全文

Introduction

Anovulation is one of the leading causes of female infertility.About 80% of anovulatory women have a WHO type II or normogonadotrophic normoestrogenic anovulation, and polycystic ovary syndrome (PCOS) is the most common form of WHO II anovulatory infertility (Nahuis et al., 2009).

Clomiphene citrate is widely used as a first-line ovulation induction agent (Kousta et al., 1997). Systematic reviews and meta-analyses have shown that clomiphene citrate is indeed the best primary treatment option in therapy-naı¨ve women with PCOS (Brown et al., 2009; Moll et al., 2008).

Although 60–85% of women starting ovulation induction with clomiphene citrate will ovulate, only about 50% of these couples will conceive after six cycles (Neveu et al., 2007). When anovulatory women ovulate after clomiphene citrate but fail to conceive after six cycles, the subsequent treatment scenarios are diverse in current practice. Treatment may continue with clomiphene citrate for another six cycles or may switch to ovulation induction with gonadotrophins.

The ESHRE consensus guideline recommends that therapy naı¨ve women should start with six cycles of clomiphene citrate, but continuation until 12 cycles could be considered (Thessaloniki ESHRE/ASRM-Sponsored PCOS Consensus Workshop Group, 2008). This guideline also recommends that second-line treatment with gonadotrophins should not exceed six ovulatory cycles, but evidence underpinning this is not provided. Costs are considered but no definite recommendation is given concerning cost-effectiveness (Thessaloniki ESHRE/ASRM-Sponsored PCOS Consensus Workshop Group, 2008).

From a cost-effectiveness point of view, it is thus unclear whether and how long one should continue with clomiphene citrate or whether one should switch to ovulation induction with gonadotrophins or even IVF in women with PCOS that ovulate on clomiphene citrate, but failed to conceive after six cycles. Therefore this study evaluated the cost-effectiveness of different treatment scenarios in these women.

Materials and methods

A Markov decision tree was constructed for women with PCOS who ovulate on clomiphene citrate but fail to conceive after six ovulatory cycles. The crucial choice was whether to stick to clomiphene citrate or to switch to gonadotrophins, and when to proceed to IVF. To balance the choices for treatment, six scenarios were defined (Figure 1). Scenario 1 represented a scenario in which women received a maximum of three cycles of IVF. Scenario 2 consisted of an additional six cycles of clomiphene citrate and a maximum of three cycles of IVF in case no live birth was achieved. In scenario 3, six cycles of gonadotrophins were followed by a maximum of three cycles of IVF.

Scenario 4 started with 12 cycles of gonadotrophins and, in case of no live birth, a maximum of three cycles of IVF was given. Scenario 5 consisted of an additional six cycles of clomiphene citrate, six cycles of gonadotrophins and three cycles of IVF. In scenario 6, six additional cycles of clomiphenecitrate were followed by 12 cycles of gonadotrophins, and a maximum of three cycles of IVF in case of no live birth. For all six scenarios, the treatment costs and pregnancy probabilities were calculated over a 2-year period. Cycle length was set at 1 month. We used a time frame of 2 years because in this period all scenarios could be tested. Due to this short time frame, no discounting was applied. Dropout rate was not included in the main analysis so that the maximum possible gain could be shown, but the effect of dropout was tested in sensitivity analysis. Risk of ovarian hyperstimulation syndrome was not included because it is virtually eliminated in chronic low-dose step-up FSH regimens (Homburg, 2005).

The outcome was defined as live birth of at least one child.

Estimates for the base-case scenario and ranges for sensitivity analyses are summarized in Table 1 and were derived from peer-reviewed literature. The cost calculation was made according to the Dutch situation in the year 2010; hence costs were adjusted according to the consumer price index (Statistics the Netherlands, 2010). It was assumed that in this period no significant cost changes in the treatment protocol occurred except for inflation.

The model was built from a healthcare perspective. Details of computer simulation model Patient characteristics.The base-case calculation was centred on a woman with PCOS who fails to conceive after six ovulatory cycles with clomiphene citrate. As the average female age is 30 years in most studies on PCOS, this age was used as the reference point (Brown et al., 2009).

Live birth probabilities

Estimates of live birth probabilities per cycle for clomiphene citrate were based on a recent Cochrane review (Brown et al., 2009). The average probability of live birth was 12.2%. As several studies have shown that the highest live birth rate is seen in the first two ovulatory cycles with clomiphene citrate, this study assumed that live birth rate was 20% in the first two cycles and decreased to 7.5% per cycle in cycles 3 to 6, reflecting an average of 12.2% per cycle. If a live birth was not achieved in the first six cycles, live birth probabilities decline, but it is not known with what rate per cycle (Hammond et al., 1983; Kousta et al., 1997). Therefore, this analysis assumed a steady live birth rate of 7.5% per ovulatory cycle with clomiphene citrate after 6 months.

