Cost-effectiveness modelling of recombinant FSH versus urinary FSH in assisted reproduction techniques in the UK

S. Daya1,8, W. Ledger2, J.P. Auray3, G. Duru3, K. Silverberg4, M. Wikland5, R. Bouzayen6, C.M. Howles7,* and A. Beresniak7,*

1 McMaster University, Hamilton Ontario, Canada, 2 University of Sheffield, The Jessop Hospital for Women, Sheffield, UK, 3 National Center of Scientific Research (CNRS), Villeurbanne, France, 4 Texas Fertility Center, Austin, Texas, USA, 5 Carlanderska Hospital, Gothenburg, Sweden, 6 IWK Grace Health Center, Halifax University, Canada and 7 Serono International SA, Geneva, Switzerland


    Abstract
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Acknowledgements
 References
 
BACKGROUND: The purpose of this study was to undertake an economic evaluation to compare the cost-effectiveness of recombinant (r)FSH with urinary (u)FSH for attaining clinical pregnancy with assisted reproduction. METHODS: Mathematical modelling was utilized incorporating a Markovian decision framework and a Monte Carlo simulation. Statistical representations of recurrent events over time were incorporated into a decision analysis involving fresh and frozen cycles in any sequence (after the first fresh embryo transfer cycle) over three successive assisted reproduction attempts. The mean values of transition probabilities were derived from randomized controlled clinical trials and published reports. The distributions of these transition probabilities were agreed upon by a panel of experts. Cost data for procedures and drugs were derived and validated according to the perspectives of the National Health Service and private clinics in the UK. RESULTS: The study involved 5000 Monte-Carlo simulations of treatment on a Markov cohort of 100 000 patients. The total number of pregnancies attained was significantly higher in the rFSH (40 575) compared with the uFSH (37 358) group. The cost per successful pregnancy was significantly lower for rFSH (£5906) compared with uFSH (£6060) and overall, fewer cycles of treatment were required with rFSH to achieve an ongoing pregnancy. The incremental cost-effectiveness ratio is £4148 for each additional clinical pregnancy with rFSH. CONCLUSIONS: In addition to the increased effectiveness of rFSH in ART, this study demonstrated that it is more cost-effective and more efficient than uFSH in attaining an ongoing pregnancy.

Key words: assisted reproduction/cost-effectiveness/mathematical modelling/recombinant FSH/urinary FSH


    Introduction
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Acknowledgements
 References
 
Approximately 10–15% of couples will experience infertility at some time during their reproductive years (Forti and Krausz, 1998Go). While infertility is not a direct threat to physical health, it has a significant effect on the psychological and social well-being of couples. During the last two decades, advances in the field of assisted reproductive technology (ART) have provided increased success and hope for couples for whom effective treatment was previously unavailable.

The increased numbers of couples requiring advanced fertility treatment and its attendant financial implications on healthcare systems, emphasizes the importance of demonstrating that clinical interventions are cost-effective. In the UK, for example, the provision of services using ART is complex, with only some regions providing health care through the National Health Service (NHS). Increasingly, the cost of ART is being assumed in large measure by the infertile couple. Therefore, it is important to ensure that treatments are both effective with respect to attaining pregnancy and cost-effective. This objective is particularly relevant when multiple treatment cycles are needed to achieve a successful pregnancy.

Ideally, efficacy analyses of ART treatment regimens should be based on prospective controlled clinical trials. The usefulness of clinical trials in determining the efficacy of a single drug or a procedure is widely recognized. However, their value is limited for determining the cost-effectiveness of complex situations such as a typical multi-step, multi-cycle ART intervention. To date, very few studies of clinical effectiveness have attempted to evaluate cost outcomes in such complex treatment situations. While retrospective surveys are informative, they are fraught with bias and are inadequate for pharmacoeconomic analyses because complete data for all subjects are usually unavailable. Furthermore, a randomized trial to evaluate cost-effectiveness in such a complex treatment as ART would require considerable resources, an enormous number of patients and a lengthy follow-up to provide any meaningful information.

