(This is a response from the virologist/epidemiologist who contributed this article.)
Thanks to both readers that have made it so far.
First: Michael is right. You can feel the rage dripping from his comments and are similar to those eloquently and bravely stated by my hero, the late Jonathan Mann, AIDS is one of the most important events in history, surpassing the Black Plague in its death toll . Measles, polio, meningitis, Kaposi’s sarcoma, obesity—almost every human affliction affects the poor and disenfranchised more frequently and more intensely. This is morally abhorrent. But distorting and denying facts will not help stop AIDS, even though we know that it is social conditions that allow this virus to flourish and to kill.
Second: George Lakoff is also right: don’t ever argue with the bad guys by letting them frame the debate. I knew that it was dumb when I did it, and I went ahead anyway.
Bialy puts up a series of silly questions that would deserve an “F” for any beginning student in an epidemiology course on causality. This isn’t just empty ad hominem pejorative as I hope to convince you below. But, I answered each question with cited research. Michael, anonymous, funkyfrankerbeany, etc. don’t attempt to address these studies and there is no evidence in their comments that they even read the abstracts. That’s not being skeptical, that’s just plain lazy (actually funk, when I was referring to swamp marsh gases, I was describing cholera, not malaria. The prevailing miasma theory for cholera was held until 1848 when John Snow, the father of infectious disease epidemiology, showed it to be bunk a London cholera epidemic and here). If you want to argue with me, bring your studies or stay home and stop wasting electrons.
The best and most eloquent single piece of evidence for HIV causing AIDS was cited in a posting on Dailykos.com regarding the infamous Harper’s article. Here is the logic: by the mid1990s, pharmaceutical scientists had developed protease and reverse-transcriptase inhibitor drugs designed on the properties of these proteins encoded by HIV. We are not talking about AZT, which was developed prior to the discovery of HIV, I’m referring to idinavir, nevirapine, etc. The reason why these drugs were developed and tested was solely their activity on the proteins that are made by HIV. If they affect the syndrome known as AIDS, and you still deny that HIV is a cause of AIDS, well, brother (or sister) I can’t help you. So what is the data for symptomatic AIDS in the era of these drugs? :
See Figure 2 here.
It is pretty clear from this graph what is going on across the US AIDS population once effective drugs that directly target HIV were approved and distributed in the US. This does not even begin to take into consideration blinded, randomized clinical trials (the gold standard for epidemiology) of each individual drug such as here, or here, or here.
So we see that antiHIV drugs are effective against AIDS on an individual level and on a population-wide level. Are some of these drugs outrageously expensive and do they have severe, potentially fatal side-effects? Yes, but that is another story and does not impact on whether or not HIV is the cause of AIDS, which in comparison has a near 100% fatality rate. To recap: 1) scientists identify key proteins encoded by HIV, 2) make drugs to specifically targeting these proteins, 3) treat AIDS patients who then have, 4) reduced mortality and morbidity of AIDS. Sounds kind of like HIV is causing AIDS, doesn’t it?
Sorry, can’t wrap my brain around the belief that the AIDS epidemic is all a surveillance artifact that we missed somehow in earlier decades before 1980 (hence HIV’s association with AIDS is an artifact). Having nine hundred thousand people keel over dead from diseases like Kaposi’s sarcoma, pneumocystis pneumonia and tuberculosis is pretty hard to miss (it could, however, be a conspiracy, see: Doctors'_plot). If the surveillance data that you see in these figures is AIDS, and if it is dropping on both population-wide levels as shown here and randomized, blinded clinical trials once HIV-specific drugs are given to patients, there are not very many other ways that you can interpret the data.
Bialy and Dean: I don’t see you as true skeptics, much as I would like to. Maybe this is just because I find your comments boring and incoherent. But, real skeptics would question the bases of these data and provide alternative explanations that might better explain them. A good skeptic should make these scientists really squirm. They might ask questions like, “What is the surveillance sensitivity?” What is the evidence that patients were truly randomized?” “What was the intent-to-treat effect?” These are reasonable skeptical questions and they are the types of questions asked every time a paper is peer-reviewed.
