“Pfizer had clues its blockbuster drug could prevent Alzheimer’s. Why didn’t it tell the world?”

Jon Baron points to this news article by Christopher Rowland:

Pfizer had clues its blockbuster drug could prevent Alzheimer’s. Why didn’t it tell the world?

A team of researchers inside Pfizer made a startling find in 2015: The company’s blockbuster rheumatoid arthritis therapy Enbrel, a powerful anti-inflammatory drug, appeared to reduce the risk of Alzheimer’s disease by 64 percent.

The results were from an analysis of hundreds of thousands of insurance claims. Verifying that the drug would actually have that effect in people would require a costly clinical trial — and after several years of internal discussion, Pfizer opted against further investigation and chose not to make the data public, the company confirmed.

Researchers in the company’s division of inflammation and immunology urged Pfizer to conduct a clinical trial on thousands of patients, which they estimated would cost $80 million, to see if the signal contained in the data was real, according to an internal company document obtained by The Washington Post. . . .

The company told The Post that it decided during its three years of internal reviews that Enbrel did not show promise for Alzheimer’s prevention because the drug does not directly reach brain tissue. It deemed the likelihood of a successful clinical trial to be low. A synopsis of its statistical findings prepared for outside publication, it says, did not meet its “rigorous scientific standards.” . . .

Likewise, Pfizer said it opted against publication of its data because of its doubts about the results. It said publishing the information might have led outside scientists down an invalid pathway.

Rowland’s news article is amazing, with lots of detail:

Statisticians in 2015 analyzed real world data, hundreds of thousands of medical insurance claims involving people with rheumatoid arthritis and other inflammatory diseases, according to the Pfizer PowerPoint obtained by The Post.

They divided those anonymous patients into two equal groups of 127,000 each, one of patients with an Alzheimer’s diagnosis and one of patients without. Then they checked for Enbrel treatment. There were more people, 302, treated with Enbrel in the group without Alzheimer’s diagnosis. In the group with Alzheimer’s, 110 had been treated with Enbrel.

The numbers may seem small, but they were mirrored in the same proportion when the researchers checked insurance claims information from another database. The Pfizer team also produced closely similar numbers for Humira, a drug marketed by AbbVie that works like Enbrel. The positive results also showed up when checked for “memory loss” and “mild cognitive impairment,” indicating Enbrel may have benefit for treating the earliest stages of Alzheimer’s.

A clinical trial to prove the hypothesis would take four years and involve 3,000 to 4,000 patients, according to the Pfizer document that recommended a trial. . . .

One reason for caution: another class of anti-inflammatory therapies, called non-steroidal anti-inflammatory drugs (NSAIDS), showed no effect against mild-to-moderate Alzheimer’s in several clinical trials a decade ago. Still, a long-term follow-up of one of those trials indicated a benefit if NSAID use began when the brain was still normal, suggesting the timing of therapy could be key.

Baron writes:

I bet this revelation leads to a slew of off-label prescriptions, just as happened with estrogen a couple of decades ago. My physician friends told me then that you could not recruit subjects for a clinical trial because doctors were just prescribing estrogen for all menopausal women, to prevent
Alzheimer’s. I’m still not convinced that the reversal of this practice was a mistake.

That said, off-label prescribing is often a matter of degree. It isn’t as if physicians prescribed hormone replacement for the sole purpose of preventing Alzheimer’s. Rather this was mentioned to patients as an additional selling point.

Here’s the bit that I didn’t understand:

“Likewise, Pfizer said it opted against publication of its data because of its doubts about the results. It said publishing the information might have led outside scientists down an invalid pathway.”

Huh? That makes no sense at all to me.

Baron also points to this blog by Derek Lowe, “A Missed Alzheimer’s Opportunity? Not So Much,” which argues that the news article quoted above is misleading and that there are good reasons that this Alzheimer’s trial was not done.

Baron then adds:

I find this issue quite interesting. It is not just about statistics in the narrow sense but also about the kind of arguments that would go into forming “Bayesian priors”. In this sort of case, I think that the structure of arguments (about the blood-brain barrier, the possible mechanisms of the effect, the evidence from other trials) could be formalized, perhaps in Bayesian terms. I recall that a few attempts were made to do this for arguments in court cases, but this one is simpler. (And David Schum tried to avoid Bayesian arguments, as I recall.)

It does appear that the reported result was not simply the result of dredging data for anything “significant” (raising the problem of multiple tests). This complicates the story.

