Likelihood of Confusion: Is 15% The Magic Number?

David Bernstein, Partner at Debevoise & Plimpton, gave an interesting presentation yesterday at NYU Law Engelberg Center's "Proving IP" Conference on the origins of the "fifteen percent benchmark" in trademark likelihood of confusion analysis. (The subject of the panel was "Proving Consumer Perception: What are the best ways to test what consumers and users perceive about a work and how it is being positioned in the market?")

In trademark law, infringement occurs if defendant’s use of plaintiff’s trademark is likely to cause confusion as to the source of defendant’s product or as to sponsorship or affiliation. Courts across circuits often frame the question as whether an "appreciable number" of ordinarily prudent purchasers are likely to be confused. But evidence of actual confusion is not required. There is not supposed to be a magic number. Courts are supposed to assess a variety of factors, including the similarity of the marks and the markets in which they are used, along with evidence of actual confusion, if any, in order to asses whether confusion is likely, at some point, to occur.

In theory.

But in practice, Bernstein asserted, there is a magic number: it's around fifteen percent. Courts will often state that a survey finding 15% or more is sufficient to support likelihood of confusion, while under 15% suggests no likelihood of confusion. See, e.g., 1-800 CONTACTS, INC. v. Lens. com, Inc., 722 F. 3d 1229, 1248-49 (10th Cir. 2013) (discussing survey findings on the low end).

For example, in James Burrough Ltd. v. Sign of Beefeater, Inc., 540 F.2d 266 (7th Cir. 1976), the Seventh Circuit reversed the lower court's finding that a survey indicating only 15% of respondents were confused was legally insufficient to support likelihood of confusion.

The plaintiff, which owned the famous Beefeater brand of gin sold in bottles bearing the traditional London Beefeater, argued that consumers "would be likely, upon seeing ... [defendant's] Restaurant's sign, to believe that [defendant] Restaurant's enterprise was in some way related to, or connected or affiliated with, or sponsored by, [plaintiff.]."  Id. at 274.



Plaintiff put forward a survey showing that 15% of survey respondents believed the restaurant was sponsored by the distiller of the Beefeater gin. The district court found that the survey "demonstrated nothing more than that a 'small percentage' of the population 'thinks of' Beefeater Gin when 'they see the word Beefeater or hear it in any context' and that that was an insufficient modicum of proof."  Id. at 278-279 (quoting district court).

The Seventh Circuit overturned this conclusion.

We cannot agree that 15% is “ small.” Though the percentage of likely confusion required may vary from case to case, we cannot consider 15%, in the context of this case, involving the entire restaurant-going community, to be de minimus.

Id. at 279. See also, e.g.,Exxon Corp. v. Texas Motor Exch. of Houston, Inc., 628 F.2d 500, 507 (5th Cir. 1980) ("The survey results themselves indicate a high possibility of confusion between Texon and EXXON. Approximately 15 percent of the individuals surveyed associated the Texon sign with EXXON.").

In his clear and engaging talk, Bernstein brought to light something very interesting about the state of the law here: many of the major precedents that announce this 15% baseline, such as Sign of the Beefeater, occurred prior to consistent use of controls in confusion surveys.

Today, Bernstein says, sophisticated survey designers use controls to make sure the confusion observed is from defendant’s use of plaintiff’s protectable trademark, and not the result of some other factor, such as unprotectable aspects of the product's appearance, a stray word on the package, or the nature of the product itself. See Novartis Consumer Health, Inc. v. Johnson & Johnson-Merck Consumer Pharm. Co., 129 F. Supp. 2d 351, 365 (D.N.J. 2000) ("[T]he control cell functions as a baseline and provides a measure of the degree to which respondents are likely to give an answer ...not as a result of the name or labeling, but because of other factors, such as the survey's questions, the survey's procedures, the nature of the product, or some other potential influence on a respondent's answer such as pre-existing beliefs.").”

Bernstein gave the example of how surveys were used to test whether the trademark SOUTH BEACH DIET was infringed by a defendant's sale of food bars that bore the name "South Beach."  The bars had many unprotectable visual features that might have caused confusion, other than the name South Beach, such as the standard shape of defendant's food bar. By using a control that replaced the word "South Beach" with “Solutions,” but kept everything else the same, the survey designers were able to find more convincingly that it was indeed the South Beach name that was confusing people, not just the fact that this was a health food bar, and the South Beach Diet is a well-known diet plan.

The upshot is that the courts in the earlier cases that accepted 15% confusion were actually basing their conclusions only on what Bernstein called "gross confusion" (total confusion that includes "noise" from other factors) rather than "net confusion" (confusion caused only by use of plaintiff's trademark).

