ChEMBL

I have just got an account to try out a cool patent database (more on this later in a future post, maybe), but it gave me a chance to do an analysis I had wanted to do for some time - well at least start the analysis. The basic question is - Is there a relationship between R&D expenditure and discovery output? (as judged instead of the usual measure of drug approval, but instead by the more frequent and hopefully 'leading indicator' metric of patent publication). Future drugs will be based on current and historical research efforts, and so, on an idealised basis, one would hope for an approximate correlation between the number of patents held and future revenues - this assumption is problematic for a number of reasons, primarily, there is a huge and wild variance on the 'value' of certain patents/products, primarily due to the presence of large blockbusters. A further motivation of doing this was to complement the analysis that are usually done on R&D costs per launched drug, and also as part of a general interest in identifying 'leading' indicators for drug discovery.

With all the mergers in the industry at the moment, this is clearly complicated, but I made the assumption, that for publicly traded companies the R&D funded research would yield patents assigned in the name of the public company; for M&A activities the assumption is that the sunk costs on discovery will be accounted for on the acquired companies balance sheets, and it would appear that many of the original 'acquired' patent filings are progressed under the acquired entities name (albeit probably funded from the acquiring companies current R&D costs). Anyway, the accounting and costing here is nothing but a gross approximation, but I am sure that within private companies, some real, properly costed calculations like this have been done. In fact, this is such a toy analysis, it may not even merit the storage on the hard disks of the blog-site ;)

Here are the data (for the 'largest' pharmaceutical company Pfizer, chosen because of their size).

Patent publication - total of 1) US granted and apps, 2) EU granted and apps, 3) WO apps

Year	Number of assigned published patents
2000	463
2001	539
2002	592
2003	830
2004	1038
2005	1093
2006	908
2007	724
2008	611

Research and Development Costs

I then looked for R&D expenditure per year (this was surprisingly difficult to do freely, quickly and accurately), so 1) it would be good to have these numbers readily available in the public domain and 2) there may be some errors in the numbers here! I then also corrected the expenditures to 2008 dollars (so as to take account of inflation, the measure I used here was GDP deflator (using the calculator on http://www.measuringworth.com/uscompare/ - I do not know enough to judge if this is the best approach, but it makes superficial sense). The basic effect of this dollar correction is to flatten out the cost per patent curve (as you would hope).

Year	R&D cost ($B)	Corrected R&D cost (2008 $B)
2000	4.4	5.38
2001	4.8	5.74
2002	5.1	6.01
2003	7.1	8.19
2004	7.8	8.74
2005	9.1	9.87
2006	7.6	7.98
2007	7.5	7.66
2008	7.9	7.90

So what does the data say - well the 'R&D cost per patent', on average, over the period 2000-2008 is $10.2M. Of course this is an average for a specific company, and other companies may well be different average numbers for a number of reasons (e.g. a more prolific or restrictive corporate policy/strategy on patenting, changing patent strategies over time). However, to my eyes there are a couple of interesting features from these graphs. 1) There is a general correlation between absolute R&D investment and the number of patents published, and 2) There is not a significant lag between the R&D investment and the patent output (you could expect that it takes some time to a) make the inventions and discoveries, and then patent, and wait for the patent to publish), so to be honest I was expecting a lag between the R&D cost curve and the patent output curve, maybe of around two years....

Here is a graph of patent output vs R&D cost.

Here is the graph on 'cost per patent' as a function over time.

Anyway, there is a lot more interesting stuff that could be simply done with this data, including 1) For fully integrated companies, is there a systematic difference in per patent cost that reflects either a fundamental difference in patent strategy, or an underlying difference in R&D productivity, and 2) More detailed and appropriate slicing of patent data by those that are granted, maintained, etc.