To determine whether routine measurement of second-trimester transvaginal cervical length by ultrasound in low-risk singleton pregnancies is a cost-effective strategy.
To determine whether routine measurement of second-trimester transvaginal cervical length by ultrasound in low-risk singleton pregnancies is a cost-effective strategy.
We developed a decision analysis model to compare the cost-effectiveness of two strategies for identifying pregnancies at risk for preterm birth: (1) no routine cervical length screening and (2) a single routine transvaginal cervical length measurement at 18–24 weeks' gestation. In our model, women identified as being at increased risk (cervical length < 1.5 cm) for preterm birth would be offered daily vaginal progesterone supplementation. We assumed that vaginal progesterone reduces preterm birth at < 34 weeks' gestation by 45%. We also assumed that a decreased cervical length could result in additional costs (ultrasound scans, inpatient admission) without significantly improved neonatal outcomes. The main outcome measure was incremental cost-effectiveness ratio.
Our model predicts that routine cervical-length screening is a dominant strategy when compared to routine care. For every 100 000 women screened, $ 12 119 947 can be potentially saved (in 2010 US dollars) and 423.9 quality-adjusted life-years could be gained. Additionally, we estimate that 22 cases of neonatal death or long-term neurologic deficits could be prevented per 100 000 women screened. Screening remained cost-effective but was no longer the dominant strategy when cervical-length ultrasound measurement costs exceeded $ 187 or when vaginal progesterone reduced delivery risk at < 34 weeks by less than 20%.
In low-risk pregnancies, universal transvaginal cervical length ultrasound screening appears to be a cost-effective strategy under a wide range of clinical circumstances (varied preterm birth rates, predictive values of a shortened cervix and costs). Copyright © 2011 ISUOG. Published by John Wiley & Sons, Ltd.
Preterm birth is a major cause of neonatal morbidity and mortality in the USA. According to the National Vital Statistics Reports, 11–12% of the 4 million neonates born each year are delivered before 37 weeks' gestation and 3.6% are delivered before 34 weeks. Early preterm birth (before 34 weeks) is particularly associated with high rates of mortality and morbidity, including intraventricular hemorrhage, necrotizing enterocolitis, respiratory distress syndrome and neurological deficits1. The relevance of preterm birth as a major public health problem is emphasized by the more than $ 26 billion spent annually in the USA to treat these preterm infants2. Additionally, over 50% of neonates born before 28 weeks are rehospitalized within 24 months of birth and almost one third are diagnosed with asthma or reactive airway disease by their fifth birthday3. 9.1% of surviving infants born between 23 and 28 weeks' gestation are diagnosed with cerebral palsy, 4.4% with mental retardation and 2.5% with behavioral or emotional disorders4.
Traditionally, physicians assessed a woman's risk for preterm birth using clinical factors including multiple gestations, history of preterm birth or prior cervical surgery. Yet, more than half of all preterm births occur to women without historical risk factors5–7. Compelling data suggest that the length of the cervix as assessed by transvaginal ultrasonography is inversely related to the risk of preterm birth, and may be a stronger predictor than the previously used clinical risk factors8–10. For example, Goldenberg et al.6 demonstrated that a cervical length of < 2.5 cm had a relative risk of 3.5 (95% confidence interval (CI), 2.7–4.6) for preterm birth, while previous preterm birth had a relative risk of 2.7 (95% CI, 2.1–3.4).
Although valuable, this information had limited clinical utility until more recent studies demonstrated interventions that could potentially reduce the frequency of preterm birth11, 12. In a randomized trial, Fonseca et al.11 demonstrated that daily vaginal progesterone reduced the risk of preterm birth by 45% in low-risk women identified with a cervical length of ≤ 1.5 cm. Given this finding, we sought to investigate the cost-effectiveness of routine cervical-length screening in the low-risk population. To this end, we constructed a decision-analysis model, which provided evidence that universal cervical-length screening would be cost-effective.