An effective method of overcoming these limitations to undertaking a pharmacoeconomic assessment of ART treatment, is to employ modelling and computer-simulation. Modelling is a powerful method for repetitively simulating and testing the conditions and outcomes of a complex treatment programme. It allows the use of data from several sources such as randomized controlled trials, national IVF registries and expert opinions. Computer-based models are constructed with the use of established mathematical simulation techniques. Markov modelling is also increasingly used in economic evaluations of other health related issues (Briggs and Sculpher, 1998Go). In a Markov model, patients are at any time in a specific `health state' that indicates their position in their treatment cycle, e.g. ovarian stimulation, oocyte retrieval, fertilization and so on. Patients progress along an ART treatment cycle through a series of health states, each with an associated probability (a `transition' probability) for that particular outcome, during a complete ART cycle, e.g. failed fertilization. Transition probabilities for all stages of a treatment cycle (e.g. cancelled ovum retrieval, successful recovery of oocytes, fertilization of oocytes, etc.) can be obtained from available clinical trials. This model therefore provides a `natural' framework for constructing simulations of temporally changing situations, such as one encounters in ART, while also assessing the economic and clinical effects of the intervention being studied.

Markov modelling has been used to determine the cost-effectiveness of medical treatments in a variety of clinical conditions (Col et al., 1997Go; Nuitjten et al., 1998; Palmer et al., 1998Go; Mantovani et al., 1999Go; van Loon et al., 2000Go). The cost-effectiveness of recombinant (r)FSH has been compared with urinary (u)FSH in the treatment of infertile women in Greece (van Loon et al., 2000Go) and in Italy (Mantovani et al., 1999Go), and in both of these studies it was demonstrated that the use of rFSH was most cost-beneficial. However, the results of these studies have limited utility because they were based on data from a small number of clinical trials and the models contained relatively few health states. Additionally, although the mean transition probabilities of the models were agreed upon by a panel of experts, critical information on the distribution of these probabilities was not provided. Such omissions result in uncertainty about the interpretation of such modelling outcomes and, despite the use of sensitivity analyses, the lack of confidence limits around the estimates of the outcomes makes it impossible to determine whether differences between treatments are statistically significant. The purpose of the present study, therefore, was to develop a detailed model of ART that could incorporate a large number of health care states, thereby providing outcome estimates (and their ranges) in order to effectively compare the cost of infertility therapy with rFSH (Gonal F®, Serono) and uFSH (Metrodin HP®, Serono) in the UK.


    Materials and methods
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Acknowledgements
 References
 
The Markov model was used as the framework for the computer simulation of the ART treatment programme, in which the efficacy and costs of rFSH and uFSH for ovarian stimulation were compared. A customized computer program was written in PASCAL language specifically for this study. We chose to use the fully customized software in PASCAL language in order for the analysis to be more transparent and to be able to test the random number generator. Currently available commercial software packages are not as powerful when it is necessary to manage a very large number of health states and the quality and characteristics of the random number generator are not disclosed or described.

The general architecture of this Markov model for ART is shown in Figure 1Go. The first cycle is a complete treatment cycle (CTC) in which fresh embryos are transferred. Should the first cycle not result in pregnancy, a second cycle could involve either another CTC with fresh embryo transfer or a non-stimulated cycle with frozen embryo transfer. Similarly, if a third attempt is needed, the third cycle could also involve either a CTC with fresh embryos or a frozen embryo transfer. Figure 1Go represents, albeit in a simplified fashion, the possible CTC and frozen embryo transfer combinations that can occur with up to three cycles of embryo transfer in ART after rFSH stimulation. An identical representation exists for ovarian stimulation with uFSH.



View larger version (28K):
[in this window]
[in a new window]
 
Figure 1. The general architecture of the Markov model in ART illustrating the various pathways available to a patient undergoing up to three treatment cycles with embryo transfer with assisted reproduction techniques using a single stimulation treatment option with either rFSH or uFSH. The insert details some of the decision pathways included in the Markov model during a single treatment cycle.

 
Considering Figure 1Go in greater detail, a CTC is composed of 61 health states, whereas a frozen embryo transfer is composed of 13 health states. The health states for CTC are obtained from the decision tree, commencing with ovarian stimulation and ending with a successful pregnancy, a repeated cycle or discontinuation of treatment. After the first health state (i.e. receiving ovarian stimulation) there are 30 health states one can encounter along this pathway for each of rFSH and uFSH. These health states are: ovum retrieval (yes or no); oocytes recovered (yes or no); IVF or intracytoplasmic sperm injection (ICSI; the subsequent health states are identical for each of IVF and ICSI); fertilization (yes or no); embryo transfer (yes or no); pregnancy (yes or no); if pregnancy is achieved, then ovarian hyperstimulation syndrome (yes or no); and for each of these last two states, the pregnancy could progress normally or end as a miscarriage.

