Threshold for Sound Prediction

Sandoz Canada Inc v Janssen Inc 2023 FCA 221 Locke JA: Mactavish, Monaghan JJA affg 2022 FC 715 Pallotta J

2,659,770 / macitentan / OPSUMIT / NOC

In this brief decision, Locke JA has provided a useful summary of the law related to the threshold for a “sound” prediction of utility. The case also implicitly raises an interesting issue relating to the standard of appellate review.

Threshold for a sound prediction

At trial, Pallotta J held that Janssen’s 770 patent was valid in the face of a variety of attacks: see here. The only argument pursued on appeal related to lack of utility [2]. Pallotta J had held that while utility had not been demonstrated, it was soundly predicted. Sandoz appealed on the basis that Pallotta J has erred in law by applying too low a threshold in concluding that the prediction was “sound.” Sandoz acknowledged that Pallotta J has not explicitly misstated the law, but argued that she must have applied the wrong threshold because, according to Sandoz, the evidence could not support a finding of sound prediction based on a correct statement of the law [18]. Locke JA rejected this argument, on the basis that the threshold for sound prediction “is not high” and the facts as found by Pallotta J were sufficient to support her holding.

Locke JA’s decision provides a useful summary of the law related to the threshold for sound prediction. The leading case on sound prediction is of course Wellcome / AZT 2002 SCC 77. The leading case on the threshold is Novopharm / olanzapine 2010 FCA 197, quoted by Locke JA at [8]:

[84] [Wellcome / AZT 2002 SCC 77] does not define the threshold required for sound prediction. However, Binnie J. states that more than mere speculation is required (para. 69). He also provides the following indicia:

• the requirement is that the claims be fairly based on the patent disclosure (para. 59);

• it must be prima facie reasonable that the patentee should have a claim (para. 60);

• it cannot mean a certainty (para. 62);

• the desired result must be able to be inferred from the factual basis (para. 70).

[85] In my view, these indicia signify that a sound prediction requires a prima facie reasonable inference of utility.

The requirement of “a prima facie reasonable inference of utility” is the best known summary statement of the threshold. Pallotta J did not mention this wording explicitly, but that in itself “is not remarkable, and does not amount to an error of law” [30]. (Indeed, Sandoz itself did not explicitly refer to this wording in its written submissions before Pallotta J [31].)

Locke JA noted that Sandoz cited several additional passages from Wellcome / AZT that indicate what is not sufficient for a prediction to be sound:

1. No more than a mere belief that something might be useful (para. 25);

2. Little more than an announcement of a research project (para. 64); and

3. Only a promise that a hypothesis might later prove useful (para. 84).

Locke JA concluded:

[16] In my view, there is nothing in [Novopharm / olanzapine] that represents any kind of departure from what one would glean from a complete reading of Wellcome. While it is not necessary that the prediction be certain (see Wellcome at paragraph 62), or to a regulatory standard (see Wellcome at paragraph 63), the public is entitled to a teaching that is solid (see Wellcome at paragraph 69) and accurate and meaningful (see Wellcome at paragraph 83), and based not on speculation but exact science (see Wellcome at paragraph 84).

In affirming Pallotta J, Locke JA stated that “I am not prepared to conclude that it overlooked the guidance therein that the teaching in a patent based on a sound prediction must be solid, accurate and meaningful, and based on exact science (not speculation),” citing Wellcome / AZT [69], [83], [84]. Given his reliance on Novopharm / olanzapine, the description of the threshold for sound prediction as being as being “solid, accurate and meaningful, and based on exact science (not speculation),” is an explanation of what it means for an inference to be “a prima facie reasonable,” and is not intended as a departure from the threshold set out in Novopharm.

Standard of appellate review

Sandoz’s argument also raises an interesting issue as to how to distinguish a holding of law from a finding of fact in applying the standard of appellate review. Sandoz characterized Pallotta J’s putative error as an error of law in order to invoke a correctness standard of review [11]. Sandoz did not argue that Pallotta J made a palpable and overriding error of fact. While Sandoz acknowledged that Pallotta J did not explicitly misstate the law, it argued that Pallotta J must have failed to actually apply the correct law because the evidence could not support her finding that the prediction was sound [18]. Sandoz characterized this as an extricable error of law [11].

Sandoz’s argument was along the following lines. Suppose the law is that “Facts A, B and C are all relevant to establishing sound prediction.” If the trial judge says “C is not relevant and I have not considered it,” this is clearly an error of law. Now suppose the trial judge correctly states the law, and then says “I find that facts A, B and C are not proven,” yet nonetheless concludes that sound prediction has been established. The finding that sound prediction has been established is a mixed finding of law and fact, and so is to be reviewed on a deferential standard unless there is an extricable error of law. But the law itself was correctly stated. And it doesn’t seem to be an error of fact, because no one is disputing the factual findings that A, B and C were not proven. Sandoz’s argument was that this kind of error is an error of law, on the view that since the facts are undisputed and the result was clearly erroneous, we can infer that even though the trial judge stated the law correctly, the trial judge cannot actually have applied the correct law to the facts. Sansoz’s argument seems right in an extreme case such as I have described. The difficulty is that on that argument, the only way for an appellate court to decide whether the trial judge made an error of law is to delve into how the facts were applied to the law; and how the law was applied to the facts is a classic issue of a mixed finding of fact and law, which is reviewed on a deferential standard.