We used a decision-tree model to compare two clinical strategic approaches to preterm birth prevention in low-risk pregnancies. The population targeted included singleton pregnancies in women without a history of prior preterm birth. The first strategy was consistent with the current clinical recommendations advocated by the American College of Obstetricians and Gynecologists: no screening for preterm birth in asymptomatic low-risk pregnant women with a singleton gestation13. The second strategy included performance of a single routine transvaginal ultrasound cervical-length measurement on all asymptomatic, low-risk singleton pregnant individuals at between 18 and 24 weeks' gestation. In the second strategy the cervical length was considered short (< 1.5 cm), mid length (1.5–2.49 cm), or normal length (≥ 2.5 cm).
Our model proposed that women with a cervical length of < 1.5 cm be offered vaginal progesterone11, 14. This strategy was based on the data presented by Fonseca et al.11 This trial enrolled 24 620 asymptomatic women, including 24 189 singletons, for cervical-length measurement. Cervical length was ≤ 1.5 cm in 413 women, 250 of whom were randomized to receive vaginal progesterone or placebo. The primary outcome was delivery before 34 weeks' gestation. Thirty-six of 112 women (32.1%) with singletons in the placebo group went on to deliver prior to 34 weeks compared to 20 of 114 (17.5%) in the progesterone group. Thus, we assumed that there would be a 45% reduction in deliveries before 34 weeks with progesterone administration. An adherence rate of 92% was also assumed based on the data of Fonseca et al.11.
The baseline probability and outcomes for each strategy were obtained based on a comprehensive bibliographic survey of the English literature in PubMed, using the following search terms: vaginal progesterone, Prometrium, preterm birth, preterm delivery, preterm prevention and cervical length, as well as combinations of these terms. Point estimates were determined from published randomized controlled trials and prospective cohorts when possible. Retrospective cohorts or review studies were used when no other sources of information were available. The decision tree was developed and the analysis performed with TreeAge Pro 2007 (TreeAge Software, Williamstown, MA, USA). The probability estimates and the references used in support of our model are reported in Table 1.
|Variable||Base case (range) (%)*|
|Preterm birth (< 34 weeks)1, 11, 15||2.1 (2.0–3.6)|
|If birth at < 34 weeks, probability of birth at < 28 weeks11||20.6|
|If birth at ≥ 34 weeks, probability of birth at < 37 weeks11||8.4|
|Prevalence of cervical length|
|< 1.5 cm9, 11, 17||1.7 (0.9–1.88)|
|1.5–2.49 cm11||8.3 (7.9–8.7)|
|≥ 2.5 cm11||90 (89.5–91.8)|
|Prevalence of inpatient admission if cervical length < 1.5 cm||0.0 (0.0–100.0)|
|Delivery at < 34 weeks if cervical length < 1.5 cm8, 11||34.1 (9.7–58.7)|
|Delivery at < 34 weeks if cervical length 1.5–2.49 cm9, 11||5.1 (4.2–14.0)|
|Delivery at < 34 weeks if cervical length ≥ 2.5 cm10, 11||1.2 (1.1–3.0)|
|Adherence to progesterone therapy11||92.8 (86.0–97.0)|
|Reduction in deliveries prior to 34 weeks with progesterone11||45 (12.0–66.0)|
|Probability of severe disability if delivery at4:|
|< 28 weeks||10.6 (9.1–17.1)|
|≥ 28 weeks, < 34 weeks||5 (3.3–9.7)|
|≥ 34 weeks, < 37 weeks||2.4 (2.1–2.6)|
|≥ 37 weeks||1.7 (1.5–1.8)|
|Probability of death if delivery at16:|
|< 28 weeks||17.9 (8.0–49.3)|
|≥ 28 weeks, < 34 weeks||0.9 (0.2–8.6)|
|≥ 34 weeks, < 37 weeks||0.2 (0.1–0.4)|
|≥ 37 weeks||0.07 (0.05–0.09)|
|Utility of neonatal outcome18|
|Severe neurologic disability||0.61 (0.5–0.8)|
The overall incidence of preterm birth at < 34 weeks' gestation was estimated at 2.1% based upon data provided by Fonseca et al.11. A preterm birth rate as high as 3.6% for birth at < 34 weeks' gestation was based upon data from the 2006 United States National Vital Statistics data1 and as low as 2% from UK data15. Given the differences in neonatal morbidity and mortality between early preterm birth at < 28 weeks' gestation compared with those delivering at ≥ 28 weeks, we utilized data from United States National Vital Statistics summaries to estimate the proportion of deliveries at < 28 weeks, 28–34 weeks, 34–37 weeks and ≥ 37 weeks4, 16. Of the deliveries that occurred at < 34 weeks, we assumed that 20% would occur prior to 28 weeks while the remaining 80% would deliver at 28–34 weeks1. Of the deliveries that occurred after 34 weeks' gestation, 8% were assumed to occur at between 34 and 37 weeks while the remaining 92% were full-term births1.
