Presentation Transcript

A Markov Model of the Cost Effectiveness of Human-Derived FSH versus Recombinant FSH Using Comparative Clinical Trial Data
presented by Hind Hatoum, PhD

view Dr. Hatoum's presentation  
view interview video in Windows Media or RealMedia
read the written transcript of the interview
read Dr. Hatoum's curriculum vitae
access the printable version of this page

This work is a collaborative effort among Ferring, Dr. Keye and Dr. Marrs and also my colleagues at the University of Illinois in Chicago. 

What we have done is, first, let’s talk a little bit about why do we need to have economics in infertility. Some useful statistics. You have one in six couples who experience problems in getting pregnant at some point in their reproductive lives. Assisted reproductive technologies have become the widely-used method to correct these problems and the health economic implications of using ARTs are not that well-established.

Understanding the economics of FSH selection - there is a particular interest in understanding what the economic implications are of selecting one FSH product over another and establishing the cost-effectiveness of competing FSH products is not feasible in clinical trials because we need a large staff size in order to come up with economic advantages or disadvantages for one agent versus another. So Markov modeling is one way of analyzing clinical trial findings on any existing clinical trial when you are trying to comparative the economic impact of one treatment versus another.

What were our objectives? We wanted to create a Markov model and we wanted to use comparative and I am emphatic about the fact that we used existing comparative clinical trials which compared human-derived FSH and recombinants, Bravelle versus Follistim. These clinical trial databases were the basis for the FDA approval for Bravelle for IVF procedures. The Markov model was used to examine the cost and effectiveness of the two FSH products and we also wanted to conduct a sensitivity analysis using the Monte Carlo simulation and I will talk a little bit about that because that is really very critical to understanding the impact of the findings.

We used an off-the-shelf Markov program, Data Pro by Treeage, and we used that to examine both the cost and effectiveness of both Bravelle versus Follistim. Drug costs for either one of the two agents were used from actual data used in the clinical trials, so there was no assumption as to how many vials, how many IUs the patients used, it was exactly what was used in the trials and all of the costs we used we credit Dr. Silverberg and colleagues when they did this study dealing with IVF and follicle-stimulating hormones. The study was conducted last year, so we inflated the costs to reflect 2003 dollars.

What does the Markov model mean? It sounds like it’s a German or Russian name and a lot of people have asked where did it originate? I could not find it exactly, but someone mentioned the railroad industry but just so we will understand its application in health economics - it is an accepted method for determining the difference in the health economic impact of two or more agents, it can be for devices, it can be for surgery, whatever it is you have to do, you can use the Markov model to do that. What the Markov model does is replicate how patients progress through the different stages in the IVF cycle. We call that transition probability. So you are talking from the oocyte retrieval until the female achieves continuing pregnancy. That is what we enter into the model. So it replicates how the patients go through that process to achieve the endpoint which is, in the clinical trial for follicle-stimulating hormone, actually achieving clinical pregnancy. The different stages of the IVF cycle from oocyte retrieval is considered a health state in the model. When a patient does not progress to the next stage in the IVF cycle, she is bumped into a brand new cycle and that is all taken into account when we calculate the economic consequences of going through the IVF cycle, one versus two or versus three.

Why select the Markov model? A lot of people ask why do the Markov model, why not cost minimization? There are a lot of different techniques in health economics that we use for IVF profit and procedures that are available to health economics other than prospectively designing a study that takes into account the health economics, in fact the full treatment versus a lot of issues that are either prohibitive in terms of cost and time to achieve and in the sample I needed to show different or equivalent. The Markov model takes into account cost for each step in the IVF cycle and it is an acknowledgement that there is more in terms of resource consumption than the cost of FSH used. You always lose sight of the fact that FSH is one component in the IVF procedure and is not everything. Anyone who has done any research or review of the literature, you find there is so much variability in the patients’ achieving pregnancy that not everything can be attributed to FSH used or the FSH that is discredited for not achieving that pregnancy. 

The probability of outcome between two FSHs are identical at each step. If they are, then the Markov model will turn into a cost minimization study. The fact that the probability of outcome between Bravelle and Follistim when the two comparative clinical trials were conducted were not identical at all steps argued for the use of the Markov model. You really don’t know when you are undertaking the work, you are taking the medical trials, we did not do any assumptions. Health analysts are wonderful in assuming anything that we don’t know, we say, let’s convene a group of experts and ask them, what is the probability of something taking place and then we use it. In our Markov, we take pride that we did not change any one of the transition probabilities for the patients achieving each phase of the IVF process. We did not change it, so whatever came from the clinical trial database, that is what we used. The fact that they are not identical, at ooctye retrieval, you have some patients bumped from one treatment arm versus another, and the cost of moving the patient from one stage of the cycle to the other, there are health economic reasonable assumptions that have to be kept in order to come up with the overall economic impact of one treatment versus the other. Cost minimization is, when we first started in the field, what we used and that is to say that the two agents are equivalent in terms of the effectiveness and let’s see what the cost difference in the price will do in terms of coming up with an economic advantage or not. It is only appropriate to use when the two agents have identical outcomes. In simplest indications, and certainly all of you would know, that IVF procedures are not one the simplest procedures to do.