There are six health states for each of rFSH and uFSH for frozen embryo transfer after the first health state, which is frozen embryo thaw. The subsequent health states are: embryo survival (yes or no); pregnancy (yes or no); and ongoing pregnancy or miscarriage. Thus, the total number of health states in this study of three cycles is:

CTC + (CTC + frozen embryo transfer) + (CTC + frozen embryo transfer + CTC + frozen embryo transfer) = 61 + (61 + 13) + (61 + 13 + 61 + 13) = 283. The total number of health states considered in this model reflects its comprehensiveness.

The insert in Figure 1Go illustrates the decision pathway that includes the 61 health states involved in a CTC. Similar decision pathways (decision trees), which for reasons of brevity are not illustrated, exist for cycles two and three, further demonstrating the complex nature of this model.

Transition probabilities for each health state were estimated for both rFSH and uFSH protocols. The means of these probabilities were calculated from data derived from several sources, including a meta-analysis of randomized controlled trials comparing the two gonadotrophins (Daya and Gunby, 1999Go), national IVF/ICSI registries (FIVNAT, 1999Go; HFEA, 1999Go; SART `97, 1999Go) and clinical expert consensus opinion when published data were unavailable. These probabilities are shown in Table IGo.


View this table:
[in this window]
[in a new window]
 
Table I. Transition probabilities
 
Each transition probability was assigned a range based on the expert opinion of clinicians experienced with clinical trials and epidemiology, and incorporating the clinical perspective of the UK. The expert panel was composed of five practising physicians considered as experienced opinion leaders in reproductive health. The role of the panel was to construct the treatment model according to medical practices and validate the resulting structure, validate the sources of the transition probabilities and express the acceptable range of variability around each transition probability. The value for the range was selected to include 95% of patients, i.e. the mean ± 2SD. For example, the mean value of the transition probability for the health state `cancelled ovum retrieval' derived from randomized controlled trials (Daya and Gunby 1999Go) was 4.4% for rFSH and 6.4% for uFSH. The expert panel agreed that the probability for this health state in 95% of patients would be 2–8% (range = 6, i.e. 0.06). In classical statistics, the standard deviation is calculated from the range by the following formula: SD = range ÷ (2x1.96). In the example then of `cancelled ovum retrieval', the respective means and standard deviations were 4.4 ± 0.0153 for rFSH and 6.4 ± 0.0153 for uFSH. As with clinical studies, the availability of an estimate and its standard deviation permits significance testing and derivation of P-values for outcomes. The statistical test used was the Student's t-test.

Figure 2Go illustrates the distribution profile of the transition probabilities for the health state `cancelled ovum retrieval' for both rFSH and uFSH. The shape of both distributions is Gaussian (normal curve) and is centred about the mean value, with a dispersion that reflects the calculated SD. The expert clinical panel validated the normality of the distributions. Furthermore, mathematical estimations of proportions are always normal, thereby allowing the construction of confidence intervals. The mean ± SD for all the transition probabilities are shown in Table IGo.



View larger version (16K):
[in this window]
[in a new window]
 
Figure 2. Distributions of the transition probabilities of the health state `cancelled ovum retrieval' for both rFSH and uFSH.

 
Using the probability range (mean ± 2SD) for each health state, data for an individual patient was randomly generated and entered into the computer model using the Monte Carlo method (Doubilet et al., 1985Go), which is widely used in situations where multiple outcomes are evaluated. Briefly, Monte Carlo methods use a random number generator to generate values at random, i.e. without bias, for each health state so that the dispersion of the outcome estimates could be derived. The randomly assigned values fall within the range of values agreed upon for 95% of the population. The benefit of such methods is that confidence intervals around the outcome estimates of interest (pregnancy rate, cost/pregnancy and number of cycles) can be obtained (Auray et al., 1996Go).

As with any other model, a number of assumptions were made in the construction of this model. These assumptions were: (i) complete therapy was defined as participation in up to three cycles with embryo transfer, each cycle constituting a treatment cycle; (ii) the transition probabilities for each treatment cycle remained unchanged; (iii) the distribution of all transition probabilities was normal; (iv) there were no spontaneous, treatment-independent pregnancies; (v) the main outcome of the study was clinical pregnancy (intrauterine pregnancy at 12 weeks gestation confirmed by ultrasound); (vi) the gonadotrophin (i.e. rFSH or uFSH) used for ovarian stimulation was randomly assigned and remained the same in subsequent treatment cycles.