The live birth probabilities per cycle for gonadotrophins were also based on a recent review, that showed an average probability of live birth of 19.6% for gonadotrophins (Nahuis et al., 2010). For IVF, live birth probabilities were derived from a prospective cohort study performed between 2002 and 2004 (Lintsen et al., 2007; Moolenaar et al., 2011). Since the population was assumed to be age 30 at the beginning of the model, IVF live birth rates in the range of 30–32-year-old women were considered. The analysis included up to three IVF cycles because this is common practice in Europe (Andersen et al., 2007). The model assumed that natural conception did not occur.

Twin probability

The probability of conceiving a twin was included in the model, since probabilities differ across treatments. The probability of conceiving a twin with clomiphene citrate was set at 10%, with gonadotrophins at 25% and with IVF at 11% (Bayram et al., 2004; Eijkemans et al., 2003; Guzick, 2007; Kousta et al., 1997; NVOG, 2010; Schenker et al.,1981; Scialli, 1986; Thessaloniki ESHRE/ASRM-Sponsored PCOS Consensus Workshop Group, 2008).

Costs

The analysis was performed from a healthcare perspective; therefore, only direct medical costs were included: the costs of clomiphene citrate per cycle, gonadotrophins per cycle, cycle monitoring, IVF per cycle, and the costs of a singleton and twin pregnancy until 6 weeks post partum. Unit costs of clomiphene citrate and gonadotrophins were derived from the Dutch healthcare insurance board (Health Care Insurance Board, 2009). Costs per cycle were calculated according to dosages in the used reviews (Brown et al., 2009; Nahuis et al., 2010). Costs per IVF cycle were derived from the Dutch Umbrella study on fertility treatments (Merkus, 2006). The cost of a singleton and twin pregnancy until 6 weeks post partum were derived from a cost-analysis in the Netherlands (Lukassen et al., 2004).

All monetary units were converted into the equivalent of 2010 using the consumer price index (Statistics the Netherlands, 2010). It was assumed that in this period no significant cost changes in the treatment protocol occurred, except for inflation. The costs were €5 for a clomiphene citrate cycle, €334 for a gonadotrophin cycle, €410 for cycle monitoring, €3119 for an IVF cycle, and €2891 for a singleton pregnancy and €15,276 for a twin pregnancy (Table 1).



Outcomes

Cumulative live birth rates of at least one child were determined for each scenario, as were the estimated costs. From these values were computed the costs per live birth and the incremental cost-effectiveness ratio (ICER). The lowest cost option was used as a reference scenario since a no treatment scenario at this stage of the treatment is not realistic. The incremental cost-effectiveness ratio represents the extra costs per live birth between two scenarios. These costs are calculated by dividing the differences in costs by the difference in live birth rate of two scenarios.

Sensitivity analysis

To address the uncertainty regarding the assumptions, this work carried out one-way and probabilistic sensitivity analyses. In the base-case calculation, no discounting was applied; in one-way and multiway sensitivity analysis, the effects of different discounting rates for costs and effects were tested. All variables in the one-way sensitivity analysis were tested independently. Thresholds of all used variables were determined and the analysis included investigating if and when a variable changed the main conclusions. The threshold value represents the value of a variable above or below which another scenario is preferred. Since the modelassumed that no spontaneous live birth could occur, the sensitivity analysis tested what the effect would be if spontaneous live birth did occur. The effect of dropout as a continuous rate in all cycles was also tested.

In probabilistic sensitivity analysis, the uncertainty in each parameter is quantified in terms of a probability distribution of this parameter. For this analysis, distributions were fitted for all parameters in the model. Beta distributions could not be set; hence, normal distributions were fitted. The normal distributions were calculated according to the confidence interval from the study or by the plausible range provided by expert opinion (Briggs et al.,2006). For the probabilistic sensitivity analysis, 5000 iterations of 5000 women were performed. To visualize the probability of the optimal scenario based on the willingness to pay, cost-effectiveness acceptability curves were computed. The willingness to pay expresses society’s willingness to pay for a live birth. The ranges and values of all variables used in the sensitivity analyses are shown in Table 1.

To address the effect of time to pregnancy resulting in live birth, a threshold analysis was performed, in which a live birth was discounted per month and per year. The effect of discounting is, for example, that a live birth achieved now is valued more than a live birth achieved 1 year from now.

The analysis was performed using a computer-generated Markov model (TreeAge Pro 2009; Tree Age, Williamstown, MA, USA). IRB approval was not needed.

