Is Sound Prediction a "Free-Standing Doctrine"?

Teva Canada Ltd v Novartis AG / imatinib 2013 FC 141, Snider J

            GLEEVEC / imatinib mesylate / 2,093,203

[164] Sound prediction is not a free-standing statutory requirement. Rather, it is a way of showing that an invention is useful when the invention has not been directly demonstrated to work. Its introduction into Canadian law was not, as I understand it, to give a crushing hammer to those who challenge patents.

This is technically and historically correct, but it does not mean that patentees have no cause for complaint. Sound prediction originated as an aid to patentees, but some developments have been very problematic. These stem from a pregnant passage in Wellcome / AZT 2002 SCC 77 [70], that said that to support sound prediction, in addition to a factual basis and a sound line of reasoning:

there must be proper disclosure. . . . In this sort of case, however, the sound prediction is to some extent the quid pro quo the applicant offers in exchange for the patent monopoly. Precise disclosure requirements in this regard do not arise for decision in this case because both the underlying facts (the test data) and the line of reasoning (the chain terminator effect) were in fact disclosed, and disclosure in this respect did not become an issue between the parties. I therefore say no more about it.

While these remarks are expressly obiter, they suggest that when utility is based on sound prediction, there may be a requirement that the facts and line of reasoning be disclosed in the patent itself. This suggestion was taken up by Hughes J in Lilly / raloxifene 2008 FC 142 [164] and affirmed by the FCA 2009 FCA 97 [15], holding that the factual basis for sound disclosure be provided in the patent itself: see my article “Must the Factual Basis for Sound Prediction Be Disclosed in the Patent?” (2012) 28 CIPR 39, draft version here. Furthermore, in its Latanoprost decision, 2011 FCA 236, the FCA strongly implied that the line of reasoning must also be disclosed in the patent.

In her imatinib decision, Snider J held that the utility of some use claims and some genus claims was established based on sound prediction. Consequently her decisions navigates the shoals of these heightened disclosure requirements.

Consider first the issue of disclosure of the factual basis for sound prediction in respect of the use claims, in particular Claim 46 read with Claim 29, which claimed imatinib for chemotherapy treatment of tumours. Snider J construed the promise of this claim as being the potential to treat tumours in warm-blooded animals [268]. There was a strong factual basis for predicting an in vivo effect, consisting of both in vitro tests and in vivo murine tests [271-90]. There was also a straightforward line of reasoning: imatinib had been demonstrated to inhibit v-ABL kinase, it would probably also inhibit the closely related BCR-ABL kinase, and it was well-established that chronic myeloid leukemia was caused by BCR-ABL kinase [313].

The difficulty was that the in vivo tests were not disclosed in the patent. Fortunately for the patentee, Snider J ultimately held that the in vitro tests alone were a sufficient factual basis to establish the sound prediction [314]. 

However, Snider J also discussed the in vivo tests at length [274-89], and in her conclusion in the final paragraph of this section [290] – “I am satisfied that the Ciba-Geigy researchers had a factual basis for predicting the utility of the [compounds in question] compounds in the chemotherapy of tumours” – implies that her conclusion is based on all of the evidence that had just been reviewed. Furthermore, after stating that the in vitro tests alone were a sufficient factual basis, she immediately went on to say that “The in vivo testing carried out by Ciba-Geigy was not essential to establish the line of reasoning; nevertheless, the fact that no experiments can be said to have “failed” strengthens the prediction and confirms the line of reasoning” [314]. This is curious. It is true enough that undisclosed tests will be relevant if they undermine the prediction by demonstrating lack of utility. But so far as establishing utility, if the factual basis must be disclosed in the patent itself, the evidence not disclosed in the patent is irrelevant, and it is hard to see how such evidence can “strengthen” the prediction. I wonder if Snider J’s extensive treatment of apparently irrelevant evidence, and her reference to it “strengthening” the prediction shows some discomfort with the raloxifene doctrine? Perhaps this is wishful thinking on my part. When she turned to consider the disclosure requirement, she accepted that the raloxifene doctrine requires disclosure of the factual basis in the patent itself [323], and certainly her reasons are formally consistent with that position.

Snider J did clarify that the raloxifene doctrine does not require the disclosure of all important test results [319], so long as the results which are disclosed as sufficient to permit a sound prediction. This holding is entirely consistent with the leading cases; it truly is no more than a clarification of the raloxifene doctrine. She also clarified that it is permissible to rely on common general knowledge as part of that factual basis [323].

There is one thing that is not entirely clear to me about this whole analysis. Why was the utility of the use claims based on sound prediction? Snider J answered this question by stating that “[f]or the use claims. . . the promise is that [a specified compound] can be used to treat the diseases or conditions specified in the use claims. There is no question in my mind that demonstration of such utility would require more than the in vitro testing that was carried out in the Ciba-Geigy laboratories. . . . Thus, I must consider the question of sound prediction” [266]. 

However, strictly, the promise of the use claims was for the “potential” to treat tumours [268]. The word “potential” is not incidental; it was the central issue in the “can be used” debate which spanned paras 139-151, and Snider J stated expressly that this “potential” might be “demonstrated or predicted” [151].

Despite this clear statement of the promise, Snider J’s discussion of the use claims seems to proceed on the basis that what must be soundly predicted is use for treating tumours, not merely potential for such use:

Thus. . . could the use of imatinib . . . for the ‘chemotherapy of tumours’ . . .be soundly predicted? [268]
I am satisfied that the Ciba-Geigy researchers had a factual basis for predicting the utility of the . . . compounds in the chemotherapy of tumours. [290]
I am satisfied that the testing that Ciba-Geigy had carried out (and disclosed in the patent) coupled with the knowledge in the area leads to a sound prediction that imatinib would be useful as a chemotherapy treatment. [333]

But if the promise is only for potential use, isn’t the question whether the evidence shows a sound prediction that imatinib would potentially be useful? At one point Snider J raises the issue of potential use: “Quite simply, that information – widely known and accepted – was to the effect that a compound that could selectively inhibit either PDGF-R or v-ABL would have the potential for the treatment of tumours. As accepted by all of the experts, this link was well known to the point of being notorious” [328]. But if that is so, why isn’t that enough to establish demonstrated utility of potential for treatment of tumours?

While this may seem like wordplay, it is important wordplay. There is a difference between demonstrated utility of a potential for treatment, and a sound prediction of utility as a treatment. The difference is that in the latter case, but not the former, the patent will have to meet the heightened disclosure requirements that the Federal Courts have discerned in the sound prediction doctrine. In particular, if the question is whether there is demonstrated utility of a potential for treatment, it is clear law that the in vivo test data could be considered. This is exactly the point made by Hughes J in GSK / rosiglitazone Hughes J 2011 FC 239, as discussed here. Of course, this is the kind of absurd hair-splitting distinction that gives law and lawyers a bad name, but it is the inevitable consequence of the raloxifene doctrine. Snider J statement that “[s]ound prediction is not a free-standing statutory requirement” [164] is true in principle, but not in practice. The sound prediction doctrine does have requirements that have no equivalent in demonstrated utility; it is a free-standing requirement, albeit one that does not have any statutory basis.