A comprehensive cost analysis accounting for all health states in the ART model was undertaken and included costs incurred for ovarian stimulation, monitoring, oocyte retrieval, laboratory procedures, luteal phase support, pregnancy determination and cryopreservation. The costs considered in this study were for the year 2000 and were obtained from the NHS clinic tariff from one clinic located in Sheffield, UK, which is a medium–low price clinic in the provision of ART services. The price of drugs are the list prices for the year 2000 (Table IIGo).


View this table:
[in this window]
[in a new window]
 
Table II. Charge categories considered in the model
 
The cost-effectiveness analysis comparing rFSH with uFSH was performed with ongoing pregnancy at 12 weeks gestation as the primary outcome from up to three cycles of treatment. To perform the analysis, a virtual population of 100 000 patients (the `Markov cohort') was `treated' in the computer simulation of ART treatment in each of 5000 Monte Carlo simulations. The cohort was composed of women with infertility, aged 18–40 years and with ovulatory cycles undergoing ovarian stimulation with FSH for IVF or ICSI. These large numbers of patients and simulations provided a high degree of statistical accuracy and allowed confidence limits around the outcome estimates to be generated with precision.


    Results
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Acknowledgements
 References
 
This model utilized clinical data for rFSH in general, while the price of rFSH was the listed price of Gonal-F® in the UK. Since subgroup analysis in the meta-analysis of rFSH versus uFSH showed a statistically significantly higher odds ratios for Gonal-F® (follitropin-alpha) versus uFSH, but not for follitropin-beta versus uFSH (for ongoing pregnancy) (Daya and Gunby, 1999Go), the cost-effectiveness of Gonal-F® could be greater. This conservative approach therefore generated outcomes that are specific for Gonal-F®, but possibly underestimated its comparative cost-effectiveness because we underestimated the clinical effectiveness data for Gonal-F®, but used the specific list price for Gonal-F®. The results obtained from running Monte Carlo simulations on the Markov cohort of patients are summarized in Table IIIGo.


View this table:
[in this window]
[in a new window]
 
Table III. Model outcomes when comparing rFSH and uFSH in 5000 Monte Carlo simulations with 100 000 patients per group
 
Effectiveness
The total number of ongoing pregnancies at 12 weeks gestation was 40 575 (95% confidence interval ± 2648) with rFSH (Gonal-F®) and 37 358 (± 2922) with uFSH (Figure 3Go and Table IIIGo). This difference of 3217 pregnancies in favour of rFSH was statistically significant (P < 0.0001).



View larger version (10K):
[in this window]
[in a new window]
 
Figure 3. The distribution of on-going pregnancy rate at 12 weeks.

 
Total costs
The total cost for one patient entering the model was £2393.15 (± 29.04) with rFSH (Gonal-F®) and £2259.69 (± 31.8) with uFSH (Table IIIGo).

Cost effectiveness
The cost per ongoing pregnancy was £5906 (± 455) with rFSH (Gonal-F®) and £6060 (± 547) with uFSH (Table IIIGo). This difference of £154 between treatments was statistically significant in favour of rFSH (P < 0.0001) and represented a reduction in overall treatment cost of 2.5%. The cost-effectiveness distribution for each FSH preparation can be seen in Figure 4Go.



View larger version (11K):
[in this window]
[in a new window]
 
Figure 4. The comparative cost-effectiveness distributions for rFSH and uFSH.

 
In addition to these average cost-effectiveness ratios, it is also possible to calculate the incremental cost-effectiveness ratio by dividing the difference in total costs by the difference in effectiveness. In this study, the incremental cost with rFSH was £239 315 074 – £225 969 395 = £13 345 679, and the incremental outcome (effectiveness) was 40 575 – 37 358 = 3217 clinical pregnancies. Thus, the incremental cost-effectiveness ratio was £4148 for each additional clinical pregnancy.

Mean number of cycles per success
This model also allows one to calculate the total number of treatment cycles needed to obtain a successful outcome (i.e. ongoing pregnancy). Taking into consideration the numbers of cycles terminated before oocyte retrieval, and taking fresh and frozen embryo transfer cycles together, the mean number of cycles required to achieve one ongoing pregnancy was found to be 4.49 with rFSH and 4.80 with uFSH (Table IIIGo).

