Skepticism about science and medicine

In search of disinterested science

Trust medical science at your peril: Correlations never prove causation

Posted by Henry Bauer on 2016/06/28

It was a long-known empirical fact that poverty, vagrancy, criminality, and apparently deficient intelligence all correlated with heredity to a considerable extent; they all ran in families and clans. The scientific confirmation that characteristics of animals are passed on from generation to generation, and the Darwin-Wallace explanation of evolution by natural selection of the fittest, made it possible to understand those aspects of human society. It was an obvious, scientifically sound conclusion that human societies could be steadily improved by restricting reproduction of the less fit and expanding the fertility of the fittest. Hence the eugenics movement, promoted by the most progressive, liberal people who were also the best educated, with an apparently justified faith in the reliability of what was at the time the most up-to-date the scientific knowledge (Trust science at your peril: Beware of scientism and political correctness). Those circumstances led to forced sterilization of tens of thousands in America and reinforced Nazis in their doctrines and practices of mass killing of the unfit — Jews, gypsies, homosexuals (Edwin Black, War Against the Weak, 2003).

Only in hindsight did the flaws and errors of the earlier scientific consensus become clear. We now appreciate that environmental and developmental influences can modify heritable traits quite dramatically. “Ill-bred” can be the result of social, economic, environmental factors as much, perhaps even more than any pre-ordained verdict of genetics; and “well-bred” individuals can spring from what might seem the least promising hereditary stock. In other words, the observed correlation between undesired social characteristics and clans was misinterpreted through neglecting the variable of environmental effects.

One lesson to be drawn is that bad science, wrong science, what some even call pseudo-science, can remain the accepted scientific consensus for decades, even in quite modern times, say, the middle of the 20th century. It is unlikely that a mere half-a-century later our societies have become immune from assuming that a mainstream scientific consensus must be true to Nature. Nothing guards our times from treating unjustified, misguided scientific claims as good science.

Unwarranted claims coming from scientists continue to be accepted if they appear minimally plausible and if they are consistent with world-views and vested interests of financial, social, or political powers.

The most sweeping lesson that remains to be learned is that correlations must never be taken as demonstrating a cause-and-effect relationship: there might always be in play an unsuspected variable. One of the earliest axioms taught in Statistics 101 is that correlations never prove causation. The evident correlation between biological kinship and undesirable behavioral traits was not a cause-and-effect relationship.

Many or most people have never learned that basic truth that correlations are not causes. Many others “know” it as a generalization but fail to apply it in specific instances, when an evident correlation could plausibly reflect cause and consequence — just as a genetic basis for undesirable characteristics seemed quite plausible to educated and expert people not so long ago.

Indeed, a large swath of modern medical practices is based on mistaking mere correlations for evidence of causation (“Correlations: Plausible or implausible, NONE prove causation”). For example:

HPV and cervical cancer

The National Cancer Institute offers a great deal of information about this:

Human papillomaviruses (HPVs) are a group of more than 200 related viruses. . . Sexually transmitted HPV types fall into two categories:
— Low-risk HPVs, which do not cause cancer but can cause skin warts (technically known as condylomata acuminata) on or around the genitals, anus, mouth, or throat. For example, HPV types 6 and 11 cause 90 percent of all genital warts. HPV types 6 and 11 also cause recurrent respiratory papillomatosis, a less common disease in which benign tumors grow in the air passages leading from the nose and mouth into the lungs.
— High-risk HPVs, which can cause cancer. About a dozen high-risk HPV types have been identified. Two of these, HPV types 16 and 18, are responsible for most HPV-caused cancers. . . .
>> Cervical cancer: Virtually all cases of cervical cancer are caused by HPV, and just two HPV types, 16 and 18, are responsible for about 70 percent of all cases . . . .
>> Anal cancer: About 95 percent of anal cancers are caused by HPV. Most of these are caused by HPV type 16.
>> Oropharyngeal cancers (cancers of the middle part of the throat, including the soft palate, the base of the tongue, and the tonsils): About 70 percent of oropharyngeal cancers are caused by HPV. In the United States, more than half of cancers diagnosed in the oropharynx are linked to HPV type 16 (9).
>> Rarer cancers: HPV causes about 65 percent of vaginal cancers, 50 percent of vulvar cancers, and 35 percent of penile cancers (. . . .) Most of these are caused by HPV type 16.

The Centers for Disease Control & Prevention offer advice on avoiding HPV cancers:

— Bivalent, quadrivalent and 9-valent HPV vaccines each target HPV 16 and 18, types that cause about 66% of cervical cancers and the majority of other HPV-associated cancers in both women and men in the United States. 9-valent HPV vaccine also targets five additional cancer causing types (HPV 31, 33, 45, 52, 58) which account for about 15% of cervical cancers. Quadrivalent and 9-valent HPV vaccines also protect against HPV 6 and 11, types that cause anogenital warts.
— Quadrivalent and 9-valent HPV vaccines are licensed for use in females and males; bivalent HPV vaccine is licensed for use in females.
What percent of HPV-associated cancers in females and males are caused by the 5 additional types in the 9-valent HPV vaccine?
— About 14% of HPV-associated cancers in females (approximately 2800 cases annually) and 4% of HPV-associated cancers in males (approximately 550 cases annually) are caused by the 5 additional types in the 9-valent HPV vaccine.

What evidence is there for these extremely specific claims of causation?

None, actually. The cited facts are merely that the stated strains of HPV have been detected in those proportions of those cancers. Those correlations don’t begin to indicate causation.

