Objective
The objective of the study was to evaluate the cost-effectiveness of risk stratification with cervical length (CL) measurement and/or fetal fibronectin (fFN) tests in women with threatened preterm labor between 24 and 34 weeks’ gestation.
Study Design
We performed a model-based cost-effectiveness analysis to evaluate 7 test-treatment strategies in women with threatened preterm labor from a health care system perspective. Estimates on disease prevalence, costs, and test accuracy were based on medical literature.
Results
We found that additional fFN testing in the case of a CL between 10 and 30 mm is cost saving without compromising neonatal health outcomes, compared with a treat-all strategy or single CL testing. Implementing this strategy could lead to an annual cost saving between €2.8 million and €14.4 million in The Netherlands, a country with about 180,000 deliveries annually.
Conclusion
In women with threatened preterm labor between 24 and 34 weeks of gestation, the most cost-effective test strategy uses a combination of CL and fFN testing.
Preterm delivery, defined as birth before 37 weeks of gestation, occurs in 5-13% of all deliveries worldwide and is a major cause of perinatal mortality and morbidity. The number of women presenting with threatened preterm labor between 24 and 34 weeks of gestational age is approximately 2-3 times as high.
At present, many women with threatened preterm labor are initially admitted to a peripheral hospital and subsequently transferred to a perinatal center. They are treated with tocolysis to postpone delivery and corticosteroids to reduce neonatal mortality and morbidity in case of preterm delivery. However, 75-95% of the women with threatened preterm labor do not deliver within 7 days, and 40% even will deliver at term. As treatment leads to substantial costs and discomfort for the mother, appropriate identification of women at low risk of immediate preterm delivery could reduce unnecessary treatment in these women, resulting in cost savings without affecting health outcome.
Cervical length measurement and the fetal fibronectin test may help to identify women who would not benefit from treatment. Several systematic reviews reported promising negative predictive values for preterm birth within 7 days for cervical length (cutoff, 25 mm) or fetal fibronectin (cutoff, 0.050 μg/mL). Yet randomized clinical trials showed that implementing the fetal fibronectin test into a clinical setting will not lead automatically to the expected clinical improvement and cost reduction. However, these trials did not follow a strict fixed protocol in which the therapy choices are mandatory and are based on the test result.
On the contrary, multiple metaanalyses concluded that the combination of fetal fibronectin testing and cervical length measurement may reduce unnecessary costs. Unfortunately, the included studies combined fetal fibronectin testing and cervical length measurement in different ways, so it remains unclear from these studies how to combine both tests to maximize cost reduction without compromising neonatal outcomes.
This economic evaluation study evaluates the effectiveness, costs, and cost-effectiveness of several testing strategies, including those incorporating cervical length and fetal fibronectin, in women with threatened preterm labor between 24 and 34 weeks of gestation from a health care system perspective.
Materials and Methods
Model structure
A decision tree was designed to compare 7 test-treatment strategies in women with threatened preterm labor, intact membranes, and a gestational age between 24 and 34 weeks. Five of these management schemes incorporated risk assessment with fetal fibronectin (fFN) testing and/or cervical length (CL) measurement; one was a treat-all strategy, and one was a no-treatment strategy. Preterm delivery in this analysis was defined as delivery within 7 days after presentation because this outcome is related to the need for immediate transfer to a perinatal center, hospitalization, and treatment ( Figure 1 ). Treatment was defined as the administration of tocolysis and steroids, combined with the transfer of women to a perinatal center if they were currently in a peripheral hospital.
The decision tree consisted of the following strategies: (1) the reference strategy consisted of treating all women with threatened preterm labor with tocolysis and steroids, and transferring them to a tertiary center; (2) the testing strategies that we evaluated were measuring CL and treating women with a CL of less than 25 mm; (3) testing fFN and treating women with a positive fFN test (fFN ≥0.050 μg/mL); (4) testing fFN and CL and treating women if both tests are positive (ie, fFN of ≥0.050 μg/mL and a CL <25 mm); (5) testing fFN and CL and treating if either test is positive; (6) testing fFN only if a CL test is 10-30 mm and treating women either who had a CL of less than 10 or a CL 10-30 mm and a positive fFN test; and (7) no test and no treatment.
The cutoff value for CL in strategy 6 was based on a Dutch pilot study investigating the prognostic accuracy of fibronectin. The other cutoff values for CL and fFN were chosen based on systematic reviews. We assumed that within each strategy, except for the reference and no-treatment strategy, the allocation of treatment would strictly depend on the test results (ie, using a fixed protocol).
Outcomes
Strategies were compared for the following outcomes: proportion of patients treated, perinatal death, a composite of adverse neonatal outcomes, and costs. The composite neonatal outcome consists of perinatal death, chronic lung disease, neonatal sepsis, intraventricular hemorrhage greater than grade II, periventricular leukomalacia greater than grade I, and necrotizing enterocolitis. Costs were calculated from the health care system’s perspective. Only costs and neonatal outcomes until neonatal discharge were taken into consideration because the necessary information to specify model parameters in a decision model with a long-time horizon is lacking.