When these data were further subdivided into fresh and frozen embryo transfer cycles, it was found that with rFSH the mean number of fresh embryo transfer cycles for one successful outcome was 3.83 compared with 4.29 with uFSH, and for frozen embryo transfer cycles the mean numbers were similar at 0.66 and 0.51 respectively.


    Discussion
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Acknowledgements
 References
 
Although the cost per ampoule is higher for rFSH, the increased effectiveness of rFSH in ART results in it being significantly more cost-effective than uFSH in achieving an ongoing pregnancy. Furthermore, the average number of cycles required to achieve an ongoing pregnancy was much lower with rFSH than with uFSH.

Ongoing pregnancy was chosen as the outcome of interest because the meta-analysis used to derive clinical probabilities for the model contained no information on live births or premature delivery, but used clinical pregnancy (at 12 weeks) gestation as the primary endpoint. Furthermore, the expert panel validated the fact that there was no evidence to suggest that the choice of recombinant or urinary gonadotrophin would have an effect on obstetrical outcomes.

One reason for the superiority of rFSH is its higher level of clinical response. The retrieval of a higher number of mature oocytes when rFSH was used for ovarian stimulation compared with uFSH has been reported in many studies (Fisch et al., 1995Go; Out et al., 1995Go, 1997Go; recombinant human FSH Study Group, 1995Go; Bergh et al., 1997Go; Manassiev et al., 1997Go; Khalaf et al., 1998Go; Franco et al., 1999Go; Frydman et al., 2000Go; Lenton et al., 2000Go). Overall, these comparative studies demonstrated that rFSH has statistically significant advantages in terms of efficacy, resulting in higher numbers of follicles aspirated and oocytes retrieved, even though the daily dose of gonadotrophins administered was lower and the treatment period shorter (Bergh et al., 1997Go; Manassiev et al., 1997Go; Khalaf et al., 1998Go; Franco et al., 1999Go; Lenton et al., 2000Go). Consequently, the total consumption of gonadotrophins was much lower with rFSH than with uFSH. As a result, the likelihood of having more embryos when rFSH is used is much greater, thereby providing more opportunities for cryopreservation of surplus embryos, which can be used for subsequent, less costly frozen embryo transfer cycles. Additionally, a meta-analysis evaluating the clinical outcomes when the two gonadotrophins were compared demonstrated a significantly higher pregnancy rate per cycle started with the use of rFSH (Daya and Gunby, 1999Go).

From an economic evaluation point of view, rFSH was clearly more cost-effective than uFSH, but did not dominate this alternative, because even though its effectiveness was higher, its cost was also higher. The incremental cost-effectiveness analysis demonstrated that for each additional clinical pregnancy, the overall cost increment was £4148. However, there would be savings generated from using rFSH because fewer numbers of cycles would be required to attain one pregnancy compared with uFSH. Furthermore, since social costs (non-medical costs, such as patients' time off work, travel, parking and so on) were not incorporated into the model, the cost differences might, in reality, be greater. The shorter duration of treatment with rFSH and the requirement of fewer cycles to achieve pregnancy might have had a greater effect in reducing total costs further, suggesting that the cost-effectiveness of rFSH may be of an even greater magnitude.

The use of clinical tariff from the medium–low price clinic in Sheffield, UK is a conservative approach since a more expensive clinic with higher tariffs would result in a increased difference between rFSH and uFSH due to the fact that use of urinary gonadotrophins requires more cycles to achieve success as a consequence of lower pregnancy rates.

The robustness of the pharmacoeconomic model that generated these findings deserves some discussion. The general robustness of a model depends on both the robustness of the structure (the model) and the robustness of the data. The structure of the model is robust because all assumptions were validated and considered as acceptable by the clinical expert panel. The data used are robust because they are based on methodologically sound studies (randomized controlled trials) and because we performed a high number of Monte Carlo simulations that reduced the magnitude of the standard deviations. Current commercial software packages are not as powerful when it is necessary to manage a very large number of health states and the quality and characteristics of the random number generator is not disclosed/described. Furthermore, our choice of the random number generator, which is necessary for the Monte Carlo method, guarantees that there will be no selection bias from repetition because the sequences of random numbers used are unique.