It may be worth recalling that the Centers for Disease Control & Prevention in the early 1990s had officially stated, on the basis of the same sort of data (epidemiology, i.e. correlations), that cervical cancer was an AIDS disease, caused by HIV.

One may sympathize with medical researchers for the impossibility of conducting experiments that would be capable of proving cause-and-effect; ethical, legal, and moral restraints make it unfeasible to use human beings as experimental guinea pigs. There would also be practical barriers: To determine whether a given treatment, in this case a vaccine, actually prevents cancer, a clinical trial would be necessary that spanned over decades and enrolled large numbers of human guinea-pigs, some of whom (controls) would not get potentially-cancer-preventing vaccine.

However, the inability to obtain proof does not justify proclaiming as fact, as these official agencies do, causative relations that are no more than speculation based on statistical correlations.

[The vaccines] “Gardasil and Cervarix have not been shown to be of any significant health benefit. They have been demonstrated to cause serious injuries. It’s scandalous that they were ever approved, and it’s scandalous that they remain on the market.

And they are far from alone on those scores among new prescription medications introduced in the last couple of decades” (Deadly vaccines, 2013/04/17 http://wp.me/p2VG42-24).

Alzheimer’s Disease

Sleep disorders may raise risk of Alzheimer’s, new research shows
Sleep disturbances such as apnea may increase the risk of Alzheimer’s disease, while moderate exercise in middle age and mentally stimulating games, such as crossword puzzles, may prevent the onset of the dementia-causing disease, according to new research to be presented Monday

A daily high dose of Vitamin E may slow early Alzheimer’s disease

Again, these are correlations speculated to be possible causes.

Semantics no doubt plays a role. One could report that sleep disorders, and lack of vitamin E, seem to be associated with a risk of Alzheimer’s. Medical jargon puts it like this: “sleep disorders, and lack of vitamin E, are risk factors for Alzheimer’s”. Then the media and public conclude that “risk factor” means something that tends to cause the associated effect.

See also “60 MINUTES on aging — correlations or causes?

Biomarkers

It is not feasible to test treatments for chronic conditions by actual outcome, because one would have to wait a couple of decades to determine whether regimen A or drug B reduces morbidity and mortality apparently associated with high blood pressure, or high cholesterol, or high blood sugar, or low bone density, etc. All those are statistically correlated with increased morbidity and mortality. They are risk factors.

Present-day medical dogma makes them biomarkers for cardiovascular disease, diabetes, bone fracture, in other words indicators of whether the disease is present. But that is tantamount to making those quantities measures of actual risk, in other words regarding them as measures of what causes those ailments, in other words equating risk factors with causes.

Official reports, however, as well as the many studies on which those reports are based, find that biomarkers are not proper measures of risk after all. See:

“Everyone is sick?”

“‘Hypertension’: An illness that isn’t illness”

“Cholesterol is good for you”

 

Unfortunately, they were not joking

“Magical statistics: Hearing loss causes dementia”

 

The overall lesson:

“Don’t take a pill if you’re not ill”

The ignorant acceptance of correlations as capable of demonstrating causation is greatly reinforced in medical matters by the pharmaceutical industry, which sells drugs as palliatives and preventatives based on nothing more than correlations with biomarkers.

Advertisements

5 Responses to “Trust medical science at your peril: Correlations never prove causation”

  1. Dan Kegel said

    Hi Henry,
    re HPV and cervical cancer, when you say
    “The cited facts are merely that the stated strains of HPV have been detected in those proportions of those cancers.”
    shouldn’t you link to
    “The causal relation between human papillomavirus and cervical cancer”,
    http://www.ncbi.nlm.nih.gov/pubmed/11919208
    ? That’s where the standard argument in favor of HPV causing cervical cancer is laid out.

    • Henry Bauer said

      Dan Kegel:
      That article underscores my point. It is replete with claims that “consistency”, “plausibility” and other such “criteria” can somehow make correlations, associations, epidemiological data, into proof of causation.
      That’s poppycock, to put it mildly and politely.
      As the National Cancer Institute and the CDC point out, HPV is one of the most prevalent sexually transmitted infections, with the large majority of people contracting it at one time or another. It will therefore be associated with innumerable conditions.
      But the main point remains, correlations cannot prove causation, no matter how consistent or plausible they may seem to be. The correlation between kinship and poverty, crime, etc., was both consistent over long periods as well as plausible, but it was mistakenly taken to be causative.

  2. AlzScience said

    Great article. The correlation vs. causation distinction is something more people need to be aware of as misinterpretations can have dangerous consequences. You use Alzheimer’s as an example, a field whose research is often too heavily based on observational studies. I’ve had to explain so many times to people that there’s no proof that higher education prevents AD, for example. Thanks for posting.

  3. Mark said

    I told you, before, that I wish you would stop using such an absolute word like “never.” In my opinion, even “almost never” would be better. I didn’t think that I was asking too much, but you dug in your heels and kept saying “never.” Well, if you won’t even move a little, for me, then maybe you’ll move a little for these people:

    https://statswithcats.wordpress.com/2015/01/01/how-to-tell-if-correlation-implies-causation/

    Whattaya think? Is it, maybe, a good time to admit that there MIGHT be SOME times when correlation DOES prove causality, even if those times are very rare? If you won’t, at least, give me that, then maybe I should leave you with this cartoon:

    http://www.smbc-comics.com/index.php?db=comics&id=1994

  4. spynose said

    Hi Mark,
    Sometimes it is better to add the adverb “almost” to “never”, but certainly not in this case. Because correlation simply cannot implie causation, rather the inverse is the case. It is like one-way traffic. There is a fundamental factor in play: time sequence.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s