Model parameters
All model parameters are presented in Table 1 , with the best available point estimate as the base case, and a plausible range expressing parameter uncertainty. All probabilities are conditional. We assumed that the risk of preterm delivery within 7 days among women with threatened preterm labor was equally likely at any gestational age. We also assumed that all women without a preterm delivery would not deliver before 34 weeks of gestation. Parameter estimates for test accuracy were derived from the literature. A MEDLINE and PubMed literature search, restricted to literature in English in the last 10 years, was conducted using the key words fetal fibronectin, fibronectin, cervical length, preterm birth, premature birth, and preterm labor.
| Costs (2011 euros) | Base case | Range a | Distribution | References |
|---|---|---|---|---|
| CL measurement | 38.2 | 10–60 | Normal | Dutch Healthcare Authority |
| fFN test | 70 | 50–100 | Normal | Benelux Price List Hologic |
| Patient transfer | 343 | 243–504 | Normal | Standard unit price |
| fFN transfer to laboratory | 50 | 25–300 | Normal | Standard unit price |
| Treatment threatened labor | 4653 | 2000–8000 | Normal | Bottom-up method |
| Watchful waiting (no treatment) | 958,33 | 500–1500 | Normal | Bottom-up method |
| Living child <34 wks | 84028 | 46,711–98,565 | Gamma | Phibbs and Schmitt |
| Death <34 wks | 90042 | 11,297–97,925 | Gamma | Phibbs and Schmitt |
| Living child >34 wks | 3931 | X | X | Phibbs and Schmitt |
| Death >34 wks | 68934 | X | X | Phibbs and Schmitt |
| Probabilities | ||||
| PTD | 0.08 | 0.02–0.30 | Beta | Sanchez-Ramos et al |
| CL 0-9 mm if no PTD | 0.05 | 0.01–0.15 | Beta | Wilms et al, Tsoi et al, Fuchs et al, Gomez et al, |
| CL 0-9 mm if PTD | 0.51 | 0.25–0.75 | Beta | Wilms et al, Tsoi et al, Fuchs et al, Gomez et al, |
| CL 10-30 mm if no PTD b | 0.43 | 0.20–0.60 | Beta | Gomez et al, Wilms et al |
| CL 10-30 mm if PTD b | 0.84 | 0.60–0.90 | Beta | Wilms et al, Tsoi et al, Gomez et al, Fuchs et al |
| CL <25 if CL 10-30 mm and no PTD | 0.76 | 0.50–1.00 | Beta | Wilms et al |
| CL <25 mm if CL 10-30 mm and PTD | 1.00 | 0.75–1.00 | Beta | Wilms et al |
| CL <25 mm if fFN negative and no PTD | 0.38 | 0.30–0.41 | Beta | Gomez et al, Wilms et al |
| CL <25 mm if fFN negative and PTD | 0.82 | 0.80–1.00 | Beta | Gomez et al, Wilms et al |
| CL <25 mm if fFN positive and no PTD | 0.61 | 0.55–0.65 | Beta | Gomez et al, Wilms et al |
| CL <25 mm if fFN positive and PTD | 0.86 | 0.70–0.94 | Beta | Gomez et al, Wilms et al |
| fFN positive if CL 10-30 mm and no PTD | 0.29 | 0.10–0.50 | Beta | Gomez et al, Wilms et al |
| fFN positive if CL 10-30 mm and PTD | 0.75 | 0.50–0.98 | Beta | Gomez et al, Wilms et al |
| fFN positive if no PTD | 0.18 | 0.16–0.21 | Beta | Sanchez-Ramos et al |
| fFN positive if PTD | 0.76 | 0.69–0.82 | Beta | Sanchez-Ramos et al |
| Perinatal death with ACS | 0.04 | 0.02–0.05 | Beta | APOSTEL-II |
| Perinatal death without ACS | 0.05 | 0.02–0.06 | Beta | APOSTEL-II, Roberts and Dalziel |
| Perinatal death after 34 wks | 0.01 | X | X | APOSTEL-II |
| Severe adverse neonatal outcome with ACS b | 0.22 | 0.18–0.27 | Beta | APOSTEL-II |
| Severe adverse neonatal outcome without ACS b | 0.37 | 0.33–0.43 | Beta | APOSTEL-II, Roberts and Dalziel |
| Severe adverse neonatal outcome after 34 wks b | 0.01 | X | X | APOSTEL-II |
| RRs | ||||
| RR on perinatal death if ACS | 0.77 | 0.67–0.89 | Log-normal | Roberts and Dalziel |
| RR on severe adverse neonatal outcome b | 0.59 | 0.41–0.88 | Log-normal | Roberts and Dalziel, raw data, APOSTEL-II |
a Range used for univariate sensitivity analyses
b Composite of perinatal death, chronic lung disease, neonatal sepsis, intraventricular hemorrhage greater than grade II, periventricular leukomalacia greater than grade I, and necrotizing enterocolitis.
