal, composite fetal/neonatal) separately, we performed a two-stage IPD meta-analysis to obtain summary estimates and 95% confidence intervals (CIs) for the intervention effects [mean difference (MD) or odds ratios (ORs)] and the interactions (subgroup effects) of interest. All participants were analysed according to the group they were randomised to. We used a two-stage random-effects meta-analysis approach, which allows for between-study heterogeneity in intervention effect (and interaction effect). In any two-stage meta-analysis, the clustering of participants within trials is accounted for by analysing each trial separately in the first stage. Women with confirmed glucose intolerance or hypertensive disorder at baseline, as defined by the primary authors, were excluded in the analysis of composite adverse pregnancy outcomes.

First stage of individual participant data meta-analysis

Continuous outcome measures were checked for normality and log transformed if applicable. Variables (covariates) were kept as continuous as well as defining subgroups for BMI and maternal age. All analyses were performed on complete cases, that is individuals who provided the outcome and (if relevant for the analysis) baseline adjustment factors. When analysing cluster randomised trials, we included a random intercept for the unit of randomisation to account for clustering. For the continuous outcome of weight gain, we used analysis of covariance in each trial to regress the final weight value against the intervention, while adjusting for baseline weight. For the binary outcome of adverse fetal/neonatal or maternal outcome, the binomial nature was modelled using a logistic regression in each trial separately, with intervention as a covariate. Stratification or minimisation factors used in the randomisation of each study were not adjusted for in any analyses. The Sweeting et al. 25 approach was applied to include studies into the analysis of composite outcomes that had no information on outcome for one treatment group. This was only done for the primary analysis without interaction terms.

When examining intervention effect modifiers, we extended the models to include interaction terms between participant-level covariates and the intervention. For the interactions, continuous covariates (BMI and age) were analysed on continuous scales and as clinically defined categorical values. In addition, effects were presented within the subgroups defined by the interactions.

All primary analyses were performed on the combined intervention and any multiple treatment arms were combined into one intervention arm. For the secondary analysis of individual intervention types, multiple treatment arms were combined if they belonged to the same type, for example brochure arm and active counselling were grouped as mixed-approach intervention) or analysed separately if the treatment arms were categorised as different types (e.g. exercise and exercise plus dietary counselling).

Second stage of individual participant data meta-analysis

We pooled effect estimates (e.g. relating to treatment effects or treatment–covariate interactions) using a random-effects model using restricted maximum likelihood (REML) to produce a summary effect estimate for the mean (or average) effect across studies. The Knapp–Hartung correction was applied when deriving 95% CIs for each summary effect, to account for the uncertainty of the estimate of between-study heterogeneity (τ²). Forest plots were generated to display the study-specific and pooled results.

Heterogeneity was summarised using the I² statistic and the estimated between-study variance (τ²) was obtained using REML. To reveal the impact of heterogeneity more clearly, we also calculated approximate 95% prediction intervals (PIs) for the intervention (or interaction) effect in a new study using the formula suggested by Higgins et al. 26

Sensitivity analyses

Small-study effects (and the potential for publication bias) were investigated by using contour-enhanced funnel plots and tests for asymmetry (using either the Egger’s test for continuous outcomes or Peter’s test for binary outcomes). In order to examine whether or not there may be availability bias in the obtained IPD, we compared summary results when including non-IPD studies with those in our IPD studies. When possible, we then incorporated this aggregate data into the second stage of the two-step meta-analysis framework (see below), to combine the IPD trials with the aggregate data from other trials, to ascertain if conclusions were robust.

We investigated the following sources of bias for all or a subset of the primary outcomes by performing the following sensitivity analyses.

Secondary analyses

All secondary analyses were performed only on participants in control arms to exclude the effect of treatment.

Overall effect

Overall, diet- and physical activity-based interventions (33 studies, 9320 women) reduced GWG by an average of –0.70 kg (95% CI –0.92 to –0.48 kg; I² = 14.1%) (Figure 4), after accounting for baseline weight and clustering effect.

FIGURE 4.

Effects of diet- and physical activity-based interventions on GWG (kg). Trt, treatment.

All three individual interventions (diet, physical activity and mixed) had a similar effect on reducing GWG by an average of 0.7 kg (Table 7).

Differential effect of the intervention on gestational weight gain in various subgroups

Thirty-one studies (9285 women) provided data to evaluate the differential effect of interventions on GWG for women with varied BMI at booking. There was no significant treatment–covariate interaction for baseline BMI (–0.02 kg change in effect per 1 kg/m² increase in BMI, 95% CI –0.08 to 0.04 kg change). We did not observe any interaction effect for other effect modifiers such as age (–0.03 kg change in effect per 1-year increase in age, 95% CI –0.08 to 0.02 kg), parity (0.10 kg change in effect for multiparous vs. nulliparous, 95% CI –0.39 to 0.60 kg change), ethnicity (0.05 kg change in effect for non-Caucasian vs. Caucasian, 95% CI –1.27 to 1.37 kg change) and underlying medical condition (1.51 kg change in effect for women with at least one condition vs. none, 95% CI –2.01 to 5.02 kg) (Table 8). The findings were consistent when we analysed the continuous covariates as dichotomised measures.