To me, it seems that “HIV-deniers” are the exact opposite of skeptics: they slavishly accept the hypothesis of Duesberg and ignore anything that disagrees with it—which pretty much includes all science generated over the past quarter century. Doesn’t sound like skepticism, sounds more like dumb.
Let’s talk about chimps and AIDS since the Bialy crowd is fixated on this, although I don’t know why. This gets back to the species specificity issue--which isn’t very interesting, I would pick a better argument if you want to deny that HIV is the cause of AIDS. Viruses, having evolved over thousands to millions of years, have genetically aligned themselves with their hosts. A reasonable skeptic always asks “Does the disease in an animal model truly represent the same pathogenic process in humans?” Personally, I find infecting chimps with anything immoral because of their high degree of social and behavioral sophistication. But they are infectable with HIV. They tend to not develop full AIDS syndrome but this hardly falsifies HIV as the cause of AIDS. In fact, HIV is a zoonotic infection spread to humans of a chimpanzee virus from the subspecies Pan troglodytes troglodytes. Beatrice Hahn and her coworkers have done brilliant work in figuring this out. There are four major subspecies of chimps, and she showed that one of them is the natural host of HIV. P. troglodytes troglodytes have widespread infection with HIV and the obvious conclusion is that chimpanzees and HIV have genetically adapted to each other and the virus does not cause symptomatic disease in this primate. Regardless of your interpretation of this data, lack of AIDS symptoms in chimpanzees after HIV infection is not a useful argument (see story about African and Indian herpesviruses).
Alright, so I addressed Bialy’s questions, now it is my turn. How do scientists determine causality? The single best and most accessible article on this is here.
I don’t agree with Fredericks and Relman on everything they say about causality but it is masterful and should be read by everyone who is interested in the subject. I go further--you shouldn’t post on a blog about HIV and AIDS causality without reading it, it is that good.
Causality is an area bridging science and philosophy. One of the first people attempting to address this question was the Scottish empiricist, David Hume who had many interesting ideas. First, he said we cannot state causality by reasoning alone and it must rely on empirical observation. This is true even for mathematical causality, an area where pure reason might apply and is seemingly an ideal case for causality through sheer reason. Principles used to secondarily derive causal inference are based on axioms derived from observation (e.g. 2 plus 2 causes 4. Okay, but what is a 2? Well, we see one thing and another thing just like, and we call them two. We are defining “two” by our experience). Although, this may seem obvious, it overthrew a line of Aristotelian thinking that held sway over 18th century science and is still in vogue with Duesberg. Secondly, as described here:
“When we examine experience to see how expectations are actually produced, we discover that they arise after we have experienced "the constant conjunction of two objects;" only then do we "expect the one from the appearance of the other." But when "repetition of any particular act or operation produces a propensity to renew the same act or operation...we always say, that this propensity is the effect of Custom" (EHU, 43)” Morris, William Edward, "David Hume", The Stanford Encyclopedia of Philosophy (Spring 2001 Edition), Edward N. Zalta (ed.).
In essence, if B always follows A, then we say that A causes B. This is short-changing Hume’s ideas but it illustrates the point.
See Figure 3 here.
Graph illustrates problem of saying that if B is always associated with A, then A causes B.
Unfortunately, life is rarely this simple and it is easy to make logical arguments against this definition of causality. For example the epidemiologic concept of confounding (a reasonable, but not perfect, explanation of confounding is here) is problematic for this:
one ruse perpetrated by tobacco companies was that it was not cigarette smoke that causes lung cancer, it was the sulphur smoke from matches that caused lung cancer. How do you separate the two? Well, this requires a study in which you recruit sufficient numbers of lung cancer patients so that a large enough fraction only used zippo lighters (by the way this reminds me of a superb piece of music by Phil Kline called Zippo songs here and here) so that you can show that folks who did not use matches but smoked cigarettes still developed cancer. You can see that it is pretty easy to trip up an epidemiologist and that it is really tough to prove causality.)
