Modeling Fast Track Designation
Since 2018, Captario has participated in formulating product strategies for a number of oncology assets. These development projects, run by a range of different companies and operations, have some key points in:
They are approaching or are in phase 2
They have a potential to be granted fast track designation by FDA
There is a complex market situation with several competitors within and outside of the class.
As decision scientists, we are in the room to bring quantitative analytics to the project strategy discussions. In other words, we are quantifying intangibles such as risk and opportunities and trying to put them into context of the decision at hand.
Participation in these project team discussions is always a learning experience, and this has helped us to develop new methods of modeling and evaluating R&D assets which in the end will help drug project sponsors make better decisions.
THE CASE STUDY MODEL
When building models, we use a graphic process modeling notation (see picture below) where rectangles are activities and diamonds are decision gateways. The activities have durations and costs, and the decision points have probabilities for each one of the outgoing paths. This is a bit simplified, but is what we need to know for now.
Now, for the real oncology cases, we have built models similar to the picture below.
Here is an overview of what is going on in the project:
After the on-going phase 1 study, the sponsor makes a decision to either continue development or terminate the project.
After the go-ahead, a study is started with very few patients; we have seen as few as 15 up to about 50. This can be considered an interim part of a larger registrational study (5), or as a separate entity. In some cases, the small first study is using a surrogate endpoint, e.g., if the primary registrational endpoint is overall survival, the surrogate endpoint could be response rate.
After the interim cut-off, there is typically a feasibility decision where the sponsor will terminate the project if there is no response at all.
The main decision is dependent on the interim readout and is really more of an outcome rather than a decision; If there is an exceptional result in the interim data, FDA may grant fast track designation. This will lead to a registration procedure directly after the interim study. If fast track status is not achieved, the sponsor will run a full registrational trial before submission.
After the registration/FDA approval process if we are successful, we will launch a new product. The market activity includes a sales model that will forecast sales based on what has happened in the preceding R&D project. This means that if the project takes the fast track path through the model, the sales are adjusted based on that.
There are a numerous modeling capabilities that are must-haves for this level of realism in the model:
Clearly, getting fast track designation and launching early is a huge upside for the project, and it is fundamentally important that a model can capture alternative paths to launch and subsequently allow them to impact the project or strategy value.
The sales model must be dynamic enough to take into account which path has been taken to get to a launch. If we launch 2-3 years early, that should have down-stream effects on order of entry and sales in the end. So rather than providing a simple sales curve, our sales model should be a function of R&D outcomes. In this case, since launch timing is so volatile depending on path, sales should be a function of at least launch time.
The durations and costs of clinical trials depend on a number of things, and in order to make realistic forecasts, we need to be able to model this. In our models, we have used ranges to estimate duration and cost in some cases, and in others we have created sub-models which take into account, endpoints, number of patients, and number of centers to model the trial cost and duration.
Now let us focus on the output from this model. When evaluating drug project strategies it is usually a good idea to look at the project from many different angles to eventually reach a conclusion. Right now we will will focus on the decision tree. A decision tree will generally focus on the decisions or pivotal crossroads in a project and will show the effect of key decisions. Our version of the decision tree does that and we have added a few things that makes ours quite useful for understanding the dynamics of the project. Please see the picture below. I will briefly go through what happens, and then highlight a couple of key takeaways.
DECISION TREE WALKTHROUGH
In our version of the decision tree, decisions are represented by clear diamonds. Leaf destinations are either red dots, which represents project terminations, and green triangles which represent product launches. Please note that each decision has a label which corresponds to a decision in the process diagram above. The data above each diamond has the following meaning (using Ph3 Start at (1) in the graph as example):
The percentage shows the likelihood that we will reach this point from the start of the process.
Date: This shows the expected date when we will reach this point (if we reach it at all!).
eNPV: Since we have a sales function in our model, we can calculate an eNPV, which we use as a representation of project value. There are many opinions regarding the usefulness of eNPV, but in the decision tree the interesting part is not the absolute value, but the difference in eNPV between milestones.
Probability of launch: The POL from the current decision point.
The first decision that we make from now will be to either continue into phase 3 or stop after the ongoing study. This decision has 100% chance of happening.
The decision point CRR > 5% is after we have run the small first study in phase 3. It is important to note that activities are not represented at all, so there is quite some time between this and the first diamond. The value of the project is now (before the decision has been made) 446 M which is substantially higher than 379 M which we had earlier. This is explained by the fact that when we get to point 2, we will have avoided the first red dot, which has about 14% risk of happening.
This is the fast track gateway. If we saw an exceptional result in the interim readout, we will take the fast track path. There is only just over 10% chance that we will have results that support fast track, but if we do get it, the project value is dramatically increased by more than 100%. On the other hand, if we do not get fast track, and that opportunity is lost, we lose 65% of the project value.
In the process map there are two ways of getting to launch, which corresponds to the two launch scenarios in the decision tree. This is a very good overview where we can compare and contrast the two scenarios, and where we can start to get an understanding what it means in terms of value if we can get the fast track designation.
To sum this up, I just want to say that a decision tree can be a fantastic way to visualize decisions and their impact on project timeline, cost and value. It is made possible by having a good flexible modeling context as base and build from there.Having automatically generated decision trees is an invaluable tool for quickly assessing different project scenarios and will definitely help the team build an understanding of project dynamics.