Overall effect

Diet- and physical activity-based interventions (24 studies, 8852 women) reduced the odds of adverse maternal outcomes by 10% (summary OR 0.90, 95% CI 0.79 to 1.03; I² = 26.7%). The effect was not statistically significant at the 5% level (Figure 5).

FIGURE 5.

Effects of diet- and physical activity-based interventions on composite maternal outcome. Trt, treatment.

The effects on composite maternal outcomes were evaluated in two-thirds of participants in studies of mixed interventions, compared with 4% of participants in studies of diet-based interventions. The effects of physical activity, diet and mixed approaches were not statistically significant (Table 9).

Sensitivity analysis by excluding studies rated as having a high risk of bias (summary OR 0.91, 95% CI 0.77 to 1.08) and women non-adherent to the intervention (summary OR 0.92, 95% CI 0.80 to 1.06) did not affect the findings. The results of all sensitivity analyses are provided in Appendix 8.

Differential effects of the intervention for composite maternal outcome in various subgroups

Twenty-four studies (8848 women) contributed IPD to assess the differential effects of interventions on the composite maternal outcome according to maternal BMI category. There was no significant treatment–covariate interaction for baseline BMI (no change in effect for every 1-kg/m² increase in BMI, OR 95% CI 0.98 to 1.02). The effects of the interventions were not significantly modified by other relevant covariates such as age (1% increase in effect for every 1-year increase in age, OR 95% CI 0.99 to 1.03), parity (3% increase in effect for multiparity vs. nulliparity, OR 95% CI 0.75 to 1.39), ethnicity (7% decrease in effect for non-Caucasian vs. Caucasian, OR 95% CI 0.63 to 1.37) and underlying medical condition (44% increase in effect for women with none vs. at least one condition, OR 95% CI 0.15 to 13.74). The findings were consistent when continuous covariates were further analysed as categorical values (Table 11).

Overall effect

Diet- and physical activity-based interventions (18 studies, 7981 women) did not reduce the odds of the composite adverse fetal/neonatal outcome (summary OR 0.94, 95% CI 0.83 to 1.08) (Figure 6) after adjusting for clustering.

FIGURE 6.

Effect of diet- and physical activity-based interventions on composite fetal/neonatal outcome.

Two studies (346 women) evaluated diet-based interventions, five (1274 women) assessed physical activity-based interventions and 12 (6494 women) studied mixed interventions on composite fetal/neonatal outcomes. None of the three interventions reduced composite adverse fetal/neonatal outcome (Table 12).

Differential effect of the intervention for composite fetal/neonatal outcome in various subgroups

Eighteen studies (7981 women) provided IPD to assess the differential effects of interventions by maternal baseline BMI on composite fetal/neonatal outcome. There was a 2% decrease in the treatment effect of borderline significance for every 1 kg/m² increase in booking BMI for composite fetal/neonatal outcomes (OR 0.98, 95% CI 0.95 to 1.00). There was no treatment–covariate interaction for other variables, such as maternal age of < 20 years (OR 1.05, 95% CI 0.33 to 3.35), ethnicity (12% decrease in effect for non-Caucasian vs. Caucasian, 95% CI 0.75 to 1.68), parity (6% reduction in effect for multiparous vs. nulliparous, 95% CI 0.64 to 1.47), baseline medical conditions (42% increase in effect for women with at least one medical condition vs. none, 95% CI 0.00 to 2440.15) (Table 13).

None of the sensitivity analyses performed showed a significant impact on the intervention effect (see Appendix 8).

Model structure

The appropriate model for this study was a decision tree because of the short-term nature of the decision problem. 98 The model was developed using TreeAge Pro 2014 software (TreeAge Software, Inc., Williamstown, MA, USA). The structure was informed by the objectives of the study, the pathways indicated by the data and trials included in the IPD meta-analysis, clinical input, NICE guidelines on the management of women in pregnancy10,99,100 and the approaches adopted in previously published model-based economic evaluations in relevant clinical areas. 101–105 For completeness, the model included all the potential pathways that could be followed by the women. Women entered the model at the point of randomisation to receive the intervention or care as usual (control). All women were assumed to follow one of six clinical pathways based on whether or not they developed pregnancy-related complications or experienced miscarriage or maternal death. These pathways were (1) PE, (2) GDM, (3) PIH, (4) no complication, (5) second-trimester miscarriage and (6) maternal death (Figure 10). Each pathway included appropriate maternal and fetal outcomes as detailed below.

FIGURE 10.

Patient pathways incorporated in the model.

For the base case, outcomes were considered until the point of discharge from hospital. Once women entered the model, it was assumed that they followed one of the clinical pathways defined in the model, based on whether or not they developed a pregnancy-related condition/complication. Complications were defined in accordance with the definitions used in studies included in the IPD meta-analysis. To illustrate the approach used for each patient pathway, a subset of the model is presented for the PE pathway (Figure 11). Women who developed more than one complication were allocated to the most resource-intensive pathway based on an analysis of NICE clinical pathways and clinical opinion. 106 The intensity of pathways was defined as follows (in decreasing order of intensity): (1) PE, (2) GDM99 and (3) PIH. 100 For the purposes of the model it was assumed that once women developed a complication they were treated in accordance with NICE guidelines and incurred associated antenatal health-care costs. Women who did not develop any of the specified conditions were assumed to receive routine antenatal care only. It was assumed that routine antenatal care would be received by all women, irrespective of whether or not they developed a pregnancy-related condition. 107 As the purpose of economic evaluation is to examine the differences in costs and outcomes between alternative courses of action,95 the costs of routine antenatal care were not included in the model as they would be identical for each arm.