The potential limitation of the model is based on the assumption that patients proceeding through second and/or third cycles would be treated with the same drug combinations as in the first cycle (a situation that may vary in practice) and the agreement by the expert panel that there were no reasons to change the value of transition probabilities for the second and third cycles because there is no evidence to suggest that this assumption is invalid.

In other studies of the relative cost-effectiveness of rFSH and uFSH in ART, classical sensitivity analyses have been performed to attempt to account for the variance of inputted data and test the robustness of the results (Van Loon et al., 2000Go). Such sensitivity analyses consist of sequentially altering the value of an important parameter, e.g. a mean transition probability, to observe the effect of this change on the overall outcomes. Such analyses are usually restricted to the effects produced by altering the values of only a few key parameters. Consequently, the choice of the examined parameter and the degree to which the parameter is altered are subject to selection bias. A further limitation is that parameters are altered one at a time to test the effect on the results, whereas in reality, there is variance around all these variables simultaneously. These limitations of classical sensitivity analyses consequently led to the development of probability sensitivity analysis using the Monte Carlo technique previously described. In the present study, probability distributions were used at every decision point to account simultaneously for variation in all the variables. The Monte Carlo method uses these distributions to produce confidence intervals around the estimates of the outcomes and, therefore, provides much more precise and useful information than do classical sensitivity analyses.

It is important to note that the findings of this cost-effectiveness study are valid only when considering rFSH and uFSH within the scenario of treatment of infertility with ART. Simple substitution of the cost of gonadotrophins other than those used in this study would fail to derive an accurate estimate of the cost-effectiveness of the substituted drug. This is because the present model utilized product specific data, i.e. transition probabilities, and in order to test other drugs reliably the model would have to be `reprogrammed' using the correct distribution probabilities and costs associated with the substituted drugs. Furthermore, although the mean figures used in the model come from a variety of sources, the use of the panel of experts who were able to review the range of costs in the UK implies that the distribution probabilities are finely honed to the situation in the UK. Similar exercises using this model can be undertaken to evaluate the cost-effectiveness of the gonadotrophins in ART in other countries.

In conclusion, a robust, statistically powerful, pharmacoeconomic model of the procedures used in the conduct of a typical infertility treatment programme using ART in the UK has been constructed. Running Monte Carlo simulations on a Markov cohort of patients has clearly demonstrated that use of rFSH for ovarian stimulation in ART and embryo transfer is significantly more cost-effective than uFSH. This model is currently being developed to assess the cost-effectiveness of other drugs used in ART.


    Acknowledgements
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Acknowledgements
 References
 
We are grateful to Dr Mark O'Brien, Serono International, for making available data from trials and for defining charge categories.


    Notes
 
8 To whom correspondence should be addressed at: Department of Obstetrics and Gynecology, McMaster University,1200 Main Street West, Hamilton, Ontario, Canada, L8N 3Z5. E-mail: dayas{at}mcmaster.ca Back

* Colin Howles is Vice President, Reproductive Endocrinology, Serono, International SA and Ariel Beresniak is Corporate Director Pharmacoeconomics, Serono International, SA. Back

Submitted on December 29, 2000; July 10, 2001


    References
 Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 Acknowledgements
 References
 
Auray, J.P., Beresniak, A., Claveranne, J.P. and Duru G. (1996) Dictionnaire commenté d'Economie de la santé. Masson, p. 289.

Bergh, C., Howles, C., Borg, K. et al. (1997) Recombinant human follicle stimulating hormone (r-FSH, Gonal-F) versus highly purified urinary FSH (Metrodin HP): results of a randomized comparative study in women undergoing assisted reproductive techniques. Hum. Reprod., 12, 2133–2139.[Abstract]

Briggs, A. and Sculpher, M. (1998) An introduction to Markov modelling for economic evaluation. Pharmacoeconomics, 13, 397–409.[ISI][Medline]

Col, N., Eckman, M., Karas, R. et al. (1997) Patient-specific decisions about hormone replacement therapy in postmenopausal women. JAMA, 277, 1140–1147.[Abstract]

Daya, S. and Gunby, J. (1999) Recombinant versus urinary follicle stimulating hormone for ovarian stimulation in assisted reproduction. Hum. Reprod. 14, 2207–2215.[Abstract/Free Full Text]

Doubilet, P., Begg, C.B., Weinstein, M.C. et al. (1985) Probabilistic sensitivity analysis using Monte Carlo simulation: a practical approach. Med. Decis. Making, 5, 157.[Medline]

Fisch, B., Avrech, O., Pinkas, H. et al. (1995) Superovulation before IVF by recombinant versus urinary human FSH combined with a long GnRH analog protocol: a comparative study. J. Assist. Reprod. Genet., 12, 26–31.[ISI][Medline]

FIVNAT (1999) Dossier FIVNAT 1999.