In the realm of infectious diseases, the next major player to address causality was the German microbiologist Robert Koch (pronounced like you-know-what, not like the soft drink). Koch is considered the “Father of Microbiology’ (actually co-Father along with Louis Pasteur and the two despised each other). Koch not only defined anthrax, discovered cholera bacteria and the bacillus that caused tuberculosis, he also asked the question “How do I know that the organism that I’ve found, actually causes disease?” Koch’s postulates are as follows:
- The agent is always present in the case of disease (see Hume above).
- The agent is never present in the absence of disease.
- The agent induces disease in a susceptible host.
- The agent can be isolated in pure culture again, and introduced into a new host, causing disease.
This was a brilliant construction appropriate for the 1880s, but I would no more have Koch be my causalist than I would have Sigmund Freud be my psychotherapist. Unfortunately, eager biology freshman are still tortured with these postulates and it is no wonder that they don’t want to become scientists. Koch did not know about asymptomatic carrier states (people are infected but don’t have disease such as with the bacteria causing meningococcal meningitis and can pass it on to others, but very few actually have disease), and he did not know about viruses (discovered by Dmitri Iwanoski in 1892. Koch did a great job, like Hume before him, but biology and life ain’t that simple.
Often you will hear the MIGHTY ROAR: NO ONE HAS PROVED THAT HIV FOLLOWS KOCH’S POSTULATES!!!! Whoa, got me there, you clever guys! The reason is that you cannot isolate a virus in free culture the way that Koch had in mind for bacteria. So, by the same reasoning, no one has proved that any virus has ever caused disease. Some people have tried to manipulate Koch’s postulates around by examining cases where HIV was transmitted through blood transfusion as an example of transmission of an “isolated agent” but it doesn’t really work well and I agree with Peter Duesberg that this is pretty twisted logic. It isn’t what Koch meant, so heed the advice of Lakoff and don’t even try to define modern causality in terms of Koch’s postulates.
Enter Austin Bradford Hill and Richard Doll.
Hill and Doll are largely credited (appropriately so) for describing the relationship (i.e., cause) between lung cancer and smoking. This was again a brilliant period for epidemiology, they developed and expanded technical methods such as case-control studies, cross-sectional studies, and cohort studies that are the bread and butter of epidemiology. Since we can’t ethically experiment on people (to cause severe illness or death), how do we take the existing data that we do have on the natural history of disease to make causal inference?
Epidemiology is fundamentally different from experimental science. Epidemiology, like astronomy, is an observational science. We have to take the data that we have and analyze it, so that we can tease apart associations and figure out the most likely explanations for what the underlying mechanisms are for how patterns of disease occur.
In terms of HIV and AIDS, we see a syndrome that we call AIDS (won’t belabor what this is, look it up here), we see certain people getting AIDS. Who are they? What is similar about them? What is different? We also find a new retrovirus called HIV. Who is infected with it? What happens to these people? Is it the cause of AIDS? How do we know?
Bradford Hill came up with a set of “criteria” which are really common-sensical. He said the following (more or less) that if an agent (cigarettes, HIV, etc.) causes disease (lung cancer, AIDS, etc.), even in this messy world it should have the following characteristics:
Association: The agent should generally be found associated with disease.
This is what Hume said. Even Duesberg et al. agree that AIDS patients are generally all HIV+. They just say that it is a passenger virus that doesn’t actually cause disease. So read on.
Generalizability: The agent should be associated with disease in different settings, by different people using different methods and so forth. Yep, HIV is seen in US AIDS, African AIDS, AIDS in old people, AIDS in young people, AIDS in red people, AIDS in blue people, etc. Okay, Peter is still with us here.
Specificity: The agent should be generally associated with just one disease or only a few diseases. This is like Koch’s second postulate but a little less strict. Hill thought this was the weakest piece to causality because it is clear that cigarette smoking not only causes lung cancer, but also emphysema, head and neck cancer and so forth. But the point is that if you postulate that an agent causes every kind of disease then you can’t prove that it causes any disease. Oxygen is always present in lung cancer. Does oxygen cause lung cancer? Maybe, but you can’t prove it so why worry about it. So, the number of diseases associated with the agent should be limited and the more limited and precise they are, the more likely the agent causes these diseases. In particular, if the diseases tend to be linked to each other and people who get one of the diseases also gets others (as lung cancer, emphysema, head and neck cancer are) then the association is specific.