FIGURE 11.

Detail for the PE pathway. C-section, Caesarean section.

For all clinical pathways, women could either experience a preterm delivery or a delivery at term; preterm delivery was defined as delivery before 37 gestational weeks. 108 Three types of delivery were included in the model: Caesarean section, assisted delivery and vaginal delivery. The outcome of the delivery was either a stillbirth (or IUD) or live baby. Stillbirth was defined as a baby born with no signs of life after 24 weeks of completed pregnancy and IUD as a baby with no signs of life in utero. 109 Women who experienced stillbirth and IUD were assumed to have received appropriate antenatal care for any condition they were recorded as developing during the trial. Additional costs associated with investigations and counselling were included in the total costs for these women. 110

The model also included pathways for second-trimester miscarriage and maternal death in order to reflect all possible pathways for the women. Second-trimester miscarriage was defined as the spontaneous loss of pregnancy after the 14th week of pregnancy and before the 24 completed weeks. 111 The risk of second-trimester miscarriage and maternal death was based on secondary sources and applied to both arms equally to preserve the face validity of the model. 112 As the purpose of the economic evaluation was to examine the differences in costs and outcomes between the intervention and control arms, costs were not included for second-trimester miscarriage and maternal death, as they would be identical for each arm.

Model assumptions

To carry out the model-based analysis some further assumptions were required. These are presented below.

Clinical data

The primary focus of the economic evaluation was the effect of the intervention on adverse maternal and fetal outcomes. The maternal outcomes included the development of three pregnancy-related conditions (PE, GDM and PIH), as well as outcomes relating to the timing and mode of delivery (preterm delivery and Caesarean section). Fetal outcomes included IUD, SGA, LGA and admission to the NICU.

Cost data

The cost components of the decision model were parameterised with data from NHS reference costs 2013/1497 and other secondary sources. Costs were calculated in 2013–14 GBP. Costs from secondary sources were inflated to 2013–14 prices using the hospital and community health services pay and prices index. 96 When necessary, costs were converted to GBP using historical annual average rates124 and then inflated to 2013–14 prices.

The estimate of the cost of the weight management intervention was based on the results of a systematic review of economic evaluations of weight management interventions in pregnancy that was conducted for this report (see Appendix 10). The review identified 11 studies that were concerned with the cost-effectiveness of weight management interventions during pregnancy. 68,101,125–133 Four of these involved women who already had DM or GDM and hence the intervention costs included additional monitoring and medication. 125,130,131,133 Two did not report a cost for an intervention. 101,126 Data on the cost of the intervention were thus extracted for five studies68,127–129,132 and the median value was used to estimate the cost of the intervention in the model (Table 24).

Estimates of antenatal and postnatal care costs were drawn from systematic reviews of the literature (as described in the previous section on assumptions). Delivery costs were based on national average costs in the UK. 97

Analysis

The decision-analytic model was constructed to compare the cost-effectiveness of diet- and physical activity-based interventions in pregnancy with usual care. Two separate economic analyses were conducted. The main analysis compared costs and outcomes for a hypothetical cohort of 10,000 pregnant women, based on the results of the IPD meta-analysis for all women. The secondary analysis compared costs and outcomes for three subgroups of women based on their pre-pregnancy BMI classification. Women were classified as normal weight (pre-pregnancy BMI of < 25 kg/m²), overweight (pre-pregnancy BMI 25–29.9 kg/m²) or obese (pre-pregnancy BMI of ≥ 30 kg/m²). This allowed an exploration of whether or not a weight management intervention in selective subgroups of women is a more cost-effective strategy than care as usual.

For both the primary and secondary analyses, the relative cost-effectiveness of the intervention was evaluated using effect size estimates from the IPD meta-analysis. An incremental approach was adopted, with a focus on the additional costs and benefits associated with a move from care as usual to diet and lifestyle interventions to manage weight gain in pregnancy. The results were reported in terms of an incremental cost-effectiveness ratio (ICER) of cost per unit benefit gained, measured in clinical outcomes. The IPD meta-analysis was based on a composite outcome, which included maternal and fetal complications. An economic evaluation based on such a composite outcome would not be meaningful, as the full extent of the cost implications based on such an outcome would be lost. The economic analysis therefore examined cost-effectiveness for each of the primary outcomes separately; for example, results are presented in terms of cost per case of PE avoided, cost per case of GDM avoided, etc. Results are also presented in terms of cost per major outcome avoided. The major outcome was predefined to include one or more of the fetal and maternal outcomes included in the composite measure that was used in the IPD meta-analysis.