The prevalence of cervical length of < 1.5 cm and of 1.5–2.49 cm is fairly consistent in the literature (0.9–1.88% and 7.9–8.7%, respectively)9, 11, 17; we used these ranges in our sensitivity analysis. However, we also explored implausible prevalences to establish the threshold at which cervical-length ultrasound measurement was no longer cost-effective. The preterm birth rates used in the model were inversely related to cervical length based on previously published data8–11.
Only the study of Fonseca et al.11 examined the effect of vaginal progesterone therapy on neonatal morbidity. It showed a reduction in morbidity but the results were not statistically significant, as the study was not specifically powered for this purpose. Thus we used large pediatric cohort studies to estimate the mortality and short- and long-term morbidities based on gestational age at birth4, 16.
Based upon the literature, utilities were given to the offspring studied in this model18. Utilities are a means of evaluating the relative quality of life as compared to health. We determined three health states that would be relevant for our analysis: normal health (utility = 1), severe disability (utility = 0.61) and death (utility = 0). Severe disability was defined as serious medical conditions that significantly limit working capacity and included cerebral palsy, mental retardation, blindness, deafness and epilepsy4. Severe disability was assigned a value of 0.61 based on the decision analysis of Odibo et al.19 for the use of 17-alpha-hydroxyprogesterone caproate for the prevention of preterm birth. A range of 0.5–0.8 was used in our sensitivity analysis based on Tengs and Wallace's quality of life data18. This range encompasses several severe disabilities and all moderate disabilities. We ran our model using many different life expectancies for the premature infants. In the final analysis, we assigned an average life expectancy of 76 years for the purpose of calculating the quality-adjusted life-years (QALYs) for all surviving offspring17. We chose to assume that the life span of premature infants who survive the neonatal period is not significantly shortened in an effort to bias our model against screening, thus further validating the efficiency of screening if the model was cost-effective, even assuming this best case scenario.
Cost data were derived from the published literature (Table 2). Screening consisted of transvaginal ultrasono-graphic measurement of cervical length at between 18 and 24 weeks' gestation. The costs associated with this ultrasound scan were determined from Medicare data using the current procedural terminology (CPT) code 7681720. In the base-case analysis, only practice expenses were included (1.95 relative value units (RVUs), using an average RVU cost of $ 36.07). In the sensitivity analysis, the total cost of a transvaginal ultrasound scan was adjusted to as low as $ 50/scan given geographic discounting and as high as $ 300/scan to include physician fees, malpractice fees, geographic increases and charges associated with repeating fetal growth assessment (CPT code 76816).