Here’s where we lose old Pete. He seems to think that HIV is a harmless virus that coinfects AIDS patients who are sick for other reasons. In other words, HIV is not specific for AIDS. If Peter D were right, lots of people would have HIV but would not have AIDS or symptoms that a reasonable person would think are a prelude to AIDS. Fortunately, HIV infection is very uncommon in the United States and only about 0.1-0.3% of people have this virus. So, not a large percentage of Americans are running around with asymptomatic HIV infection.
What happens to people that are infected with HIV? Unfortunately, HIV is an uncommonly vicious virus and most if not all people infected with HIV eventually develop AIDS. Rutherford and colleagues followed a group of 489 gay and bisexual men who had been infected with HIV by the early 1980s . They followed this group for 11 years, when 46% had developed AIDS, and another 43% had “preAIDS symptoms” (e.g. swollen lymph nodes, dropping CD4 cell counts, etc.) More importantly, when they plot the rate that people with HIV get AIDS, it looks like the incubation period is very long, about 10 years but there is no evidence that a large fraction of people are protected from getting AIDS. This means that it looks like almost everyone with HIV, if left untreated, will develop AIDS. This is very different from most infections. The bacteria that causes meningococcal meningitis, for example, is present in the throats of about 10% of us at any one time but the chances of getting meningitis are small (about 3 per 100,000 people per year). Similarly for the virus that causes cervical cancer: many women are infected at one point or another, but only a small fraction develop cancer. Not so with HIV. So on the basis of this alone, it is very striking that HIV is causal for AIDS. But let’s go on….
Oh wait, says Duesberg and Bialy, people with AIDS have many, many diseases! This also violates your specificity rule! Let’s think about this for a second. AIDS patients have one fundamental underlying disease: their immune system is crapping out. This means that they secondarily get lots of different infections but the underlying commonality is that they all have severe immunosuppression. You do not find people with AIDS who have an intact, healthy immune system. So, if you define disease this way, which makes a lot of sense, HIV is very specific. If you wanted to be as pedantic as Pete D., most lung cancer patients don’t die of lung cancer either. They die of septicemia (overwhelming infection), pneumonia (infection of the lungs), respiratory failure, cerebral bleeding (from brain metastasis), liver failure (from lung metastasis), etc. Get the pikshur?
Dose-response (also Biologic-gradient): The more someone is exposed to an agent, the more likely that they will get sick with the disease. This makes a lot of sense for environmental exposures like cigarette smoking (a 40 pack-year smoker is more likely to get lung cancer than a 15 year old who steals his Mom’s Marlboros, lights up, gets green-sick and never touches cigarettes again). But it isn’t so good for infectious agents like HIV, since these agents replicate. Getting infected with one virus will get you as sick as getting infected with 10 viruses. So this is sometimes interpreted as, do we find the agent at the site of the disease and less so in more distant sites?
For HIV, this would be in the lymph nodes and thymus where CD4 cells live and that is precisely where the virus is found. So, what’s next?
Temporal Association: This was Hill’s key criteria, the big enchilada. I f an agent causes disease then it MUST precede the presence of the disease. If it doesn’t, then it would be a passenger virus, as Duesberg suggests. There are many, many cohort studies showing seroconversion to HIV and then subsequent development of AIDS. Go look them up. It is good mental exercise for you and I’m getting tired of doing your heavy lifting. HIV infection does not come after AIDS. Just doesn’t. Do an internet search using the terms “cohort study” with “AIDS” or “HIV”. I recommend PubMed.
