Correlation or Causation? Our Search for Certainty

Julian L. Seifter, MD


March 10, 2015

When we clinicians diagnose disease, we inevitably are involved in questions of cause and effect. You can't treat effectively without understanding pathogenesis and etiology—that is, causation—and causation is often the first thing that comes up in office visits.

Most patients, confronted with an illness, ask the question, "Why?" Or, even more painfully, "Why me?" Despite these questions, they may already have arrived at their own theory of causation, very often involving temporal sequence. There's a fancy name for this in Latin: post hoc ergo propter hoc— After this, therefore because of this."

Consider the very anxious Russian physician who came to me complaining of edema in his legs, which he felt came on after drinking a bottle of Bulgarian brandy a month earlier.

"Did you eat Russian salami with that?" I asked.

"No," he said. "Better! I had some big pickles and marinated tomatoes..."—and, as it turned out, a long list of other salty foods, all favorites of his. Sensing that he might have some prejudice involving Bulgarians and their brandy, I mildly suggested that it was the diet high in salt, not the brandy, that caused the edema.

Of course, it is not necessarily "something I ate." It's irresistible to make an association between two events that occur together, and often these associations are right—but sometimes they're wrong. Distinguishing correlation from cause depends on the quality of the evidence and the strength of the association.

The search for cause is a serious enterprise because falsely assigning cause can do significant harm. The recent outbreak of measles at Disneyland had its origins in an antivaccine movement led by those who believe that vaccination causes autism. The association was proposed in a poorly designed study published in the Lancet in 1998 (and retracted in 2010) that led tens of thousands of parents to leave their children unimmunized.

From the "Sippy diet" for ulcers and the radical mastectomies of 50 years ago to today's news about caffeine or dietary cholesterol or mercury in fish, we clinicians are always revising our ideas about association and our theories of causation in light of new evidence.

Sometimes, new evidence affirms traditional ideas. We in medicine often suspect home remedies because they don't propose cause, merely association, but occasionally an evidence-based study comes along that supports a bit of folk wisdom. I recently read an abstract of a study that suggested an association between cardiac inotropy and a substance in beans, thereby confirming something I learned 50 years ago at overnight camp: "Beans, beans, they're good for your heart; the more you eat the more you..."

Establishing a definitive cause can be difficult because circular events don't have a clear point of origin. The very notion of homeostasis—A causes B affects A—depends on feedback loops, making it difficult to identify a "prime mover." What is primary and what is secondary? What is the independent vs the dependent variable? It's possible to get it completely backwards: We used to think that renal vein thrombosis (RVT) caused nephrotic syndrome; now we know that nephrotic syndrome causes RVT.

Many difficult patients present with "chicken and egg" scenarios. Someone comes in urinating a lot: Is it diabetes insipidus making them thirsty, as they suspect? Or is it primary polydipsia—in other words, is the patient drinking a lot because he read somewhere that drinking 8 glasses of water a day was a good idea in order to stay hydrated? A psychiatrist I treated who thought he must have diabetes, given the volume of urine he was producing, turned out to be just one thirsty psychiatrist.

Clinicians cite behavior as a cause at our own peril. If you say anything about habits or diet, patients can feel defensive because then it's their "fault." They would rather have their kidneys be at fault.

Russian salami or excessive drinking may, in fact, be the root cause of a problem. But association alone doesn't prove a causal connection.

What does all of this mean for us as clinicians? We search for the etiology of disease, but often we can't assign an ultimate cause. The recent finding that two thirds of all cancers seem to be due to random mutations in tissues where cell turnover is frequent suggests that the intense focus on the causes of various cancers may not yield the answers we long for. Some cancers have environmental or genetic triggers, but the majority do not. This is inherently unsatisfying; we want to know the cause so we can treat, or prevent, or eradicate a disease.

In the classic article "The Environment and Disease: Association or Causation?"[1] Sir Austin Bradford Hill outlines the aspects of an association between two variables that might legitimately lead to the assumption of a causal connection: strength of the association, followed by consistency, specificity, temporality, biological gradient, coherence, experiment, and analogy. In the case of temporality, sequence can be an indicator of cause, but Sir Austin points out that sometimes the cart and the horse get reversed. Psychosis often follows experimentation with recreational drugs, but does pot smoking induce schizophrenia, or does the onset of schizophrenia induce pot smoking?

What about the patient who adamantly refuses a spinal tap when presenting with a number of neurologic symptoms because (in his own post hoc ergo propter hoc) he blames an earlier spinal tap for all of the problems he had in the past year?

Teasing apart apparent causes from actual causes requires thinking through all of the available evidence. At the same time, clinicians should not be paralyzed by epistemologic uncertainty. Particularly in the case of occupational or preventive medicine, Sir Austin believes that it's not always necessary to sit around waiting for the results of extensive research; a few links in the causal chain may be sufficient to take action. In 1965, the strong association between smoking and lung cancer was sufficient evidence for the Surgeon General to issue a warning to the public.

If this wasn't already complicated enough, two theories of causation are often cited in the diagnostic literature. Of the two, the more appealing to medical sleuths may be Occam's razor, which posits that the simplest explanation is the most reliable. According to this view, if we could just get to the bottom of it, we'd find a single disease at the root of all of the patient's symptoms. Saint's triad proposes just the opposite: three different findings can sometimes mean three separate disease processes. Or, as we sometimes say on rounds, "A man can have as many diseases as he damn well pleases."

The search for cause, among doctors and patients alike, is in some sense a search for control. The whole concept of luck, good or bad, is alien to the medical enterprise.

Doctors, like their patients, want to know why. If someone is an outlier—surviving a usually fatal cancer or falling victim to a very rare disease—we want to get to the bottom of the puzzle. I think most doctors emphatically don't want to be helpless bystanders; they don't want to be merely consolers and allies, however important these roles are. They want to cure, or prevent, or outright "stomp out" disease. Even—and maybe especially—when causes are complex and outcomes uncertain.


Comments on Medscape are moderated and should be professional in tone and on topic. You must declare any conflicts of interest related to your comments and responses. Please see our Commenting Guide for further information. We reserve the right to remove posts at our sole discretion.