|Variable||Base case (range*) in 2010 ($ )|
|Cervical-length ultrasound scan (cost per scan)20||70 (50–300)|
|Vaginal progesterone supplementation (total cost for gestation)21||206 (100–400)|
|Admission cost for short cervix, including corticosteroids22||0 (0–10 000)|
|Cost of maternal care if delivery at23:|
|< 28 weeks||10 953 (5477–21 907)|
|≥ 28 weeks, < 34 weeks||8153 (4077–16 307)|
|≥ 34 weeks, < 37 weeks||4627 (2314–9254)|
|≥ 37 weeks||3577 (1789–7155)|
|Cost of neonatal care if delivery at23, 25:|
|< 28 weeks||207 927 (85 251–415 922)|
|≥ 28 weeks, < 34 weeks||37 159 (15 235–74 329)|
|≥ 34 weeks, < 37 weeks||5460 (2239–10 922)|
|≥ 37 weeks||1806 (741–3613)|
|Cost of early intervention (0–3 years) if delivery at16:|
|< 28 weeks||8847 (4427–17 699)|
|≥ 28 weeks, < 34 weeks||4201 (2102–8406)|
|≥ 34 weeks, < 37 weeks||1632 (816–3264)|
|≥ 37 weeks||862 (431–1725)|
|Cost of severe disability24||262 667 (131 333–525 335)|
|Income and domestic productivity losses due to bed rest30||0 (0–11 310)|
If the cervix was short (< 1.5 cm), maternal costs included nightly progesterone administration until delivery or 36 weeks' gestation21. These women also received two follow-up cervical-length ultrasound scans. In the sensitivity analysis we varied the number of follow-up scans from none to four. In the base-case analysis, we did not account for any hospital admissions purely for a short cervix. However, in the sensitivity analysis we varied admission rates from 0 to 100% for a cervical length of < 1.5 cm and 0 to 50% for a cervical length of 1.5–2.49 cm. As we adjusted the admission rates we also varied admission costs from 0 to $ 10 000, with a median cost of $ 3000. This median was selected as it is consistent with a 48-h stay to facilitate steroid administration22. We included a maximum of $ 10 000 to include up to 1 week of inpatient care. Within the sensitivity analysis, we performed a similar evaluation of potential hospital admissions and treatment without clinical benefits in women with mid cervical length (1.5–2.49 cm).
Delivery costs were based upon the gestational age at delivery, as the length of stay and the percentage of Cesarean deliveries are known to be inversely related to the timing of delivery23. Offspring costs were broken down into: neonatal care costs, costs of care in the first 3 years, and long-term costs associated with severe neurologic disability4, 16, 23–25. The cost of care in the first 3 years was included in an attempt to account for the early intervention and special education that many of these children receive. Long-term care costs included only direct medical costs, thus productivity losses, both to the affected individual and their family members, are not included. We based long-term disability costs on cerebral palsy data as that was the most prevalent disability noted in this population4.
All costs are presented in 2010 US dollars and were adjusted based on the use of the medical care component of the Consumer Price Index. Costs and utilities were discounted at a baseline rate of 3% based on average inflation, although the range was varied from 1–5% in the sensitivity analysis. All analyses were from a societal perspective.
For a cohort of 100 000 women, we calculated the cost of care for each strategy. The primary outcome of the study was cost-effectiveness, measured as the incremental cost-effectiveness ratio (ICER). Cost-effectiveness was defined as an ICER of $ 100 000. We performed univariate sensitivity analyses by varying the values of the variables in the model to their plausible extremes. Other parameters we estimated included: total cost of each strategy, total QALYs per strategy, incidence of preterm birth and incidence of adverse neonatal outcomes such as fetal death or disability.
The results for the base-case model are presented in Table 3. Our model predicts that the current standard of care costs $ 1 314 520 247 per 100 000 low-risk women, while the care model involving routine screening would cost $ 1 302 400 300 per 100 000 low-risk women. We estimate that screening would prevent 248 births before 34 weeks' gestation and 22 neonatal deaths or neonates with long-term neurologic deficits per 100 000 deliveries. Thus, screening is the dominant strategy, saving cost with improved outcomes.
|Variable||Standard procedure||With screening|
|Neonatal death (n)||170||159|
|Severe neurologic deficits (n)||1827||1816|
|Neurologic deficit/death averted (n)||—||22|
|Births < 34 weeks' gestation (n)||2106||1858|
|Births < 34 weeks' gestation averted (n)||—||248|
|Total QALY||2 954 795||2 955 218|
|Marginal QALY gained||—||423.9|
|Total cost ($ )||1 314 520 247||1 302 400 300|
|Marginal cost savings ($ )||—||12 119 947|
|Marginal cost ($ )/QALY gained||—||28 592|
We performed a univariate sensitivity analysis to evaluate the impact of changing the probability and cost variables on the ICER of screening with transvaginal cervical-length ultrasonography (Table 4). In this study, a negative value (denoted by parentheses) equates to a cost saving. The model was robust for all parameters at the ranges that we examined. As expected, the model was sensitive to changes in the cost of cervical-length ultrasound scans, the effectiveness of progesterone in preventing preterm delivery, the predictive value of a shortened cervix and the prevalence of a shortened cervix.