While this is very interesting to know, I am not sure what we are to make of this discrepancy. Should we take the earlier opinions to mean that even lower numbers today could potentially be enough? If a survey today with good controls shows, say, 7% net confusion, then shouldn't this be as compelling if not more compelling than a survey in the old days showing 15% gross confusion? Or do we just throw out the old numbers altogether as too "noisy," and start over?

The topic became still more interesting after Professor Graeme Dinwoodie spoke about growing skepticism of consumer surveys among courts in the UK. Dinwoodie's take-home seemed to be that, although surveys may have their flaws and can raise the costs of litigation, those who decry use of surveys should be careful what they wish for. If courts do not rely on surveys to discern consumers' perceptions of trademarks, it can be tempting for judges to simply substitute their own understandings.  Professor Barton Beebe, moderating the panel, stepped in with an anecdote. In a well-known trademark case involving the sale of girdles under the trademark "Miss Seventeen," the judge in the case apparently wandered the courthouse asking teenage girls whether they would think the girdles were put out or sponsored by plaintiff's magazine, "Seventeen." The panelists were in agreement that this approach seems problematic on many levels.

Still another wrinkle, introduced by Professor Beebe, was what these numbers tell us about everyone else. If 15% of people are confused, and this is enough to support infringement, what is the impact of this holding on the 85% of consumers who are not confused? Professor Beebe suggested that their interests should matter too, and that they could be benefiting from information imparted by trademarks, even if a (lesser) number of consumers are getting the wrong message. So, for example, if you are among the 85% of people who love the Sign of the Beefeater and find it a helpful and creative way to advertise a restaurant, it may seem unfair, perhaps inefficient, to make defendant use another name just because others aren't as savvy. At the least, Professor Beebe said, courts should continue to rely on percentages that indicate how many people are not confused as well as how many people are confused, rather than only looking at aggregate numbers of confused individuals. 

How do we encourage innovation on “long COVID”?

By Jacob S. Sherkow, Lisa Larrimore Ouellette, Nicholson Price, and Rachel Sachs

Since the pandemic began, numerous recovered COVID-19 patients have reported having “long COVID”: COVID-19 symptoms persisting well beyond the underlying viral infection period. Whether such a condition is specific to COVID-19, or more generally a form of “post-acute sequelae”—or even a discernable condition—has bedeviled scientists and clinicians alike. The fact remains, though, that likely millions of people in the U.S. alone will continue to report a variety of challenging symptoms more than 6 months after they’re infected. Despite this magnitude of reports, confusion regarding defining the condition and identifying its etiological basis has presented significant challenges to innovating treatments for it. In this post, we explore some of the current evidence surrounding “long COVID,” some of the difficulties in developing long COVID treatments, and how policymakers can move things along.

What’s the current evidence for long COVID and how prevalent is it?

Figuring out what long COVID is has proved frustratingly elusive. The CDC uses “long COVID”—or “post-COVID conditions”—as an umbrella term for a range of health problems experienced 4 or more weeks after infection, including shortness of breath, fatigue, “brain fog,” “post-exertional malaise,” cough, headache, diarrhea, dizziness, changes in smell, and more. Guidance from the UK’s National Institute for Health and Care Excellence (NICE) draws a distinction between “ongoing symptomatic COVID-19” from 4 to 12 weeks after infection, and “post-COVID-19 syndrome” for symptoms that continue for more than 12 weeks. In November 2021, NICE published detailed evidence reviews on defining post-COVID-19 syndrome and the most prevalent symptoms, both of which noted the low quality of the existing evidence base. In particular, most long COVID studies are cross-sectional retrospective surveys of self-reported symptoms, which increases the risk of recall bias. These studies have also almost exclusively been conducted in high-income countries, and survey respondents are not even representative of the populations within those countries: they have mostly been white, female, and of higher socioeconomic status.

SARS-CoV-2 is not the only virus that results in long-term reported symptoms—generally known as “post-acute sequelae”—although the risk of sequelae after COVID-19 appears higher than the risk after influenza, suggesting that long COVID might not just be a general post-viral complication. But the lack of a consistent case definition has confounded studies, and the risk of biased surveys was recently illustrated by a recent French study that suggested that reporting long COVID symptoms “may be associated more with the belief in having been infected with SARS-CoV-2 than with having laboratory-confirmed COVID-19 infection.” Some researchers have thus focused on more objective markers of persistent clinical symptoms, which still leaves “many available hypotheses” to explain COVID’s long-term effects—to the understandable frustration of patients who are looking for more answers from the medical community.