Now, I talked a little bit about building the Markov model. Let me repeat a little bit. The transition probabilities from one health state to a subsequent one came from the clinical trial that Dr Dickey and colleagues have published in 2003. Effectiveness, as I mentioned earlier, was measured as an on-going pregnancy defined as established clinical pregnancy verified by ultrasound. Since we deem the limitation of doing the Markov model from clinical trial database is the fact that you do not have a sample size large enough to inflict a true difference between the two treatments. What you do is, you take a cohort of patients and you enter them into the Monte Carlo simulation program and the program will choose, at random, for each one of the females who enters in that cohort, chooses at random whether she will end up at oocyte retrieval or whether she is going to continue until she achieves on-going pregnancy. That is really based on the variation that you have between the two distributions of the two treatments. It is a little bit complex and I hope that I am trying to make it a little bit, even for us health economists, we have a lot of difficulty appreciating and understanding the Markov model, but it is one of the techniques that is really beautiful in terms of its ability to take a set of data from two agents or two procedures and come up with an economic argument.

The Monte Carlo simulations were tested for robustness; that is, how many changes do you need in any one of the input variables, the health state you have taken and you have most patients from one stage in the cycle to the other, how much variation do you need to have in order to alter your findings appreciably. That is really the idea to doing this. It’s kind of like being able to expand yourself in size in order to see how much or how robust your findings are. Monte Carlo simulation was based on the distributions for the transition probabilities. When I say transition probabilities, it sounds really like a fancy term, but all that it is is how many patients progress percentage-wise from, let’s say, oocytes retrieval to the next stage because these are now the new cohorts that you are dealing with at that stage. So what percentage of those patients transitioned, if you will, to the next health state leading up to a clinical pregnancy. Important statistics that we have used, we vary in the margin and that is not really our doing. It is picked at random, the computer program picks at random different transition probabilities for each of the patients entering into the model to see what exactly you come up with. Then what we did, in order to verify because the trials were done in rather younger females and not too many of them had had IVF procedures before and we do know that the more IVF procedures a female goes through, the less likely the chances of her becoming pregnant. It is not a function of IVF, it is a function of the fact that there are no issues, and most likely the IVF cycle would successful at the first stage. So, in order to verify that, what we have done with the applicable international levels, we used the results of Meldrum and colleagues in 1998, a national registry of 54 centres in the US, feeding into the database in order to come up with, on average, the chances of a female getting pregnant after one IVF cycle. It was about 37%. So we used it in order to verify what would happen if we had used the other statistics and not the females who are most likely to get pregnant.

This is a fancy way of showing the equivalent of what we used to call a long time ago a decision tree, except when you have a decision tree, it does not talk about too many likely outcomes. If you look through the different stages, at each one of these stages, the patient has a chance of either transitioning to the next health state or she is bumped out of the cycle and she has to go back and repeat the cycle. If you will, it is one cycle of the model. What I am going to present to you is the results from exposing the two treatment outcomes into one cycle versus modeling the three cycles of IVF.

This is just to let you know, I do not expect you and I am not going to quiz you at the end of the session about the actual probabilities, but just to let you know and emphasize the point, that we have not changed any transition probability. Whatever the outcome in the comparative clinical style database, this is what went into the Markov model index that we used. These are the transition probabilities at each one of the stages for the two treatments, Bravelle and Follistim. 

As I mentioned, the cost is inflated from Silverberg and colleagues to 2003 dollars and each cost in that tree you saw earlier, we include the patients who come to their cycle and drop out, we include the cost of going through that stage of the cycle. If the female progresses to the next stage, we include the cost of the second stage in that cycle and that is why it is so important to use the Markov model rather than cost effectiveness because each stage of that IVF process you have cost impacts that you have to include in your health economic impact analysis.

Running the Markov model with one treatment cycle resulted in a probability which was that means plus or minus standard deviation of pregnancy at .4 for Bravelle and .37 for Follistim. When we applied the code for the females progressing or dropping out from that IVF cycle at any point in time, we came up with a corresponding cost of $11,584 for Bravelle versus $12,762 for Follistim.