|Variable||Variable minimum||Variable maximum|
|Preterm birth (at < 34 weeks' gestation)||(26 814)||(38 175)|
|Prevalence of cervical length < 1.5 cm||(14 100)||(29 667)|
|Prevalence of inpatient admission if||(28 652)||(584)|
|delivery at ≥ 34 weeks|
|Delivery at < 34 weeks if cervical length||33 275||(38 237)|
|< 1.5 cm|
|Adherence to progesterone therapy||(26 327)||(28 965)|
|Reduction in deliveries prior to 34||35 264||(34 801)|
|weeks with progesterone|
|Probability of severe disability||(28 058)||(29 004)|
|Probability of death||(35 404)||(23 133)|
|Utility of disability||(26 069)||(32 175)|
|Cost of cervical-length ultrasound scan||(33 519)||27 461|
|Cost of progesterone||(28 951)||(27 794)|
|Cost of maternal care||(26 578)||(30 919)|
|Cost of admission for short cervix||(28 652)||(10 632)|
|Cost of neonatal care||(4307)||(68 236)|
|Cost of early intervention||(26 919)||(30 258)|
|Cost of severe disability||(25 080)||(33 933)|
While the variables listed above cause screening to shift from a cost-saving to a cost-effective strategy, there was no plausible situation in which the no-screening strategy was dominant. For example, when the cost of a single transvaginal ultrasound scan passes $ 187, cervical-length screening is no longer cost saving, but it remains cost-effective. Even at an extreme cost of $ 300 per transvaginal scan, screening costs $ 27 461 per QALY gained. When administration of vaginal progesterone reduces preterm birth by less than 20% instead of the previously predicted 45%, screening ceases to be a cost-saving strategy but remains cost-effective. Likewise, if the probability of delivery before 34 weeks with a shortened cervix is only 9.7%, not the estimated 34%, screening costs $ 33 276 per QALY gained.
We varied the prevalence of cervical length < 1.5 cm from 0.9 to 1.88% based upon values found in the literature9, 11, 17. In this range, cervical-length screening is cost saving. In order to identify transition values, we also ran the model with a prevalence as low as zero. When the prevalence of a cervical length < 1.5 cm falls below 0.8%, screening is no longer cost saving, but it remains cost-effective. Below the implausible prevalence of 0.35%, the no-screening strategy becomes dominant.
Monte Carlo Simulation (a computational algorithm that relies on repeated random sampling) was also used to simultaneously vary all variables across the extreme ranges listed in Tables 1 and 226. With 100 000 simulations, the screening strategy was cost-effective 99.4% of the time. In most of these instances (68%) screening was also cost saving.
Although only approximately one sixth of preterm births occur before 34 weeks' gestation, these neonates account for the majority of morbidity, mortality and cost1, 4, 16, 23, 27. Assessment of maternal history alone misses significantly more than half of women at risk for preterm birth5. Cervical-length measurement by transvaginal ultrasound provides a necessary screening tool to better predict which women are at risk for preterm birth7–9. Moreover, a combination of cervical-length measurement and vaginal progesterone supplementation has been shown to reduce the risk of preterm birth in a low-risk population11. Given this information, one could argue that expanding cervical-length screening to include the low-risk population would be reasonable. However, before any major screening initiatives are adopted, an understanding of the costs (intended and unintended) associated with screening is needed. We undertook this decision analysis to estimate the implications to the health of the offspring as well as the additional costs to the healthcare system associated with a comprehensive program using transvaginal ultrasound screening and progesterone intervention. Our analysis demonstrates that a screening program with appropriate interventions to reduce preterm birth would be cost-effective and in many cases cost saving.