An improved clinical understanding of what long COVID is—and how to treat it—will become even more important as people recover from the massive Omicron wave that has been circling the globe. While few long COVID cases result in hospitalization, some are serious, and they may put further strain on an already splintering healthcare system. Long COVID is also affecting many individuals’ ability to go back to work—in some cases, due to cognitive impairments—and may be exacerbating the current labor shortage. Patients are also desperate for treatments and sometimes turning to ineffective “alternative” therapies, which could create additional health risks.

What are the innovation challenges for developing therapies for long COVID?

Developing therapies for long COVID involves several interlinked challenges relating to definitions, reimbursement, clinical trials, competing incentives, and non-excludable treatments.

Perhaps the biggest challenge involves concretely defining long COVID. Medicine is typically bad at defining chronic conditions with variable and often non-specific symptoms using diagnostic exclusionary criteria—that is, when it’s difficult to say exactly what counts as having the disease (and what counts as being successfully treated). Not knowing the etiology (the causal pathway) of a disease makes it difficult to develop drugs for the disease, all things being equal. This parallels some of the failures of developing treatments for Alzheimer’s: it’s not clear what the molecular basis of the disease is (i.e., what causes Alzheimer’s) or whether it’s truly a single “disease” or a collection of diseases with potentially different treatments. For long COVID, the lack of a known mechanism (though scientists are starting to learn more) is also linked to unfortunately widespread skepticism and under-reporting, which is likely to reduce incentives for developers. Diffuse, hard-to quantify symptoms have long had trouble winning developer attention, as individuals who suffer from endometriosis, for instance, have known for far too long.

Second, and relatedly, because long COVID is not a well defined condition, reimbursement is more challenging. Reimbursement can function as both a private and a public source of incentives for development, whether deliberately or not. If a condition is poorly defined and not readily reimbursable, that decreases the incentives to develop treatments for it in the first place.

Third, clinical trials on possible treatments are likely to be both expensive and long. Trials are likely to be expensive because they would need to be large when the endpoint is relief from non-specific symptoms. Controls may also be difficult to recruit, since so many people have been infected by COVID by now. This is especially true because of the length of trials likely needed; if long COVID is defined by having several months of symptoms, for how long must symptoms remit to achieve an endpoint?  Long trials unfortunately tend to discourage therapeutic innovation. 

Fourth and finally, there is some evidence that non-excludable treatments work to help treat long COVID, including exercise, occupational therapy, and cognitive therapy. R&D on these treatments is likely to be underincentivized by existing innovation institutions. And to the extent these are successful and become the standard treatment, they’ll compete with excludable therapies—and accordingly diminish the incentives to develop them in the first place.

What can policymakers do to support innovation related to long COVID?

A first item on which policymakers have already made progress in supporting innovation related to long COVID is to expand access to insurance and care for those experiencing its symptoms. As we have written previously, ensuring that patients have access to health insurance (which provides reimbursement for treatments and services) can serve itself as an innovation incentive for pharmaceutical companies. The 2010 passage of the Affordable Care Act and its guarantee that insurers cannot discriminate against patients with pre-existing conditions (such as having previously had COVID-19) are a key step in this direction. HHS’s more recent guidance that long COVID can qualify as a disability for purposes of anti-discrimination laws is also helpful. 

But more could be done on this front. One step that could be taken is for the Social Security Administration to add long COVID to its list of qualifying conditions, which would assist patients in becoming eligible for Medicaid or Medicare benefits. 

A second item is for policymakers to invest in research enabling scientists to learn more about the long-term effects of COVID-19. In December 2020, Congress provided $1.15 billion to study long COVID, and in February 2021 the NIH launched an initiative to use this funding. This support was extended by the American Rescue Plan, and in fall 2021 NIH awarded nearly $470 million to study these issues. This research could provide physicians with novel ideas for treatment in the short-term, and pharmaceutical companies with potential drug targets in the more medium-term. But it will be important to ensure that these research efforts are accessible to patients broadly, not only those who may have the connections and resources to travel to large academic medical centers. 

Third, policymakers could consider additional ways to support the development of treatments for long COVID. Existing market-based reward incentives may be sufficient to encourage the development of novel prescription drugs or medical devices, but as noted above, they may be less effective at encouraging the development of information about the efficacy of non-excludable interventions like exercise. Trials on non-pharmaceutical interventions for long COVID might be directly supported by public funding, given the lack of private incentives to do so. 

Fourth and finally, policymakers could support ongoing efforts to repurpose existing drugs that might be effective in treating long COVID. For similar reasons as above, pharmaceutical companies may have limited incentives to test their old drugs for new conditions. But, just as some existing medications have been effective in treating COVID-19, the same may be true for long COVID. Direct funding of trials or advance purchase commitments may be useful in this area. 

This post is part of a series on COVID-19 innovation law and policy. Author order is rotated with each post.