Friday, July 07, 2006

Modeling prize fund rewards

by James Packard Love
As a general issue, I an not too keen on prize fund approaches for medicines that focus too narrowly on a specific disease or medical outcome. I think that a broader set of targets and more flexability in terms of earning participation in the rewards is quite important, given the stochastic nature of the R&D process. This note looks at a different issue, the way you structure rewards in different cases.

Aidan Hollis has emphasized the potential role of Quality Adjusted Life Years (QALYs), as a basis for rewards in a prize fund system. I have suggested several times that rewards should not be simple linear multiples of QALYs, as the costs of drug development are insensitive to population size, and high income societies also have strong willingness to pay for inventions that serve small client populations. There are other issues that are important also, such as the importance of developing medicines like antibiotics or medicines used in stockpiles for emergencies, where immediate usage is not the objective, but the medicine is desired as a back-up in cases of resistance, or for emergency use or other contingencies. There are separate issues associated with development of medicines for certain global health problems, like malaria, Chagas disease or visceral leishmaniasis.

These are among the reasons that HR 417 does not mandate a simple reward = QALY system, leaving the actual reward structure up to the managers of the Fund, with a flexiable legal structure that allows some learning by doing.

In thinking about the structure of rewards from a fixed prize fund, such as the one envisioned by the 2005 proposal for the Medical Innovation Prize Fund (HR 417), one can propose various ways of modeling the rewards.

If the rewards are to be based upon the incremental benefits of new (or improved) products, we have to identify the things we value.

One thing we want are improved health outcomes, as measures for example by QALYs. And while we certainly value more QALYs than less QALYs, it is not necessarily optimal to have a linear reward structure. One can imagine, for example, that the rewards for QALYs should follow a simple decay function, such as:

Reward = a + b * ( QALYs ^ k ) ,

where k (less than 1) is the decay parameter, and a and b are parameters that reflect the fixed and variable value of new products, both determined within the context of a budget constraint.

Products with larger populations (or greater benefits per patient) would receive more, but less on the margin. It could or could not also be combined with the notions of set-asides for orphan (rare) diseases (or other priorities) that is now part of HR 417.

One can also imagine different values for the parameter "a," depending upon the nature of the new product. For example -- the degree to which the new product itself represents an improvement over existing medicines, regardless of the number of patients, such as the current FDA S and P catagories, or even the catagory for products used to treat severe illnesses.

This is fairly simple, but I present it as an initial illustration of how one might go beyone the reward = QALY approach. I'll return to this later, with a number of other approaches, after a bit deeper look at different ways of modeling efficient reward systems.

In the case of antibiotics that are best used in case of failures of first line regimes, it seems important to think about rewards not tied to current usage of the new products. Use of the current stock of antibiotics can be modeled as a depletion of a resource, and the new antibiotics as replenishment.

In the case of a medicine to be used for a stockpile for a low probability event, such as a bird flu pandemic, SARS, or a bio-terrorism attack, one could imagine different approaches for medicines that have another use, such as Tamiflu, and medicines that would only be used for the low probability event (a vaccine for Bird Flu or Anthrax). I have suggested elsewhere that in the first case, the reward for the stockpiled products should be contingent upon actual use. But in the second case, methods of valuing options would seem more appropriate.


Post a Comment

Links to this post:

Create a Link

<< Home