Personalized medicine uses various types of diagnostic information to improve outcomes and reduce healthcare costs by determining
in advance whether a patient is likely to respond to a proposed treatment.
Michael J. Shuster
Of course, the best-known and earliest success of the use of diagnostics in personalized medicine relates to measuring HER2/neu
amplification in breast cancer patients. Only those patients whose tumors carry this amplification respond to Herceptin (trastuzumab),
and the drug's use is indicated only in such patients. Current study results indicate that treatment for a year reduces risk
or recurrence by half in these patients.1 The cost of Herceptin treatment is about $3,000 per month and is easily justified because of the benefits it provides.
NEW CHALLENGES IN PATENT PROTECTION
Getting patent protection for such diagnostics, however, recently became more difficult. The United States Patent and Trademark
Office (USPTO) has made it difficult to obtain patent protection for certain types of personalized medicine technology. Unlike
the test used for distinguishing Herceptin responders, i.e., a simple measurement of amplification of a single gene, other
personalized medicine diagnostics require measuring gene or protein levels for a number of biomarkers. These measurements
are used with sophisticated predictive modeling algorithms to generate a score associated with drug response or disease severity.
One difficulty encountered in obtaining effective patent protection for this type of diagnostic arises from the ability to
use various combinations of biomarkers to develop equivalent predictive models.
For example, while the expression of a handful of genes can be used in any particular predictive model, those genes can often
be selected from a larger set of informative biomarkers. Consider a situation in which gene expression analysis identifies
30 genes whose expression significantly differs between identified groups of responsive and nonresponsive patients. Informative
predictive models are generated using expression values for four or more genes. Consider further that many combinations of
four genes out of the 30 provide good predictive power in the model. Ideally, a patent application would claim a method of
determining the efficacy of the treatment by measuring the expression of four or more of the 30 genes and including those
values in the predictive model. Such a patent claim encompasses 27,405 different combinations of genes. Because this is such
a large number, a patent examiner may object that the large number of combinations creates an undue search burden for the
examiner.2 Consequently, the examiner issues a "restriction requirement" that causes the applicant to choose one combination of, in
this example, four or more genes whose expression values are used in the claimed method. Claims to other combinations would
then need to be pursued in separate patent applications, raising the cost of protection to unaffordable levels.
This creates a difficult position for the patent applicant. If many combinations of four or more of the 30 genes are provided
adequate predictive power, a patent for a single combination of four provides essentially useless patent protection. If restriction
among each combination is required, the applicant is forced to choose which of the many combinations it developed and disclosed
will it protect. The combinations that it cannot afford to protect can be used by competitors to develop essentially equivalent
tests that are not protected by the applicant's patents.