The analysis was conducted from the perspective of the health service (NHS) and only direct health service costs were included. The time horizon adopted for both the primary and secondary analyses was from the start of pregnancy until the mother and infant were discharged from hospital following the birth. This was considered appropriate to reflect all key differences in terms of costs and benefits for the options compared, given the time horizon adopted in the trials included in the IPD meta-analysis. The effect of considering a longer time horizon was explored as part of the sensitivity analysis.

Sensitivity analysis

Deterministic and probabilistic sensitivity analyses were conducted to explore the effects of the inherent uncertainty in the parameter estimates on the results produced by the model. 98 Deterministic sensitivity analysis involves varying one or more parameters while keeping the others at their baseline value. Such analysis can help to identify which model inputs are important in leading to a particular decision from the model, and can help to identify threshold values. A probabilistic sensitivity analysis (PSA) was also undertaken to allow uncertainty to be represented more comprehensively. A PSA involves varying all parameters simultaneously, and multiple sets of parameter values are sampled from defined probability distributions. 134 Monte Carlo simulation was used to sample from these distributions to allow the effect of parameter uncertainty to be evaluated. This involved 1000 repeated random draws from the distributions to indicate how variation in the model parameters would affect the results and hence illustrate the decision uncertainty. 135 Beta distributions were used for binomial data, log-normal distributions for ORs and gamma distributions for costs. 98 Ideally, when there are more than two possibilities at a chance node, a Dirichlet distribution would be applied. But to populate a Dirichlet distribution it is necessary that all included studies have reported data for all the branches from the appropriate chance node. This could be done if all data sources reported results for all possible branches. However, in this case, some studies did not report numbers for all branches in the model. For example, the number of LGA infants born to mothers with GDM was reported in 27 studies, but the numbers of SGA infants was reported in only 26 studies. Therefore, it is feasible to assign separate beta distributions only for the probabilities of each of these two outcomes. For consistency, we considered it appropriate to apply beta distributions throughout. The potential limitation of applying a beta distribution is that some correlation between parameters could be lost. However, in this study, this is unlikely to have had an impact on results.

Using the net monetary benefit for each of the 1000 simulations, the proportion of times the intervention had the highest net monetary benefit was calculated for a range of threshold values for the maximum willingness to pay for a major outcome averted. These values were summarised graphically using a cost-effectiveness acceptability curve (CEAC) to show the uncertainty surrounding the cost-effectiveness of the intervention, for a range of thresholds for cost-effectiveness. A value of information analysis was also conducted to estimate the expected costs of uncertainty. The expected cost of uncertainty is calculated by estimating the probability of making a wrong decision based on existing evidence and the consequences of this wrong decision. The expected value of perfect information (EVPI) estimates the difference between the expected value of the decision made with perfect information and the decision made on the basis of existing evidence. 98 EVPI was calculated based on the methods described in Claxton and Posnett. 136

Primary analysis

The results of the primary analysis are reported in Tables 25 and 26. Care as usual was the least costly arm, with the average cost estimated at £3242 (excluding routine antenatal care). However, the intervention arm was only slightly more expensive, with average costs estimated at £3390 (excluding routine antenatal care). In the base-case model (using point estimates) there was a reduction in pregnancy-related complications in the intervention arm (see Table 26). Overall, there were 55 fewer women who experienced any major outcome in the intervention arm, five fewer cases of PE, 113 fewer cases of GDM and 29 fewer women who developed PIH. No reduction was found in other outcomes such as preterm delivery, Caesarean section, IUD, LGA and NICU admission for the intervention arm. The ICERs indicated that there was an additional cost of around £306,000 for each case of PE avoided, £13,000 for each case of GDM avoided and £27,000 for each woman with a major outcome avoided. This means that, if for example, the outcome of interest is the reduction in cases of PE, an additional £306,000 is required for each additional woman who avoids the development of PE compared with usual care. These results taken in isolation would suggest that the intervention would be preferred at any willingness to pay for a case of PE avoided above £306,000.

These results need to be viewed together with those from the PSA (Tables 27 and 28). The PSA results show the modelled uncertainty in the cost and effectiveness between the intervention arm and the care-as-usual arm, from 1000 Monte Carlo simulations. These results show that it is uncertain whether the intervention is more effective than usual care or less effective (with respect to all the outcome measures) and whether the intervention is more costly or less expensive than the alternative. Hence, the most reasonable interpretation is that there is no significant difference between the intervention and control arm results for either cost or clinical effectiveness. This is further illustrated by Figure 12. The graph plots the result of each simulation on the cost-effectiveness plane, which gives information about the joint density of differences in cost and effectiveness between the two model arms. It is evident that the data in the scatterplot go into all quadrants of the incremental cost-effectiveness plane and, therefore, it should be assumed that there is no significant difference between the intervention and control arms for either cost or clinical effectiveness. This was the case for all of the outcome measures (for more details see Appendix 11). The CEAC (Figure 13) shows that the probability that the intervention is cost-effective increases as the willingness to pay for a major outcome averted increases. Thus, if the maximum willingness to pay for a major outcome averted for all pregnant women was £30,000, then the probability that the intervention was cost-effective would be 0.55.

FIGURE 12.

Incremental cost-effectiveness scatterplot of intervention compared with care as usual for pregnant women: major outcome averted. The mean of the distribution is highlighted.