By the way, technically this is the very most difficult thing for epidemiologists to determine. The reason is simple math: if someone develops a disease at a rate of 1 case per 1000 people per year (a very common disease, most diseases including AIDS are less common), you would need to follow 100,000 people for a year or 1000 people for 10 years to see only 100 people develop disease. Very expensive, and tenure committees rarely let you sit on your hands for 10 years until you collect your data. If you are looking for uncommon associations, then it can be very hard to see them this way although it is the best way. This is called a cohort study. Because cohort studies are really tough, most disease are studied by collecting disease cases (that could come from populations of millions of people) and then comparing them to control healthy persons. This is called a case-control study.
What’s next? Biologic Plausibility: Hill thought that for an agent to be causal for disease, it should make biologic sense that it is capable of doing so. For example, some people are concerned about electromagnetic (EM) radiation from electric poles causing cancer. Biophysicists tell us that this EM radiation is several orders of magnitude too weak to break DNA molecules, causing mutations and therefore cancer. This doesn’t mean that EM radiation does not cause cancer, it is just as of right now there is no biologically plausible mechanism to explain how it could do this. So this is strike against EM causing cancer unless science comes up with a plausible explanation of how this could occur.
Duesberg cites this for HIV as well. Once upon a time in the 1970s, he was a respected retrovirologist (which he and his followers will pummel you with if you let them). He hasn’t done any significant work in the field for nearly 30 years but when he was a cracker-jack, he was right, there weren’t a lot of other retroviruses that could cause diseases like this. But surprisingly enough, science has progressed since 1975 and indeed there are viruses like SIV/HIV-2, which have been found to behave similar to HIV in causing immunosuppression in monkeys and in people. Most of Bialy’s silly questions wash up on the shore here of biologic plausibility and were addressed in my previous comment. Interestingly, biologic plausibility is a minor criteria for causality, and Bialy seems likely to be unaware of Hill’s other major criteria as outlined above.
Epidemiologic coherence: By this, Hill meant, when taken together does all the data make sense? So if HIV were common in South Africa (which it is) but there was no AIDS there, it would not make sense. Gay, bisexual men, intravenous drug users and hemophilics have high HIV rates. Guess what? These are the same groups in the US who have high rates of AIDS. Think of any group, condition or incident and the patterns of HIV infection and AIDS follow each other.
Experimental data: As epidemiologists, we don’t experiment on people, do we? Actually, we do. There is a study design called a clinical trial where you take a group of people, randomly pick half and treat them one way, and either don’t treat the other half or you treat them in another way. Then you find out if they have different outcomes. A blinded trial is where the investigator is purposely kept in the dark as to who is who, until the study is finished. Then the identity of who got treatment A and who got treatment B is compared to their clinical outcome. This is precisely what was done to determine the effect of neveripine, etc. as described above. This is actually a very, very strong piece of evidence if you have a drug or vaccine that can target an agent such as HIV. Since the participants are randomly assigned to receive the drug or not, this evens out differences between individuals if a large enough group is studied.
Phew, okay so that is how scientists have and do determine causality. That was a long one. Still with me? The big point here is that Hill was not dogmatic. He didn’t say you had to have all of these criteria exactly nailed to say that an agent is causal for disease (although they are nailed cold for HIV and AIDS, case closed and has been closed for nearly 20 years). He used a lot of modifiers and adjectives, like “should be associated”, “often is present”, “usually is the case” and so forth. Why? Because the real world is messy and people are messier. Sometimes they lie, sometimes tests are a little inaccurate, sometimes the epidemiologist is an idiotic and mislabels something. Or sometimes people use PCR. Also, notice, that causality cannot be determined by one, single study because no matter how good it is, it won’t have been validated by multiple investigators and thus violates the generalizablity rule.
So, not all studies will be perfectly concordant and not all data will perfectly point to a cause. A.B. Hill was smart enough to realize this. You are suppose to interpret the data, not simply regurgitate it! So, if when you look at all of the different studies on HIV and AIDS, you have to ask how many studies support each one of Hill’s criteria and how many don’t? Can they all be logically reconciled in an unbiased and fair way? Are there studies that are so bad, so stinkeroo (a technical epidemiologic term), that they should be ignored?
