Study design
The economic evaluation was conducted during a randomised controlled trial (RCT) of the FitFor2 exercise program for pregnant women at risk for GDM. The protocol was approved by the Medical Ethics Committee of the VU University Medical Centre, Amsterdam, the Netherlands. Written informed consent was obtained from all participants. Participants were followed from 15 weeks of pregnancy till 12 weeks postpartum. More details about study design and methods of this RCT have been described elsewhere [12].
Participant recruitment
Recruitment took place in five hospitals and 20 midwifery practices in the Netherlands, between November 2007 and April 2010. Participants were pregnant women at increased risk for GDM. Women were considered to be at increased risk for GDM if they were overweight (BMI ≥ 25) and had at least one of the three following characteristics: 1) history of macrosomia (offspring with a birth weight above the 97th percentile of gestational age), 2) history of GDM, or 3) first-grade relative with diabetes mellitus type 2 or if they were obese (BMI ≥ 30). Exclusion criteria included recruitment after 20 weeks of gestation; age less than 18 years; inadequate knowledge of the Dutch language; diagnosed with (gestational) diabetes mellitus before randomisation (fasting glucose >6.0 mmol/l); hypertension; alcohol abuse; drug abuse; use of medication that affects insulin secretion or insulin sensitivity; serious pulmonary, cardiac, hepatic or renal impairment; malignant disease; serious mental or physical impairment that made understanding or implementation of the study protocol/aim difficult.
Randomisation and blinding
Eligible women were randomised into the intervention or control group. Randomisation was stratified for the hospital where participants were offered the exercise program. Within each stratum, block randomisation with blocks of four was used to ensure that each group consisted of an equal number of participants. The researcher, patients, healthcare providers and research assistant were not blinded for allocation after randomisation due to the nature of the intervention. All outcome measures were assessed by independent research assistants who were unaware of group allocation.
Intervention
Women in the intervention group participated an exercise program twice weekly in a group during the remaining duration of their pregnancy. Each exercise session lasted for 60 minutes. The exercise sessions consisted of aerobic and strength exercises that help to control blood glucose levels. The training intensity was carefully and individually controlled. All exercise sessions were completed under the guidance and supervision of a specifically trained physiotherapist, and took place in the Department of Physiotherapy of the participating hospitals. Details of the exercise program were described previously by [Oostdam et al. 12].
Women in the control group were not offered an exercise program and received standard care from obstetricians and/or midwives. The primary task of the Dutch midwife is to closely follow the health status of the pregnant woman and her unborn child. Midwifes see their clients on average 13 times during pregnancy on an individual basis. In the Netherlands, overweight or obese women receive the same care as healthy-weight women. The control group was followed throughout the entire pregnancy period.
Outcomes/effects
Outcome measures were maternal fasting blood glucose, insulin sensitivity, infant birth weight, and quality-adjusted life years (QALYs). Outcomes were assessed at baseline (approximately 15 weeks of pregnancy), 24 and 32 weeks of pregnancy by laboratory tests and self-administered questionnaires.
Maternal fasting blood was drawn from the antecubital vein after the participant had fasted for at least 10 hours. In this blood sample, glucose and insulin were measured. For insulin sensitivity, the homeostasis model assessment (HOMA) was calculated.
In the Netherlands, birth weight is routinely measured and recorded by the obstetrician, midwife or the nurse at delivery. Birth weight was obtained from a questionnaire filled out by the women 12 weeks postpartum.
Health-related quality of life was measured using the EuroQol-5D questionnaire [13]. Utilities were determined using the Dutch tariff [14]. Quality-adjusted life years (QALYs) were calculated by multiplying the utilities by the time spent in a given health state. Transitions between health states were linearly interpolated.
Cost data collection and valuation
The economic evaluation was conducted from a societal perspective. Resource utilisation was assessed during pregnancy with three questionnaires at 15-, 24-, and 32-weeks of pregnancy. Direct costs included the costs of visits to healthcare providers, medication, and informal care. Indirect costs were costs related to sick leave. Costs of delivery were assessed using a questionnaire 12 weeks postpartum and were included in the direct costs.
Costs were reported in Euros and the index year was 2009 (this was the year that most data were collected). Costs were calculated by multiplying the respective units of resource use by standard costs according to Dutch Manual for Costing [15]. For visits to care providers for whom standard cost prices were not available, prices according to professional organisations were used. Medication was valued using unit prices published by Royal Dutch Society for Pharmacy [16]. Costs of absenteeism from paid work were calculated using The Friction Cost Method (FCM). The FCM assumes that costs are limited to the period necessary to replace a sick worker, the friction period. A friction period of 154 calendar days and an elasticity of 0.8 were used. The number of sick leave hours was multiplied by the cost of productivity loss per hour, based on age and gender.
Data analysis
A power calculation was done for the primary outcome measure of maternal fasting glucose. It was determined that adequate power (>0.80) and a 5% significance level would be achieved with 80 pregnant women in both groups. The power calculation allowed for a 20% dropout rate.
The economic evaluation was performed according to the intention-to-treat principle. Due to the considerable amount of missing follow-up data, missing data were imputed using multiple imputation (MI) based on Multivariate Imputation by Chained Equations (MICE) [17]. The MI procedure was performed in SPSS 18.0, in which five complete data sets were generated. Using Rubin’s rules, effects and costs from the five complete data sets were pooled [18].
Participants with high blood glucose (>6.0 mmol/l) at baseline (n = 4) and a twin pregnancy (n = 1) were excluded from the analyses, because they did not fulfil the inclusion criteria. Insulin levels that were out-of-range, and therefore very unlikely, were coded as missing and subsequently imputed (n = 7).
The incremental cost-effectiveness ratio (ICER) is defined as the ratio of the difference in costs and the difference in effects between the intervention and standard care. ICERs were calculated for maternal fasting blood glucose, insulin sensitivity, infant birth weight and QALYs. The ICER indicates the additional investment needed to gain one additional unit of effect. For the outcome infant birth weight, the cost-effectiveness analysis was done with total societal costs including delivery costs.
Non-parametric bootstrapping with 5000 replications was used to estimate “approximate bootstrap confidence” (ABC) intervals around cost differences [19]. Bootstrapping was also used to estimate the uncertainty surrounding the incremental cost-effectiveness and cost-utility ratios (5000 replications). The bootstrapped cost-effect pairs were plotted on a cost-effectiveness plane (CE plane) [20]. The CE plane shows the difference in effect on the x-axis and the difference in costs on the y-axis. Cost-effectiveness acceptability curves (CEA curves) were also estimated. CEA curves show the probability that the intervention is cost-effective in comparison with the control treatment for a range of ceiling ratios. The ceiling ratio is defined as the amount of money society is willing to pay to gain one unit of effect [21].
Data processing was performed in SPSS 15.0. The cost-effectiveness analyses were conducted in R.
Sensitivity analysis
In a secondary analysis, the Human Capital Approach (HCA) was used to estimate costs due to sick leave. In the HCA, total sick leave time is neither “capped” as in the FCA, nor is elasticity considered.