FIGURE 13.

Incremental CEAC of intervention for pregnant women: major outcome averted.

The analysis of the EVPI demonstrates that at a willingness-to-pay threshold of £30,000 to avert a major outcome, the expected value of resolving the uncertainty around the decision is £142 per patient to whom the decision would apply (Figure 14). Thus, for a cohort of 10,000 women, the EVPI would be £1,420,000.

FIGURE 14.

Expected value of perfect information: major outcome averted.

Secondary analysis

Obese women

The results of the secondary analysis for obese women are reported in Tables 29 and 30. Care as usual was the least costly arm for obese women, with an average cost estimated at £3428 (excluding routine antenatal care). However, the intervention arm was only slightly more expensive, with average costs estimated at £3613 (excluding routine antenatal care). In the base-case model (using point estimates) there were 21 fewer obese women who experienced any major outcome in the intervention arm and 125 fewer cases of GDM. No reduction was found in other outcomes such as the development of PE, PIH and rates of preterm delivery, Caesarean section, IUD, LGA and NICU admission. The ICERs indicated that there was an additional cost of around £88,000 for each major outcome avoided and £15,000 for each case of GDM avoided. The PSA results show the modelled uncertainty in the cost and clinical effectiveness between the intervention arm and the care-as-usual arm and demonstrate that it is uncertain whether the intervention is more or less effective than usual care and whether the intervention is more or less costly than the alternative (Tables 31 and 32). These results suggest that there is no significant difference between the intervention and the control arm results for either cost or clinical effectiveness for obese women. This is further illustrated by Figure 15, which plots the result of each simulation on the cost-effectiveness plane and shows that there is no significant difference between the intervention and control arms for either cost or clinical effectiveness for all outcome measures (for more details see Appendix 11). The CEAC (Figure 16) shows if the maximum willingness to pay for a major outcome averted for obese pregnant women was £120,000, then the probability that the intervention was cost-effective would be 0.52.

TABLE 29 - Average costs (£) for intervention group compared with care as usual: obese women

As it was assumed that routine antenatal care was received by all women this was not included in the total costs, as there would be no difference in these costs between model arms.

Group allocation	Cost
	Mean (point estimate)	Difference
Intervention	3613	185
No interventiona	3428

TABLE 30 - Results for analysis for a cohort of 10,000 obese women

Cost per complication avoided is rounded to the nearest £1000.

Intervention is less clinically effective than care as usual/no difference in clinical effectiveness.

Outcome	Number of women experiencing complications (point estimate)		Number of complications avoided per 10,000 women: outcome averted (point estimate)	Cost (£) per complication avoideda
	Intervention	No intervention
Any major outcome	7248	7269	21	88,000
PE	550	530	–20	b
GDM	1512	1637	125	15,000
PIH	562	514	–48	b
Preterm delivery	1021	1018	–3	b
Caesarean section	3514	3514	0	b
IUD	25	25	0	b
LGA	1587	1589	3	712,000
SGA	905	902	–3	b
NICU admission	1192	1193	1	2,057,000

TABLE 31 - Results of PSA: average costs (£) for intervention group compared with care as usual for obese womena

Note that the mean results in the PSA are not the same as the point estimates used in the base-case analysis because of the use of log-normal distributions which create a non-linearity in the model.

As it was assumed that routine antenatal care was received by all women this was not included in the total costs, as there would be no difference in these costs between model arms.

Group allocation	Cost (95% CI)
	Mean	Difference
Intervention	3675 (1794 to 6321)	191 (–247 to 809)
No interventionb	3484 (1608 to 6129)

TABLE 32 - Results of PSA: analysis for a cohort of 10,000 obese womena

Note that the mean results in the PSA are not the same as the point estimates used in the base-case analysis because of the use of log-normal distributions which create a non-linearity in the model.

Outcome	Number of women experiencing complications, mean (95% CI)		Number of complications avoided per 10,000 women (95% CI)
	Intervention	No intervention
Any major outcome	7252 (6992 to 7524)	7268 (7108 to 7444)	16 (–194 to 197)
PE	553 (396 to 750)	529 (437 to 630)	–24 (–189 to 108)
GDM	1517 (1164 to 1940)	1636 (1482 to 1786)	119 (–262 to 445)
PIH	564 (403 to 778)	513 (420 to 615)	–51 (–223 to 95)
Preterm delivery	1021 (912 to 1130)	1018 (911 to 1125)	–3 (–17 to 10)
Caesarean section	3518 (3329 to 3700)	3518 (3336 to 3699)	1 (–19 to 21)
IUD	25 (10 to 50)	25 (10 to 49)	0 (–1 to 1)
LGA	1587 (1463 to 1716)	1590 (1463 to 1720)	3 (–17 to 25)
SGA	906 (820 to 993)	903 (822 to 985)	–4 (–27 to 16)
NICU admission	1192 (1070 to 1316)	1193 (1076 to 1319)	1 (–15 to 16)

FIGURE 15.

Incremental cost-effectiveness scatterplot of intervention compared with care as usual for obese pregnant women: major outcome averted. The mean of the distribution is highlighted.

FIGURE 16.

Incremental CEAC of intervention for obese pregnant women: major outcome averted.