I also think that part of the problem is the high cost of clinical trials. In my book “Against bioethics” I argued that some of the problems were the result of “ethical” rules, such as those that regard high pay for subjects as “coercive”, thus slowing down recruitment. But I suspect that FDA statistical requirements may still be a problem. I have not kept up with that.

Conflict of interest statement: I’ve done some work with Novartis.

47 thoughts on ““Pfizer had clues its blockbuster drug could prevent Alzheimer’s. Why didn’t it tell the world?”

  1. “That makes no sense at all to me.”

    But as a company that sells proprietary drugs it makes sense: if some new information comes out that changes the analysis, the company still has the option to move forward with it’s confidential advantage in tact.

    But from the article:
    “Having acquired the knowledge, refusing to disclose it to those who might act upon it hides a potential benefit, and thereby wrongs and probably harms those at risk of developing Alzheimer’s by impeding research”

    This is the same argument that was floated by treatment advocates when analysis revealed that aggressive early treatment of some cancers caused as much harm as good.

    Companies can’t do anything right! Stupid companies.

    • Jim:

      I don’t think that “companies can’t do anything right.” When I say, “That makes no sense at all to me,” I just mean that it makes no sense to me right now. I assume there’s some reason for such decisions, even if the reason is that they’ve always done things that way. We should be able to question a company’s decision without jumping to the extreme position that “companies can’t do anything right.”

      • Andrew:

        Apologies, I didn’t intend to attribute that sentiment to you.

        It’s in reference to the preceding quote from the article and to the more general milieu that whatever a drug company does, some advocate somewhere will twist it into an ethical failure that’s all about unethical corporate behavior – all because of ****DIRTY MONEY!!!!*** :)

      • I think you are right and that company’s do right means maintaining profitability over time and increasing them when you can without negatively impacting future profitability over time.

        Now that will based on what expertise they relied on, but my guess is their guess was that doing the trial would do right for them.

        Not releasing the suggestive study material is a bit puzzling given it like would have increased off-label sales of their drug.

        Concerns about legal liabilities, regulatory tolerance about off-label drug use that might affect other products, competitors position to make better use of what was in that study material and who knows what. Or maybe they did not have the staff capacity?

        Now, from what I learned MBA school this “It said publishing the information might have led outside scientists down an invalid pathway.” would be a benefit especially i many off those scientists worked for a competitor.

        (My last meeting with the president of a large corporation, which was my last MBA type job ended with “If you decide to work with our competitor X, we will find out whatever you are working on and supply whatever funds our people need to make sure it fails.” I did consider approaching competitor X but I had be warned about the real boss there, so did not. Hey that competitor X boss did end up in prison, so maybe I made a good decision.

        • > I think you are right and that company’s do right means maintaining profitability over time and increasing them

          Yeah not releasing the data seems to make a lot of sense here. They’re trying to make dollarydoos, right?

          Certainly doesn’t seem right in terms of figuring out how to treat Alzeihmer’s, but that’s not their first and foremost goal?

        • I don’t know the reason why Pfizer didn’t publish their results but i certainly don’t believe their public statement that it might lead scientists down the wrong path. Keith, I think you are probably correct that concerns about off-label use may have been one reason. Yes, off-label sales would have increased but so would off-label serious adverse events. Serious off-label adverse events in an elderly, compromised population would concern both the FDA and the company. I did laugh out loud regarding your speculation that they may have lacked staff. We are talking about the largest pharma company in the world with literally an army of statisticians, epidemiologists, medical writers, etc. – if they wanted to publish their findings, it would have been published.

      • Andrew:

        Apologies, I didn’t mean to attribute that sentiment to you. It was in reference to the preceding quote. Different topic but I didn’t make the distinction clear.

  2. In the sense that the plausability of effect was very low because of physiolgical reasons, it does make sense that they didn’t pursue. But I don’t see how Bayesian statistics intervene here ; it rather plays into the “asking the right questions” you outlined several times imo.

    • I was referring to Bayesian tree-like models, not statistics, although the line may be hard to draw. For example, you have nodes in the tree like “crosses the blood brain barrier”, yes/no, then, after “no”, “has a metabolite that does cross the b/b/b”, and, if no to that, “has some other indirect effect on the brain”. And the probability of each transition can itself be estimated by referring both to data and known physiological effects.