Results

Cumulative live birth rates after 2 years were 58% for scenario 1, 74% for scenario 2, 89% for scenario 3, 97% for scenario 4, 93% for scenario 5 and 98% for scenario 6. Costs per couple were €9518, €7530, €9711, €9764, €7651 and €7684, respectively (Table 2).



The lowest cost option was scenario 2. The extra cost for at least one live birth (ICER) for scenario 5 was €629 and forscenario 6 €630 compared with scenario 5. These scenarios dominated the other three scenarios (Figure 2).



Sensitivity analyses

One-way sensitivity analysis showed that, if live birth probabilities of gonadotrophins increased above 21% per cycle or if the probability of conceiving a twin after gonadotrophins decreased below 22.8%, scenario 6 would be the dominant scenario. Also, if the probability of conceiving a twin with IVF increased above 15%, scenario 6 would become dominant. If the dropout rate per cycle of clomiphene citrate, FSH or IVF was more than 1%, scenario 2 became the lowest cost option. Scenario 4 became the next best option, with an ICER of €123,959, compared with scenario 2.If the cost of cycle monitoring decreased below €357 or the cost of a twin pregnancy decreases below €13,840, scenario 6 would become the dominant scenario. The remaining variables were robust (i.e. no threshold could be found within the plausible ranges).



Time to pregnancy resulting in live birth

The impact of time to live birth was addressed by discounting a live birth per month and per year. If live birth was discounted, the conclusions based on the extra cost per live birth did not change. This work found a threshold for costs per live birth: if live birth was discounted more than 11% per month, scenarios 3 and 4 became cheaper than scenarios 5 and 6.

Probabilistic sensitivity analysis

The results of the probabilistic sensitivity analysis remained stable and did not alter the baseline results (Table 2). The cost-effectiveness acceptability frontier showed that if willingness to pay (expressing society’s willingness to pay for a live birth) was assumed to be less than €800, scenario 2 had the highest probability of being the most cost-effective scenario. If willingness to pay rose above €800 per extra live birth, scenario 6 had the highest probability of being cost-effective.

Discussion

This study evaluated the cost and effects of different treatment scenarios for women with PCOS who do not conceive after six ovulatory cycles with clomiphene citrate. The study showed that, after an initial failed number of six cycles, a continuation with clomiphene citrate for six cycles followed by three cycles of IVF (scenario 2) was the least expensive strategy. Continuation of clomiphene citrate for six cycles, followed by 6 or 12 cycles of gonadotrophins and three cycles of IVF in case of no live birth (scenarios 5 and 6) was more expensive but also generated a higher live birth rate (i.e. 93% and 98%). Whether this scenario is considered to be cost-effective depends on the willingness to pay per extra live birth.

It is counterintuitive that a long treatment protocol is more cost-effective than a short protocol. This is due to the low costs of clomiphene citrate; as more women conceive during treatment with clomiphene citrate, less women will proceed to the more expensive treatments with gonadotrophins and the even-more expensive IVF. However, not only costs but also time to pregnancy resulting in live birth is an important factor. This study discounted live birth to test the effect of time to pregnancy. Since discounting did not affect the conclusions, the issue of time to pregnancy had no influence from a cost-effectiveness point of view.

Most studies concerning pregnancy or live birth rates in women who failed to conceive after clomiphene citrate have focused on rates per cycle or on the first few cycles. No data of long-term treatment scenariosn are available, which are necessary for patients, clinicians, health economists and policy makers if they are to make an informed choice about treatment strategies for PCOS, since treatment may involve many cycles over a long period of time. It is, therefore, essential to quantify the assumptions used in this model in a randomized controlled trial.

The trend of only reporting the first or the first few cycles is found in the data of clomiphene citrate; there is little known about the live birth rates in consecutive cycles of clomiphene citrate. Two studies showed that live birth rates decline in subsequent cycles, but it is unclear with what rate, especially after six cycles (Hammond et al., 1983; Kousta et al., 1997). Therefore, the current work assumed a constant rate of live birth in the clomiphene citrate cycles, which was tested in sensitivity analyses. Irrespective of the probability of live birth after a clomiphene citrate cycle, the conclusions remain the same.

Also, the effectiveness of gonadotrophins in repeated cycles is uncertain, as most gonadotrophin studies only report on the first few cycles (Nahuis et al., 2010). If live birth probabilities per gonadotrophin cycle increase above 21%, a scenario with six cycles of clomiphene citrate, 12 cycles of gonadotrophins and three cycles of IVF would become the dominant scenario. If live birth per cycle declined, the conclusions would remain the same.