Overweight women

The results of the secondary analysis for overweight women are reported in Tables 33 and 34. As for obese women, care as usual had the lowest cost, with the average cost estimated at £3114 (excluding routine antenatal care). The intervention arm was slightly more expensive, with average costs estimated at £3326 (excluding routine antenatal care). In the base-case model (using point estimates) there were 115 fewer overweight women who developed PIH (see Table 34). However, for all other clinical outcomes, the intervention did not lead to a reduction in adverse outcomes. The PSA results allow an assessment of the uncertainty in the estimates of cost and clinical effectiveness between the intervention arm and the care-as-usual arm (Tables 35 and 36). It is evident that it is uncertain whether the intervention is more or less effective than usual care and whether the costs associated with the intervention arm are greater or lower than the alternative. Hence, there is no significant difference between the intervention and control arm results for either cost or effectiveness for overweight women. This is further illustrated by Figure 17, which plots the result of each simulation on the cost-effectiveness plane and shows that no significant differences were found between the intervention arm and control arm for either cost or effectiveness (for more details see Appendix 11). The CEAC (Figure 18) shows that, even if the maximum willingness to pay for a major outcome averted for overweight pregnant women was £500,000, then the probability that the intervention is cost-effective would be < 0.5.

TABLE 33 - Average costs (£) for intervention group compared with care as usual: overweight women

As it was assumed that routine antenatal care was received by all women this was not included in the total costs, as there would be no difference in these costs between model arms.

Group allocation	Cost
	Total (point estimate)	Difference
Intervention	3326	211
No interventiona	3114

TABLE 34 - Results for analysis for a cohort of 10,000 overweight women

Intervention is less clinically effective than care as usual/no difference in clinical effectiveness.

Outcome	Number of women experiencing complications (point estimate)		Number of complications avoided per 10,000 women (point estimate)	Cost (£) per complication avoided
	Intervention	No intervention
Any major outcome	6915	6903	–11	a
PE	459	439	–21	a
GDM	996	871	–126	a
PIH	293	407	115	18,000
Preterm delivery	1039	1034	–5	a
Caesarean section	3531	3521	–10	a
IUD	25	25	0	a
LGA	1618	1614	–4	a
SGA	879	881	1	1,621,000
NICU admission	1206	1198	–8	a

TABLE 35 - Results of PSA: average costs (£) for intervention group compared with care as usual for overweight womena

Note that the mean results in the PSA are not the same as the point estimates used in the base-case analysis because of the use of log-normal distributions which create a non-linearity in the model.

As it was assumed that routine antenatal care was received by all women this was not included in the total costs, as there would be no difference in these costs between model arms.

Group allocation	Cost (95% CI)
	Mean	Difference
Intervention	3399 (1546 to 6419)	231 (–279 to 980)
No interventionb	3169 (1334 to 6002)

TABLE 36 - Results of PSA: analysis for a cohort of 10,000 overweight womena

Note that the mean results in the PSA are not the same as the point estimates used in the base-case analysis because of the use of log-normal distributions which create a non-linearity in the model.

Outcome	Number of women experiencing complications, mean (95% CI)		Number of complications avoided per 10,000 women (95% CI)
	Intervention	No intervention
Any major outcome	6930 (6613 to 7318)	6902 (6720 to 7095)	–28 (–351 to 206)
PE	472 (248 to 800)	438 (324 to 569)	–34 (–325 to 158)
GDM	1019 (581 to 1624)	871 (712 to 1031)	–148 (–703 to 257)
PIH	299 (159 to 530)	405 (299 to 530)	107 (–72 to 265)
Preterm delivery	1039 (921 to 1159)	1033 (915 to 1150)	–6 (–31 to 12)
Caesarean section	3535 (3328 to 3729)	3525 (3328 to 3714)	–10 (–43 to 14)
IUD	25 (9 to 52)	26 (9 to 53)	0 (–2 to 2)
LGA	1616 (1482 to 1759)	1614 (1475 to 1754)	–2 (–25 to 32)
SGA	882 (785 to 974)	882 (796 to 970)	0 (–40 to 27)
NICU admission	1206 (1073 to 1341)	1197 (1068 to 1332)	–9 (–33 to 9)

FIGURE 17.

Incremental cost-effectiveness scatterplot of intervention compared with care as usual for overweight pregnant women: major outcome averted. The mean of the distribution is highlighted.

FIGURE 18.

Incremental CEAC of intervention for overweight pregnant women: major outcome averted.

Normal weight women

The results of the secondary analysis for normal weight women are reported in Tables 37 and 38. The intervention arm had the lowest costs, with an average cost estimated at £3056 (excluding routine antenatal care). This was because the additional costs associated with delivering the intervention were outweighed by savings associated with reduced health-care use overall. However, the care-as-usual arm was only slightly more expensive, with average costs estimated at £3063 (excluding routine antenatal care). In the base-case model (using point estimates) there were 108 fewer women who experienced any major outcome in the intervention arm, 231 fewer cases of PE and 53 fewer women who developed GDM. There were also reductions in the rate of preterm delivery, Caesarean section, IUD, SGA, and NICU admission.