      I’m not sure how to incorporate base-rate data into these estimates. Loew points out that “most promising treatments don’t work”. Yes, this is relevant to some decisions. But I also want to know statistics about “not crossing the b/b/b and affecting the brain anyway”. And “how often general findings about a class of drugs fail to apply to individual members of that class.” And, even with those, we may know things about the specific case that lead to modifications of the observed relative frequency.

      Anyway, from such a tree, you work back from the nodes that lead to true effects to figure their probability given the assumptions you make. The general category of this approach is “conditional assessment”, that is, using yourself to get a second or third opinion about a probability judgment by breaking it down into separate components and then combining them.

      A big issue in medicine and elsewhere is how much weight to put on data vs. our understanding of how things work. There are occasional treatments that work well and are adopted on the basis of data alone, with understanding coming, if at all, much later. (Semmelweis’s discovery of the benefits of hand washing is an example. His theory about why it worked was wrong.) And there are other cases (e.g., surgical procedures) where treatments are adopted largely on the basis of understanding without much data at all.

      If I thought I was getting Alzheimer’s, I would very likely start taking one of these drugs. My costs and benefits are not the same as those of the company.

      (That said, I seem to be a risk taker. I tried to persuade my eye doctor to supervise an experiment in which I tried to dissolve my early-stage cataracts with lanosterol, which seems to work on dogs. She said, “No way am I going to recommend this. If you do it, you are on your own.” I decided to wait, since my condition seems stable for now, and I could not figure out how to assess small effects to see if it was working, without her help.)

      • Pretty sure things RA treatments like Enbrel have side effects and risks that you’d be forced to take into account before making such a decision. For my own part, an apparent unexplained statistical association from a secondary data analysis isn’t nearly a sure enough benefit to outweighs years of drug side effects (just on the off-chance there’s some real protective effect).

  3. I sometimes struggle to guess what Andrew’s default position (or that of the majority of commentors on this blog) will be on certain matters. This is one of those.

    There’s a lot of preaching here on how statistical models should very closely reflect the precise mechanistic nature of the thing they’re trying to quantify. In the absence of any known or even postulated way for Enbrel to cross the blood/brain barrier and prevent or treat dementia, what sort of model do we think is appropriate for analyzing the results of a big clinical trial with that purpose?

    Or flip it the other way around. If someone were to publish results of a secondary analysis making big claims for the preventative powers of Enbrel on dementia, wouldn’t this blog likely be shouting it down as an example of the sorts of a-mechanistic fishing expeditions that result in big claims failing to replicate or turning out to need retracting?

    • Different purposes different analyses. *for science* in which the point of science is more or less to create models of how the world works… you should do mechanistic modeling. If you don’t it’s like being a surgeon but refusing to ever do surgery.

      On the other hand for measurement, or decision making, using alternative models that only incorporate very limited information is sometimes all you have.

      For example, to make the decision “should we start to study this to understand the mechanism by which this drug helps with Alzheimers” you obviously can’t do the scientific analysis first to decide whether you should do the scientific analysis.

      The big problem in my opinion is when people see as their goal “get statistical significance” instead of “understand how this drug affects molecules in the brain”

      • Do Big Pharma researchers have to pass the usual publication filter of “novel findings”? If not, then there seems to be less incentive to find statistical significance in that world. I find that appealing.

  4. I would support Pfizer releasing a big trove of results of analyses similar to this one across all drugs and outcomes (I’m sure they’ve conducted them). However, I think they made the right decision not to selectively publish this one. The analysis appears to have been a crappy case-control design:

    “They divided those anonymous patients into two equal groups of 127,000 each, one of patients with an Alzheimer’s diagnosis and one of patients without. Then they checked for Enbrel treatment. There were more people, 302, treated with Enbrel in the group without Alzheimer’s diagnosis. In the group with Alzheimer’s, 110 had been treated with Enbrel.”

    Even ignoring the pharmacodynamic argument that the drug doesn’t reach the brain, I would say that such a study is so likely to be confounded that it conveys virtually no information. By publishing it, Pfizer would be implicitly endorsing the idea that it was suggestive of a causal effect. And we have seen time and again that publishing bad observational studies does indeed lead to subsequent wasted scientific effort. If they released this result with other similar crappy analyses, then that implicit endorsement could be avoided. But if they only cherry-pick this result, the implication that they think it’s particularly important is unavoidable and likely to be net harmful. (I say this as someone with some skin in the game.)