Studies used for data extraction had to meet the following criteria: (1) the population included women between the 24 and 34 weeks’ gestational age with complaints or symptoms indicating threatened preterm delivery, including contractions, vaginal blood loss, or abdominal/back pain. Both singletons and twin pregnancies could be included; (2) the primary outcome had to be preterm delivery (PTD) within 7 days after inclusion/or presentation (this outcome suits the urgency of immediate referral to tertiary centers); and (3) the study results allowed data extraction, meaning that the CL cutoffs and definition of preterm delivery mentioned in previous text were used. Calculating sample size-weighted means of estimates was not possible because the studies used different cutoffs for CL measurements. Ranges used for the sensitivity analyses were based on the maximal range reported in the literature.
In addition to the estimates from the literature, we used the raw data of the Assessment of Perinatal Outcome with Sustained Tocolysis in Early Labour (APOSTEL)-II trial in combination with a systematic review for the health outcomes. The APOSTEL-II trial investigated the effect of prolonged tocolysis in women with threatened preterm labor between 24 and 34 weeks with intact membranes. Because all of these women received tocolysis and corticosteroids, we could calculate the health outcomes in nontreated women with the relative risks from a systematic review.
To estimate treatment costs, we assumed that all women had threatened preterm labor such that they would be tested for CL and fFN (if applicable for the test strategy). In the base case situation, fFN testing was available in all perinatal centers, whereas CL measurements could be performed in all hospitals. All women who have a positive test result receive a full course of corticosteroids with tocolysis. Direct medical costs, due to CL measurement and fFN testing, treatment during maternal and neonatal admissions, and costs of transport, were included. Unit cost estimates for neonatal hospitalizations were derived from literature, costs due to maternal admission and treatment from the financial department of a Dutch academic hospital, transport costs from Dutch guideline prices, and fFN costs from the Hologic price list.
The fFN costs included equipment and maintenance costs. Costs were expressed in 2011 euros and were inflated where appropriate using the consumer pricing index. Annual discounting was unnecessary because the time horizon was within 1 year. Cost ranges used for sensitivity analyses were based on 95% confidence intervals (CIs) derived from the literature for neonatal costs, on minima and maxima derived from the literature for transfer costs, and on a fixed ratio of half and double the point estimate in case of a bottom-up approach (such as for CL and fFN costs).
During multivariate sensitivity analyses, we assumed beta distributions for the probability parameters (prevalence, test accuracy, health outcomes) and gamma distributions for the cost parameters.
Analysis
Base-case analyses compared strategies using fFN and/or CL measurement (strategies 2-6) with the reference strategy (strategy 1), in terms of percentage of patients treated, costs, perinatal death, and a composite of adverse neonatal outcomes. First, we identified dominated strategies, which are less effective and more costly than others strategies. Because these dominated strategies are not viable options, they were excluded in further analyses.
Second, we performed univariate sensitivity analyses, in which we varied parameters within their predefined ranges separately to mimic the influence differences in health care systems and parameter variance. Changes in dominance (in comparison with the base case analyses) were described because this indicates model robustness.
Then we performed a probabilistic Monte Carlo simulation to determine the simultaneous effect of uncertainty across multiple parameters on the cost-effectiveness estimates. We compared each strategy with the next more expensive strategy in terms of cost savings and neonatal health loss, starting with the most costly reference strategy (treat-all). Based on the Monte Carlo simulations, we calculated 95% CIs of health and cost differences between the strategies. These data were used to select the most cost-effective strategy, in which we weighed the magnitude of cost-savings against the amount of neonatal health loss. If a strategy could lead to significant cost reductions with only a nonsignificant deterioration in neonatal health outcomes, it was considered as cost-effective.
Finally we performed 2 scenario analyses for implementation of the most cost-effective strategy in The Netherlands because variance in (current) practice is possible. First, we compared the most cost-effective strategy to current practice in The Netherlands, which encompasses 180,000 deliveries and 25,000 cases of threatened preterm labor annually, of which approximately 5000 present before 34 weeks’ gestational age. Because the percentage of women who receive CL measurements for treatment decisions in The Netherlands is unknown, we varied this percentage of CL measurements among different scenarios.
The most cost-effective strategy was compared with these different scenarios; for each scenario we calculated the potential effects of implementing the most cost-effective strategy in term of cost savings and additional adverse neonatal outcomes with 95% CIs.
Second, we compared scenarios with different locations at which the fibronectin assay was performed: (1) all peripheral (referral) hospitals, (2) a central laboratory, in which case the fibronectin sample was transferred for analysis, and (3) perinatal centers, in which case the patient was transferred to perform the fibronectin test (reference scenario).
The decision tree was created and analyzed with TreeAge Pro 2009 Suite software (TreeAge Software, Inc, Williamstown, MA).
Results
Comparing the base case point estimates of the 7 strategies showed that no treatment at all (strategy 7; mean, €11,840 per patient) and treating all patients (reference strategy 1; mean, €15,872 per patient) were the least and most costly strategies, respectively ( Figure 2 ). Strategy 3 (fFN only) was dominated because strategy 6 (fFN in case of CL, 10-30 mm) was more effective at lower costs. The remaining strategies were combinations of fFN testing and CL measurements and CL measurement only.