TABLE 37 - Average costs (£) for intervention group compared with care as usual: women of normal weight

As it was assumed that routine antenatal care was received by all women this was not included in the total costs, as there would be no difference in these costs between intervention arms.

Group allocation	Cost
	Mean (point estimate)	Difference
Intervention	3056	–7
No interventiona	3063

TABLE 38 - Results of analysis for a cohort of 10,000 normal weight women

As it was assumed that routine antenatal care was received by all women this was not included in the total costs, as there would be no difference in costs between intervention arms.

Intervention dominates usual care.

Intervention is less clinically effective than care as usual/no clinical difference in effectiveness.

Outcome	Number of women experiencing complications (point estimate)		Number of complications avoided per 10,000 women (point estimate)	Cost (£) per complication avoided
	Intervention	No interventiona
Any major outcome	6677	6785	108	b
PE	275	506	231	b
GDM	573	626	53	b
PIH	274	274	0	b
Preterm delivery	1029	1056	28	b
Caesarean section	3518	3541	24	b
IUD	25	25	0	b
LGA	1639	1621	–18	c
SGA	854	880	26	b
NICU admission	1197	1213	17	b

The results suggested that, for most clinical outcome measures reported, the intervention was less costly and more effective than care as usual. These results need to be considered alongside the results of the PSA (Tables 39 and 40). The PSA results show the modelled uncertainty in terms of the cost and the effectiveness for the intervention arm and the care-as-usual arms. The findings suggest that it is uncertain whether the intervention is more or less effective than usual care and whether the intervention costs are higher or lower than the costs associated with usual care. Hence, there is no significant difference between the intervention and the control arm results for either cost or effectiveness for women of normal weight. This is further illustrated by Figure 19, which plots the result of each simulation on the cost-effectiveness plane and shows that, although there is a trend towards lower cost and increased effectiveness for the intervention arm, there is no statistically significant difference between the intervention and the control arms for either cost or effectiveness (see more details in Appendix 11). The CEAC (Figure 20) shows that as the willingness to pay for a major outcome averted for normal weight women increases, the probability that the intervention is cost-effective also increases. Thus, if the threshold for cost-effectiveness was £30,000 to avert a major outcome for normal weight women, then the probability that the intervention was cost-effective would be 0.67.

TABLE 39 - Results of PSA: average costs (£) for intervention group compared with care as usual for normal weight womena

Note that the mean results in the PSA are not the same as the point estimates used in the base-case analysis because of the use of log-normal distributions which create a non-linearity in the model.

As it was assumed that routine antenatal care was received by all women this was not included in the total costs, as there would be no difference in costs between intervention arms.

Outcome	Cost (95% CI)
	Mean	Difference
Intervention	3140 (1244 to 6264)	22 (–563 to 741)
No interventionb	3118 (1265 to 6041)

TABLE 40 - Results of PSA: analysis for a cohort of 10,000 normal weight womena

Note that the mean results in the PSA are not the same as the point estimates used in the base-case analysis because of the use of log-normal distributions which create a non-linearity in the model.

Outcome	Number of women experiencing complications, mean (95% CI)		Number of complications avoided per 10,000 women (95% CI)
	Intervention	No intervention
Any major outcome	6704 (6403 to 7068)	6784 (6599 to 6972)	80 (–240 to 286)
PE	290 (121 to 580)	505 (376 to 662)	214 (–44 to 394)
GDM	603 (275 to 1125)	627 (504 to 757)	24 (–466 to 340)
PIH	299 (96 to 725)	272 (186 to 379)	–27 (–421 to 195)
Preterm delivery	1028 (908 to 1155)	1056 (933 to 1179)	28 (4 to 59)
Caesarean section	3520 (3302 to 3722)	3545 (3345 to 3736)	24 (–5 to 62)
IUD	26 (8 to 54)	26 (9 to 53)	0 (–2 to 2)
LGA	1636 (1493 to 1780)	1621 (1481 to 1760)	–15 (–43 to 24)
SGA	858 (761 to 954)	882 (793 to 972)	23 (–13 to 54)
NICU admission	1195 (1055 to 1337)	1212 (1080 to 1352)	17 (–6 to 50)

FIGURE 19.

Incremental cost-effectiveness scatterplot of intervention compared with care as usual for normal weight pregnant women: major outcome averted. The mean of the distribution is highlighted.

FIGURE 20.

Incremental CEAC of intervention for normal weight pregnant women: major outcome averted.

Deterministic sensitivity analysis

As demonstrated in Table 41, the results of the deterministic sensitivity analysis were as follows: (1) varying the cost of the intervention affected the average cost of the intervention arm and the cost per major outcome avoided, with this rising to £170,000 per major outcome avoided for the highest intervention cost; (2) improving the effect of the intervention meant that the intervention arm was more cost-effective than the control arm, the model was particularly sensitive to the estimate of the effect of the intervention on the odds of developing of PE; (3) varying the timing and mode of delivery did not change the overall result for the development of pregnancy-related conditions; (4) increasing and decreasing the costs for each condition had some impact on the overall results; (5) varying the costs associated with various types of delivery had a negligible effect on the overall results; and (6) increasing the costs of IUD had a negligible impact on overall results. As expected, the multivariate analyses demonstrated that using the highest estimates of effectiveness meant that the intervention arm dominated the control arm (it was less costly and more effective), as the additional costs of the intervention were outweighed by lower health-care resource use.