So, suppose a case involved a tort action in which the question was whether the defendant had been driving negligently, and the court stated “The law is that the defendant must act as a reasonable driver would in the circumstances. I find as a fact that the defendant was driving at 45kph in a 50kph zone, while sober, on dry roads, at noon, in heavy fog. I find that the defendant was not negligent.” If the factual findings are not disputed, this is a classic example of a question of mixed fact and law, to be reviewed on a deferential standard: Housen [2002] 2 SCR 235 [28]. Now suppose the court had instead stated “The law is that the defendant must act as a reasonable driver would in the circumstances. I find as a fact that the defendant was driving at 160kph in a 50kph zone, while drunk, on black ice, at dusk, in heavy fog. I find that the defendant was not negligent.” This must be an error of some kind; but the law is correctly stated and no one is disputing the specific factual findings. Is it an extricable error of law, as Sandoz argued in this case? It is true that an extricable error of law need not be explicit and can be inferred from the result on the facts, but an extricable error law is typically capable of being articulated as a specific legal rule, even if the trial court did not do so. For example, in Galaske v O’Donnell [1994] 1 SCR 670, discussed in Housen [31], the trial judge had found the driver of a car not negligent in failing to ensure that his young passenger wore a seat belt. This implies that the driver did not owe a duty of care to his underage passenger. The failure to identify this duty was an extricable error of law. Similarly, in Rhône (The) v Peter A.B. Widener (The) [1993] 1 SCR 497, discussed in Housen [34], the extricable error of law was failure to recognize that the key factor distinguishing directing minds of a corporation from normal employees “is the capacity to exercise decision-making authority on matters of corporate policy, rather than merely to give effect to such policy on an operational basis.” Again, the extricable error can be separately articulated as a rule of law. In my negligent driving example, there is no separate rule that can be extracted.

Now, in this case, Sandoz did suggest that there is an extricable rule, namely the threshold for sound prediction. This argument makes sense. A threshold is a point on a continuum, so there may be cases that are so far from the threshold that we can say that the trial judge must have applied the wrong threshold, even if the trial judge articulated the threshold correctly. The same can be said about my negligent driving example. So far, so good.

The problem is that even if we accept that the threshold is a matter of law, actually extracting the threshold that was implicitly applied by the trial judge may be impossible except in the easiest cases. Put another way, in Galaske and The Rhône, there was a clear rule that could be extracted and a specific error identified. When the legal rule is a threshold that turns on a holistic consideration of the facts, it will not be possible to extract the rule from its application in any case which is anywhere close to the line. So, in the tort example I gave above with the extreme set of facts, we can say holistically that there must have been an error of law, even if we cannot specifically identify it. But as we get closer to the line (what if it was dusk, the driver was going slightly over the speed limit etc), it becomes increasingly difficult to extract any legal error, even though one may have been made. Put another way, the trial judge who found the driver was not negligent on the extreme facts clearly had an erroneous conception of reasonableness in their mind. As the facts get closer to the line of ‘true’ reasonableness, the trial judge’s decision will become more and more defensible, even though the trial judge is applying exactly the same erroneous legal concept. When the facts are anywhere close to the line, an appellate court will say “We affirm on the basis that the holding that there was no negligence is a question of mixed law and fact to be reviewed on a deferential standard.” But as the facts get further from the line, it becomes clear at some point that the trial judge must have been making an error of law all along.

Maybe this is all a long-winded way of saying what the SCC said in Housen [28], that the distinction between an error of law and an error of mixed law and fact “can be difficult in practice because matters of mixed law and fact fall along a spectrum of particularity.” But it seems to me that Housen was making a somewhat different point, namely that we call something a rule of law when it is broadly applicable to a range of facts and we call a finding a question of mixed law and fact when it is unique to the circumstances of the case: “If a court were to decide that driving at a certain speed on a certain road under certain conditions was negligent, its decision would not have any great value as a precedent” [28], quoting Southam [1997] 1 SCR 748 [37]. My point is somewhat different. Even when we can state a generalizable precedent, it may be difficult to extract that rule from its application, particularly when the rule involves a threshold (or more generally, any kind of line drawn along a continuum, such as reasonableness in tort), and the facts of the case are somewhere close to the line. In that context, it makes sense to say that a rule of law of is involved, as there are cases at either extreme that we can say are clearly correctly decided one way or the other, even without disputing the facts. But it is nonetheless problematic to apply a correctness standard to a decision involving the application of that law to the facts in a case which is close to the line.

This was arguably such a case, which tested the limits of what exactly constitutes a sound prediction. How should the courts deal with such a case? Locke JA held that the threshold for establishing sound prediction is “not high” and “[t]he terms ‘prima facie’ and ‘reasonable inference’ leave considerable space for a fact-finding body in reaching its conclusion” [24]. He further noted that “[t]he fact that more experimentation was required after the rat studies did not necessarily take the utility of the invention outside the scope of a sound prediction” [25]. This “considerable space” remark, in combination with his look at the facts, strikes me as being a kind of intermediate level of review. It seems to me that Locke JA did take a look at the facts, albeit somewhat cursorily, to satisfy himself that Pallotta J had not obviously misapplied the law to the facts. That is not exactly standard doctrine on the standard of appellate review—Locke JA doesn’t seem to have been applying either an entirely deferential standard, or a correctness standard—but it strikes me as a sound approach.