Our findings are consistent with the only other study to examine cervical-length screening in low-risk patients, that of Cahill et al.28. While this prior study was thorough in its analysis, it did not account for many of the subsequent and unintended costs that a screening strategy would elicit. Thus, its conclusion that cervical-length screening is cost-effective needed to be verified in a study that included these unintended expenditures and consequences. The costs for serial cervical-length ultrasound scans, inpatient hospitalizations and progesterone administration for the mid-length (1.5–2.49 cm) cervix were included in our cost analysis despite the lack of evidence that any of these interventions improve neonatal outcome.
The model of Cahill et al. also used a significantly lower cost per transvaginal scan—$ 52, with a range of $ 43 to $ 74. These rates are consistent with Medicaid reimbursement for non-facility charges only at inexpensive locales. This range does not include technical fees, which are a significant portion of the cost of ultrasound services, or additional ultrasound fees from repeat cervical assessments. It also does not account for many sites, including many cities, where higher Medicaid rates are charged using an adjustment factor, the Geographic Practice Cost Indices. As the model is from a societal perspective, it is important that the range of costs includes geographic price differences and physician fees. This was particularly important as the cost per transvaginal ultrasound scan was one of the few variables to which the model was sensitive.
While our model is somewhat biased against screening compared to the Cahill et al.28 model (higher costs for ultrasound screening and more accountability for false-positives), we reached the same conclusion: a strategy of universal cervical-length screening would be cost-effective. This strengthens the growing evidence in favor of routine assessment of cervical length by ultrasonography.
As with any decision analysis, the accuracy of our outcomes depends on the quality of the data used within our model. Although we were fortunate to have a randomized trial that demonstrated the effects of progesterone supplementation at reducing the risk of preterm birth in low-risk women with a short cervix, this could also be considered a weakness, and it certainly would be ideal (and a luxury) to have multiple randomized controlled trials that demonstrate this finding. Based on our sensitivity analysis, if progesterone is found to be significantly less effective at preventing preterm birth before 34 weeks, then cervical-length screening may not be cost saving. Therefore, our study highlights the importance of verifying the efficacy of progesterone at reducing the risk of preterm birth with future randomized controlled trials.
Our analysis was from the perspective of society. Direct costs were ascertained and accounted for and we did not use indirect costs in our model. However, most of these costs are likely to favor screening, such as the psychological and financial impact of caring for a premature neonate. However, we did examine the potential effects of productivity losses from prescribed bed rest due to a short cervix. While bed rest has never been shown to decrease preterm birth rates in women with shortened cervices, a recent survey of members of the American College of Obstetricians and Gynecologists demonstrated that 34% of obstetricians prescribed bed rest for a cervical length of < 2.5 cm29. Thus, we explored the impact of bed rest on the cost-effectiveness of universal cervical-length screening in women with short and mid-length cervices (< 1.5 cm and 1.5–2.49 cm, respectively). The losses in work productivity and domestic productivity are based on data from Goldenberg et al.30 and adjusted to 2010 dollars. Even if all women with cervical length < 1.5 cm were placed on bed rest, cervical-length screening would remain cost saving. However if all women with a cervical length of < 2.5 cm were placed on bed rest, screening would shift from a cost-saving practice to a strictly cost-effective strategy, with each QALY costing $ 82 000. This highlights the large cost to society of prescribed bed rest despite its unclear efficacy.
We believe that this study provides a comprehensive analysis of the intended and unintended costs of transvaginal cervical-length screening in women at low risk for preterm birth. When this screening protocol is combined with vaginal progesterone treatment, screening is cost-effective for the healthcare system and should be investigated further.
We reanalyzed our model incorporating the recently published data of Hassan et al.31. We added an additional assumption to the base case that vaginal progesterone treatment reduced preterm birth rates in women with mid-pregnancy cervical lengths between 1.5 cm and 2.5 cm. With these adjustments, universal cervical length ultrasound screening continued to be the dominant strategy. For every 100 000 women screened we predict a net health improvement of 735 QALYs and net savings to the healthcare system of $ 19 603 380 with universal cervical length screening. The results by Hassan et al. strengthen the evidence that universal cervical length screening could both improve quality of life and be cost-saving under a wide range of circumstances.