Forti, G. and Krausz, C. (1998) Evaluation and treatment of the infertile couple. J. Clin. Endocrinol. Metab., 83, 4177–4189.[Free Full Text]

Franco, J., Baruffi, R., Coehlo, J. et al. (1999) A prospective and randomized study of ovarian stimulation for ICSI with recombinant FSH versus highly purified FSH (abstract). 11th World Congress on in vitro Fertilization and Human Reproduction Genetics. Sydney, NSW, Australia, 9–14 May 1999.

Frydman, R., Howles, C. and Truong, F. (2000) A double-blind, randomized study to compare recombinant follicle stimulating hormone (FSH; Gonal-F®) with highly purified urinary FSH (Metrodin HP®) in women undergoing assisted reproductive techniques including intracytoplasmic sperm injection. The French Multicentre Trialists. Hum. Reprod. 15, 520–525.[Abstract/Free Full Text]

HFEA (1999) The Human Fertilisation and Embryology Authority Eighth Annual Report 1999. Her Majesty's Stationery Office, London.

Khalaf, Y., Taylor A., Pettigrew, R. et al. (1998) The clinical efficacy of recombinant follicle stimulating hormone to the highly purified urinary preparation (abstract). Hum. Reprod., 13 (Abstract Bk 1), p.191.

Lenton, E., Soltan, A., Hewitt J. et al ( 2000 ) Induction of superovulation in women undergoing assisted reproductive techniques: recombinant human follicle stimulating hormone (rFSH; Gonal-F®) versus highly purified urinary FSH (Metrodin HP®) Hum. Reprod. 15, 1021–1027.[Abstract/Free Full Text]

Manassiev, N., Davies, W., Leonard, T. et al. (1997) Initial results from the comparison of recombinant FSH and urinary FSH in an IVF programme (abstract). Hum. Reprod. 12 (Abstract Bk 1) , p.265.

Mantovani, L., Belisari, A., Szucs, T. (1999) Pharmaco-economic aspects of in-vitro fertilization in Italy. Hum. Reprod. 14, 953–958.[Abstract/Free Full Text]

Nuijten, M., Hadjadjeba, L., Evans, C. and van den Berg, J. (1998) Cost effectiveness of fluvoxamine in the treatment of recurrent depression in France. Pharmacoeconomics, 14, 433–445.[ISI][Medline]

Out, H., Mannaerts, B., Driessen, S. et al. (1995) A prospective, randomized assessor-blind, multicentre study comparing recombinant and urinary follicle-stimulating hormone (Puregon® versus Metrodin®) in in-vitro fertilization. Hum. Reprod., 10, 2534–2540.[Abstract]

Out, H., Driessen, S., Mannaerts, B. et al. (1997) Recombinant follicle-stimulating hormone (follitropin beta, Puregon®) yields higher pregnancy rates in in vitro fertilization than urinary gonadotrophins. Fertil. Steril., 68, 138–142.[ISI][Medline]

Palmer, C., Revicki, D., Genduso, L. et al. (1998) A cost-effectiveness clinical decision analysis model for schizophrenia. Am. J. Managed Care, 4, 345–355.[ISI][Medline]

Recombinant Human FSH Study Group (1995) Clinical assessment of recombinant human follicle-stimulating hormone in stimulating ovarian follicular development before in vitro fertilization. Fertil. Steril., 63, 77–86.[ISI][Medline]

SART `97 (1999). Assisted reproductive technology in the United States: 1990 results generated from the American Society for Reproductive Medicine/Society for Assisted Reproductive Technology Registry. Fertil. Steril., 71, 789–807.

Van Loon, J., Liaropoulos, L., Mousiama, T. (2000) Economic evaluation of a recombinant follicle-stimulating hormone (Follitropin Beta, Puregon®) in infertile women undergoing in vitro fertilisation in Greece. Clin. Pharmacoeconomics, 19, 201–211.

accepted on September 22, 2001.