I would like to emphasize that because I had some feedback from the people who have used the data. Why must you use the final transition probability which was the rate of on-going pregnancy between the two agents? As you can see, it is not consistent throughout the IVF cycle and since there are cost implications for each one of these IVF cycle transitions to health state, we needed to account for that.

Now I mentioned running the treatment, the clinical trials were only for one cycle treatment, but the reality of the situation, especially in the US, I know that we have some international guests attending, I do not know outside the US, but it is customary to think that a managed care administration might be inclined to pay for three IVF cycles for a female before she is not able to be reimbursed for more than three cycles, so it is customary to do three cycles even if the clinical trial was done only for one cycle. When we did that, the actual rate of pregnancy, Bravelle was .78 and Follistim was .75, and the corresponding costs were $22,000 for Bravelle versus $24,000 for Follistim. More or less, the same cost difference was multiplied for the three treatment cycles, rather than one. It is important to tell you that we kept the transition probability constant over time because we forget that whatever factors there are in terms of the females not achieving the same rate as their one cycle, it is constant between the two treatments and we forget this way that opens up less criticism in terms of what rate of reduction did we use for the transition probabilities past the first cycle. However, in order to represent what actually happens in real practice, not in clinical trials, we use as I mentioned earlier the Meldrum trial to see if our conclusions need to be changed if we used national statistics from actual practice rather than clinical trial database and the findings do not change usually. We have the same magnitude of difference in terms of efficacy and the same magnitude of difference in terms of cost effectiveness found in the three cycles versus the one cycle, versus usually the national statistics probability versus what we used in the trial.

This is, in essence, what you see, the summary statistics for the three treatment cycles: chemical pregnancy for urinary-derived treatment, 0.94, clinical pregnancy was .83 and on-going pregnancy was .78. Similarly, values for the competitive agents were also calculated.

In conclusion, the model indicates that there is an incremental advantage for human-derived FSH of 3% for continuing pregnancy rate and, after that, as Chappel discussed, you need to know this is based on the comparative clinical trial that was used in comparing the two agents and a cost difference of $1,178 was calculated and saved on the human-derived FSH and the human-derived FSH was found to be a more cost-effective treatment for infertility than Follitropin-beta.

Before I do that, I want you to know that when we ran the Monte Carlo simulation, we found that, regardless of how many patients you enter through the treatment cycle, Bravelle is less costly 100% of the time and 65% of the time, regardless of how you change it, it was more cost-effective. The 35% of the time we found that Follistim is more cost-effective was when women who paid for treatment somewhere in the neighbourhood, if my memory serves me correctly, over $30,000, then Follistim tends to be more cost-effective. So there is a way to use the information and see how much she is willing to pay for treatment and come up with a cost-effectiveness argument. But nowhere in the analysis did we find Follistim to be less costly, it was always more costly, and 65% of the time, Bravelle was more cost-effective. As I mentioned, we also did the transition probabilities using Meldrum and actually that is a very nice way of verifying whether the findings were really so different from the national statistics. I believe he rate of successful females going through IVF cycles the first time around was around 37% in the national statistics. Results effected in modest degrees the success rate of paying for the two agents; however, the overall incremental advantage of human-derived FSH over Follitropin-Beta was convincing.

I like and I take pride in what I do because your name and reputation are always on the line and I would like to say that the study was really important to us in that we did not use any assumptions. We did not, say, not like any of the numbers in the comparative trial and say, okay, let’s adjust them to see what we could come up with. This is what we got from the clinical trials whereby the FDA ended up approving Bravelle from the study and that is really important for me to see. We did not use any assumptions and there is really nothing wrong with assumptions because in the reality of the situation, it is very uncommon to come across a study where all of the data you need to build your economic analysis were readily available. So there is nothing wrong with assumptions, it just opens up the door for professional interpretation. Why did you assume a 20% rate of failure? Why did you do this? So we didn’t have to deal with that and that made our lives, if you will, as scientists, that much easier. The study meets the highest requirements for transparency and reporting cost-effectiveness studies. Transparency for those of you who come from outside the US means that anyone who has a computer and actually knows what he or she is doing can take our data and plug it in their computer program and come up with comparable findings. There is nothing in a black box, no magic formula, no magic wand - everything we use can be duplicated. I have to forewarn you that this is not an easy technique to do, but whoever has the skill and time and knowledge to do it can do our study and come up with comparable results to what we have done. So it meets the transparency requirement for reimbursement for us, as in European countries. In European healthcare, financiers would like to be able to see where the data came from. That is why the need for transparency. As far as we are concerned, the study meets the highest requirements for transparency and reporting cost-effective analysis.

Thank you very much for your attention.