Principal findings

The results of this analysis suggest that there is no evidence that mixed interventions in pregnancy are cost-effective compared with care as usual. Although the primary base-case analysis indicated a small reduction in pregnancy-related complications, the PSA demonstrated that there is no evidence of a significant difference between the intervention and the control arms for either cost or clinical effectiveness. Similarly, the results of the secondary analysis suggested that for obese, overweight and normal weight women it is uncertain whether or not the intervention is more clinically effective than usual care (with respect to all the outcome measures) and whether or not the intervention is more costly than the alternative. The results of the deterministic sensitivity analyses demonstrated that the results were particularly sensitive to the estimates of the treatment effect in terms of the odds of developing PE.

Strengths and limitations of the economic study

The strength of this model-based economic evaluation is that the effect of interventions to manage weight gain in pregnancy was estimated via an IPD meta-analysis, involving data relating to 17,727 women and 30 studies. Furthermore, the resource use data were collected via a series of systematic reviews to identify studies that collected primary cost data. These reviews involved wide and detailed search and inclusion strategies. In addition, the study benefited from significant clinical input throughout its design and development. This study contributes to an area in which there is a paucity of economic studies. 101 The current public health emphasis on obesity and healthy lifestyles139 highlights the importance of contributions to understanding the costs and benefits of interventions in this area.

There were also some weaknesses. First, limited evidence was available about the resource use associated with some conditions in pregnancy. Although resource use data from high-quality RCTs were used to inform the model, the paucity of studies available limited the comparisons that could be undertaken to examine variations in costs for different groups of patients. A second limitation was that the study included a wide range of studies with different kinds of intervention models. This meant that it was difficult to estimate the costs associated with the intervention precisely; instead a median value was used, based on the findings of a systematic review. The cost of the intervention varied widely in the sensitivity analysis to account for the diversity of intervention types included in the IPD meta-analysis. In addition, for some outcomes, such as admission to the NICU of an infant whose mother did not have a pregnancy-related complication, it was difficult to obtain robust estimates of resource use. To address this limitation, the sensitivity analysis explored a wide range of plausible values for the costs associated with pregnancy-related conditions. A further limitation of the study is that outcomes were expressed in terms of clinical effectiveness rather than in terms that would allow comparison across programme areas, such as quality-adjusted life-years (QALYs). The absence of robust QALY data for women who are experiencing pregnancy-related conditions101,136 meant that a full cost–utility analysis incorporating QALYs was not possible. This means that some of the results are difficult to interpret as no willingness-to-pay threshold exists for individual clinical outcomes. As the results of the sensitivity analysis demonstrated that there was no significant difference between the intervention and the control arms results for either costs or clinical effectiveness, longer-term economic modelling was not undertaken. Finally, the economic evaluation was based on the results of the IPD meta-analysis. The findings may have been different if the evaluation had been based on the results of the meta-analysis using aggregated data (as these showed a statistically significant reduction for some outcomes). Further work could be undertaken to explore the cost-effectiveness of diet- and physical activity-based interventions using data from the aggregate meta-analyses, as this was beyond the scope of the current project.

Comparison with other studies

A limited number of studies were identified that were concerned with the costs and benefits of interventions to promote weight management during pregnancy. Only one study was identified that concluded that such an intervention was cost-effective. Dodd et al. 128 concluded that an intervention involving a lifestyle advice service was likely to be cost-effective using a monetary value of AUD 20,000 as a threshold for avoiding additional infants with a birthweight above 4 kg. However, several other studies have found no evidence of statistically significant differences in outcome measures and concluded that the intervention to manage weight gain in pregnancy was not cost-effective. For example, Oostdam et al. 132 examined the cost-effectiveness of an exercise intervention and concluded that the intervention was not cost-effective based on a range of outcome measures including infant birthweight and QALYs. Similarly, Kolu et al. found that a mixed intervention to prevent GDM was not cost-effective based on improvements in birthweight or 15D instrument scores. 129 A large-scale study concluded that a health training intervention was not cost-effective compared with usual care, based on comparison of QALY gains and costs for women. 68 Finally, a cost-comparison study for a weight-gain restriction programme for obese women found that that the weight-gain restriction programme was effective but had higher costs. 127

Meaning of the study

The results of the economic evaluation suggest that there is no evidence of cost-effectiveness for mixed interventions to manage weight gain in pregnancy. However, the lack of robust data on the quality of life of women and infants in the perinatal period means that further research is needed to fully understand the benefits of such programmes.

Unanswered questions and future research

The results of this economic evaluation highlight the need for accurate data on the quality of life of mothers and infants in the perinatal period, particularly around the impacts on quality of life for women with pregnancy-related conditions. This would enable a fuller analysis of the impact of interventions to manage weight gain in pregnancy on women’s health and that of their children. There is also a need for further work exploring the longer-term costs of weight gain in pregnancy for the mother and infant. This would need to include consideration of the wider societal costs of weight gain during pregnancy, as these are likely to be broader than health alone.