Okay, now that you know how scientists approach questions of causality, I’m going to get really perverse. At a more fundamental level, causality is relative. Yes, that’s right, I’m about to throw out everything I said above. Does having unprotected anal intercourse cause AIDS? Yes, but only if the insertive partner is HIV. Does having receptive any intercourse with an HIV positive partner cause AIDS? Yes, but only if you don’t have a specific CCR5 allele, and so forth. For every disease there are a series of causal links. For infectious diseases such as AIDS, these links include the agent (e.g. HIV) but not exclusively. So at a fundamental level, when Michael says that poverty, racism and homophobia cause AIDS, he is absolutely right and a properly designed epidemiologic study can measure these things.
With the possible exception of rabies virus, HIV/AIDS is actually the most clear case where the virus alone is causal for disease. For just about every other infectious disease the case for proving a causal relationship between the agent and the disease is much, much foggier. Epstein-Barr virus (EBV) was discovered in 1964 (hmmm, Epstein-Barr virus was actually found by Epstein and Barr--and Achong who was an excellent Ghanaian scientist but who was conveniently left off of the name the virus. Don’t want to be too facile about racism but this is pretty damning. Barr’s first name is Yvonne so at least sexism wasn’t also at play) in an African lymphoma called Burkitt’s lymphoma. After 42 years, there are still people arguing about whether or not EBV causes Burkitt’s lymphoma. It does, but that is another story. The point is that most infections are asymptomatic and never cause disease. Other risk factors, such as nutrition, previous immunity from related viruses, and so forth, actually determine whether a person develops disease once they are infected. So, these other noninfectious factors are causal for disease as well. HIV is unusual and unusually bad since it seems like to cause disease in a much higher proportion of people infected compared to other viruses.
Now there is a second thing that epidemiologists do not like to talk about: Hill’s criteria for causality are BIASED. They are based on pathoepidemiologic assumptions. The easiest way to see this is the dose-response or biologic gradient criteria. Here he assumes that an agent will be either more likely to be at the site of disease than elsewhere, if the agent causes disease. But, what about a virus or bacterium, let’s call it Agent X, that causes an autoimmune disorder? What if the disease manifestation is results because someone’s immune response is too good and they not only get rid of the invading organism, but they also start attacking their own cells? In this case, the opposite of Hill’s dose-response criteria would occur. People who are sick because of Agent X infection will paradoxically have lower rates of prevalent infection with Agent X than healthy controls. Yeowhhhh!!!
It is possible we will find new infectious agents that cause disease in ways that we haven’t previously thought of. But once these mechanisms are figured out, Hill’s criteria may not be appropriate anymore.
Here comes the third piece of heresy: causality is normative. If you go through Hill’s criteria there is no magic point at which an agent is proven to cause disease. In fact, you can never prove an agent causes disease. You can only prove that it does not. So, eventually enough smart people who know enough about all the evidence, eventually start to say, “hey, guess what? HIV causes AIDS”. And the rest of the world believes them. That’s it. So Peter Duesberg is right: there is a big conspiracy and all of the scientific community and all the physicians are in on it. Drug companies, too. White House as well.
Now that we know something Duesberg doesn’t know, which is how to determine a causal relationship, let’s beat Peter to the punch and steal his thunder.
Come on, just shout it out: “They say that HIV causes AIDS but they C-A-N-N-O-T prove it!” And, truth be told, Duesberg is right and anyone who has seriously thought about the issue knows it. We also cannot prove that poliovirus causes polio, Corynebacterium diptheriae causes diptheria, Bacillus anthracis causes anthrax, etc, and we cannot prove that HIV causes AIDS. Can’t be done so that everyone in the world including Peter Duesberg and his ilk will believe it. The reason is because causality is normative and at some point intelligent people weigh all of the evidence for and against a causal relationship and then they decide. But there is one truth that my medical resident taught me in medical school: “Stupidity is an incurable disease”, and there will always be a far right portion of the bell curve that won’t believe that HIV causes AIDS. But if it saves millions of lives by treating HIV-infected people with antiretroviral drugs and developing a vaccine to stop this pandemic, well, I’m willing to fudge a little and say that HIV causes AIDS. The only blog inspired by a Bumper Sticker.