These analyses did not include intrauterine insemination (IUI) because there is no evidence in favour of IUI versus coitus in women with PCOS that do not conceive after six ovulatory cycles with clomiphene citrate. As IUI requires more hospital visits and is more expensive and burdensome for couples, applying IUI should only be considered if it results in significantly more pregnancies (Dickey et al., 1993; Kolibianakis et al., 2004; Randall and Templeton, 1991; Thessaloniki ESHRE/ASRM-Sponsored PCOS Consensus Workshop Group, 2008).

Another issue that needs to be addressed is that the guidelines in economic research recommend using the most cost-effective currently available alternative intervention as a comparator (Drummond and Jefferson, 1996). Since in ovulation induction, the guidelines do not agree on the most cost-effective intervention, this work decided to use the lowest cost option as a comparator. Also, the main outcome was the birth of at least one child. The analysis did include the costs for a singleton or twin pregnancy but did not consider twin pregnancy as resulting in two children, because this would favour scenarios in which twins occur.

Another bias of this study is the data used to calculate live birth after IVF is based on a prospective cohort that did not include live births after frozen embryo cycles. Therefore, the IVF live birth probabilities could be underestimated. For the sensitivity analysis, the base-case assumptions of IVF live birth probabilities were changed between zero and one. Within these ranges, scenario 2 remained the lowest cost option and scenarios 5 and 6 are cost-effective depending on the willingness to pay per extra live birth. Therefore, if the IVF live birth probabilities increase due to the not-included cryopreserved embryo transfer cycles, the conclusions would remain the same. This study is a long-term cost-effectiveness study comparing different treatment scenarios for women with PCOS who do not conceive after six ovulatory cycles. There is one other study reporting on long-term treatment scenarios for women with PCOS (Eijkemans et al., 2005), which compared a standard scenario of clomiphene citrate, followed by FSH and IVF to a scenario at which the choice between clomiphene citrate, FSH or IVF was dependant on age, body mass index, androgen concentrations and cycle duration.

The analysis was based on overall live birth rates and not, as in the current study, on cycle level and is therefore difficult to compare to this work. The study found that clomiphene citrate followed by FSH and IVF was an efficient treatment protocol for women younger than 30 and that for women older than 30, gonadotrophins could be omitted.

Since the costs of clomiphene citrate are low, one could also assume that continuation of clomiphene citrate for more than 12 cycles could be even more cost-effective. Since there is a consensus that it is best to limit a patient’s exposure to clomiphene citrate to 12 treatment cycles, as additional cycles may place the woman at increased risk of borderline ovarian tumours, this model had a maximum of 12 cycles (Rossing et al., 1994). The current analysis showed that in women who fail on clomiphene citrate, the continuation for six cycles of clomiphene citrate, followed by 6 or 12 cycles with gonadotrophins, followed by IVF were the most cost-effective scenarios. The sensitivity analysis found that if IVF cycle costs increased to €3240, if the cycle monitoring cost decreased to €357, if the gonadotrophin cost decreased below €281 or if the cost of a twin pregnancy decreased below €13,840, then six cycles of clomiphene citrate, with 12 cycles of gonadotrophins followed by IVF would become the dominant scenario (scenario 6). Also, if the probability of twins after gonadotrophins decreased below 22.8% or if the probability of twins after IVF increased to 15%, then scenario 6 would become the dominant scenario.

At present, because there is a lack of data, it might be difficult to convince women to continue clomiphene citrate again for another 6 months and thereafter to switch to 12 months of gonadotrophins. Consequently, according to current standards, scenario 6 might be unrealistic from a clinical perspective. It is these authors’ unreserved feeling that the purpose of science is to challenge existing practice with hypotheses and data. This article addresses the cost-effectiveness of continuing treatment of women with clomiphene citrate after six cycles and shows that this strategy could potentially be very cost-effective, which is an important finding in this era of health budget restraints.

Fertility care specialists are obliged to take costs and effects into consideration and to promote the best balance between the two. This is what doctors in the 21st century should be doing and in itself also supports the notion of patient centredness and shared decision making. Therefore, it is concluded that in women who fail to conceive after six cycles of clomiphene citrate, the most cost-effective treatment scenario seems to be a continuation of six cycles of clomiphene citrate, followed by 6 or 12 cycles of gonadotrophins and three cycles of IVF. Obviously, because of the uncertainty of the base-case assumptions, the effectiveness of clomiphene citrate and FSH in these long treatment protocols needs to be confirmed in randomized clinical trials.

Such a trial is currently being performed (NTR1449) (Consortium Study, 2013). Until such trials are completed, it is proposed that continuation of six cycles of clomiphene citrate after six cycles of clomiphene citrate, followed by gonadotrophins should be the treatment of choice recommended by the guidelines.