    • This is actually an interesting special case of a more general phenomenon where believing that people should always do X does not force you to support someone doing X in a particular circumstance given that they do not always do X. Another example could be prosecuting corruption. I would not support a particular case of a leader prosecuting a corrupt political enemy if they do not generally prosecute corrupt allies as well.

      I suspect Andrew favors publishing everything and so supports publishing these results as a subset of “everything”. But there are harms to cherry-picked publishing that we’re all well aware of.

  5. Students and faculty are always asking me not to share their research. When I told Andrew I didn’t want to participate in secret research, he said he liked the term “private”. I contend it’s secret in that people tell me something, then ask me not to share, just like any other secret (i.e., “something that is kept or meant to be kept unknown or unseen by others.” according to Oxford). Maybe “confidential” is better (“intended to be kept secret” accordig to Oxford).

    It’s gotten so bad with people mailing me data and models off-list and asking me to keep them a secret, that I had to develop a form letter to return to people telling them that I’m not going to look at their secret work.

    Back when I was thinking about leaving academia in the mid-90s, I took the MBA finance class at Carnegie Mellon (where I was a professor). It was one of the worst classes I’ve ever taken—on a par with the one 200 level psych class I took at Michigan State in that it was something anyone who was halfway decent at math could’ve aced in high school. Our first exercise after computing the net present value of a B-school degree was to optimize payments to a pharma company that developed an AIDS cure (this was in the mid-90s). Of course, that meant selling it at a premium in the first world and letting it out to the rest of the world later (the exercise was again to maximize NPV). So not only awful, but creepy. It totally put me off a career in business.

    • “that meant selling it at a premium in the first world and letting it out to the rest of the world later ”

      But without that business model there would be **no** cure.

      You could have public investment by tax dollars but it would take so much longer to reach the goal that the cure would come quicker to the third world via corporate investment. Not to mention the exploding cost. If that kind of public investment thinking really worked we’d all be Soviets today.

      • I think there’s plenty of evidence that public dollar “prizes” for validated solutions to problems provided by companies produces good results. The biggest concern is how to avoid gaming the validation, but then, that’s a problem with snake oil salesmen as well… The advantage of the up-front prize rather than patent type system is that as soon as the solution is figured out, competition comes into play in the actual mass production of the good.

        So, I don’t think there’s any reason to believe that what we have now is some kind of optimal good thing, it’s better than Soviet command and control, yes, but has plenty of room for improvement. Patents and Copyrights are after all ways to *foil* the free market in order to induce people to invest in public goods. We’d be better off not foiling the market and just induce people to invest in things by up-front paying them when they demonstrate the information/public good itself.

        • “The advantage of the up-front prize rather than patent type system is that as soon as the solution is figured out, competition comes into play in the actual mass production of the good.”

          With the T-Mobile merger progressives wanted to block the merger to maintain high intensity price competition. Then these same people are complaining that companies don’t offer good wages, good health care, are always laying off people etc etc. If the price competition forces the seller to bare bones prices, then you can forget about high wages.

        • You must mistake me for a progressive perhaps.

          I think wages should be market wages as well. I also think we should have a flat tax, and use it to redistribute 10-15% of all income into a Universal Basic Income approximately equal for all (maybe with some age and health adjustments), thereby allowing wages to signal the value of the work they reflect (the marginal product), while still enabling people to live without excessive hunger, homelessness, poverty etc.

          In the absence of government monopolies on production of healthcare for example costs of healthcare would plummet, without interference in prices of housing and where people can build, housing would be built, and costs of housing would plummet, meaning the lower wages could let people afford to live on them.

          The assumption that it’s a good thing that the govt gets involved and allocates monopolies or oligopolies to people is extremely problematic.

        • “You must mistake me for a progressive perhaps.”

          hmmm…no. Don’t know where you are politically other than “not conservative”. No, check that. Nope. I do know. You’re a Utopian Idealist.

          My comment was that monopolies have benefits. The point about “progressives” was secondary.

        • Monopolies have benefits to the monopolists for sure. Having benefits globally or even mildly broadly is a very questionable assumption.

          I’m definitely not a Utopian Idealist, but I can probably see that from where I’m standing. Most people I think are deep in a valley somewhere digging a hole and trying to get others to jump in with them.

        • Well, I think if you look at a lot of the “monopolies” in the US, starting with Standard Oil and going forward, you’d find they achieved that status largely – but not exclusively – through superior business practices and that one of the key factors in their success was reducing prices.

        • The RIAA called, they have a subpoena for you, they plan to sue you into bankruptcy like the other 10000 customers they sued, you know, the ones who on average actually buy more stuff than the others

      • > But without that business model there would be **no** cure.

        Wait, but there actually is no AIDS cure, right?

        > If that kind of public investment thinking really worked we’d all be Soviets today.

        So you’re saying maybe we should try being Soviets then?

        • “Wait, but there actually is no AIDS cure, right?”

          Rob used AIDS as an example from his experience but the contexts suggests the story was meant to illustrate a larger concept about the pharmaceutical industry. My comment reflects the broader context: while the profit-driven business model of the pharmaceutical industry creates discrepancies in treatment opportunities, it’s still the fastest way to create medications and get treatment to the largest number of people, and probably by a large margin.

        • Makes you wonder, doesn’t it?

          What if, what is *ACTUALLY* needed are doctors wanting to fight diseases, and giving sufficient resources to them? I know, radical idea.

          (Also, health care and medical research in socialist/communist nations is/was actually better than what our biased view makes us think. Because maybe those nations also have doctors wanting to fight diseases? Again, radical idea.)

        • Furthermore, *NOT* having to limit yourself to *ONLY* developing something like a drug that can be sold (otherwise you make no money), but instead looking to causes of medical problem, might actually be quite liberating to the medical profession.

          Solutions to medical problems are more than just giving drugs – at least that’s what I tell myself when I think about medicine.

      • The U.S. government exclusively funded and controlled the Manhattan Project, the outcomes of which were split atoms and nuclear bombs (not saying that bombs are a good outcome for humanity, just that serious science was done with government funding and control). A few decades later, that same kind of government funding took us to the moon. I’d say those are pretty amazing outcomes based on tax dollars.

        Even something like Stan, our probabilistic programming language and inference system was almost entirely government grant funded for version 1. Now, we have open-source contributors from industry and grants from private foundations, so we’re no longer entirely government funded, just mostly government funded. The result is that we now have a free software system everyone can use and expand. And we beat industry to it, though they’re catching up.

        Aren’t most Nobel prizes awarded to university professors? Aren’t they largely government grant funded rather than driven by a profit motive? If you make funding available to scientists, they’ll do science.

        The opposite side of this is that major world problems like malaria aren’t tackled by big pharma because the potential for profit is too low because mostly poor people get malaria. This seems like a strong sign that capitalism isn’t the right model to solve pressing world health problems.

        Let’s look at a place like Bell Labs in the 20th century. They invented the transistor, information theory, the laser, sonar, C, and Unix, etc. etc. They did that while entirely government funded. Congress set Bell Labs’ budget because AT&T was a monopololy. They were not allowed to profit from their inventions. So C and Unix are also free. After AT&T was split up in the early 1980s and again in the mid-1990s, the research labs became more profit driven. It was a long slow decline the resutl of which was no more big breakthroughs in science and technology (I worked for Bell Labs after the second split from 1996 to 2000).

        National health systems have better outcomes for the population as a whole (life expectancy, infant mortality, etc.) at a fraction of the cost of the U.S. system.

        I’m hardly an economic history scholar, but my understanding is that there was a lot else going on in the Soviet Union besides public investment.

  6. It seems to me the next step shouldn’t be a clinical trial, but focused research trying to map out potential biological pathways for anti-inflammatory drugs to impact the brain. If there is an effect, it’s likely secondary or tertiary…but potentially clinically important. The rush to a clinical trial to demonstrate a causal link could then reasonably follow.

    That being said, I’m aware of more than a few drugs that are routinely prescribed or sold over the counter that work by mechanisms we don’t yet fully understand. I’m just saying that, in this case, the call for large-scale testing in the absence of a causal theory seemed rushed.

  7. he WaPo article notes “Enbrel has reached the end of its patent life.”

    So, they would have been conducting a study which would seem to be of primary benefit to the world at large, rather than Pfizer. What’s the point of that? So I can see not conducting a clinical trial. Not publishing the data is more puzzling.

  8. Enbrel blocks the tumor necrosis factor, it messes with your immune response. Impressive list of potential side effects. It’s hardly something that would ever be prescribed as a preventive medication, for anything. Also it costs a pretty penny.

Leave a Reply to Thomas Cancel reply

Your email address will not be published. Required fields are marked *