Opinion, Berkeley Blogs

Thinking under the influenza

By Fyodor Urnov


It is flu season, and an opportunity to, first, remind ourselves that in 1918-19 the virus killed approximately 30 million people, and to take appropriate precautions; second, ponder a general human weakness the disease name reflects.

Symptoms of the flu include fever, aches and pains, headache, and weakness, so why is this condition called “influenza”? The word is Italian for “influence.” The disease was thus named because people thought that its comings and goings were due to the influence of the planets. Our understanding of disease etiology has changed very substantially, but the name stayed. Similarly, “mal’ aria” is Italian for “bad air” (malaria was thought to be caused by the unpleasant smell emitted by marshes).


Why should we care? Plenty of words do not stand for what they actually mean. Our files and music do not live in an actual cloud and visiting this web site will not leave you a physical cookie. Even the atom has the least accurate name possible (“a tomos” is Greek for “indivisible” – tell that to Berkeley’s own Ernest Lawrence!).

Through their names, influenza and malaria teach us about an ancient, and powerful, logical fallacy that humans succumb to. Formally known as “coincidental correlation,” it is classically referred to by the phrase “post hoc, ergo propter hoc.” The lofty Latin means this: if B happened after A, then this must mean that A caused B. A recent tongue-in-cheek example used real data to “prove” that organic food causes autism (for aficionados, the correlation coefficient is 0.9971).

Are we doomed to forever conflate correlation with causation? Not in certain scientific neighborhoods. A marvelous recent effort by scientists and physicians in Japan shows how to stare down “coincidental correlation” and defeat it decisively (a more layperson-oriented description of the work can be found here).

The tragic burden of severe genetic disease can be lessened by carrier screening and prenatal diagnostics; a striking example is summarized in the title of a New York Times article about it: “Using Genetics Tests, Ashkenazi Jews Vanquish a Disease.” For this approach to work, physicians must know where in the DNA of the prospective parents the mutations causing the genetic condition lie. This is not as easy as it sounds; as folks from the wonderful show Radiolab remind us, if you read your DNA, one letter per second, it will take you A CENTURY to get through the whole thing. Our genome is large.

The paper from Japan describes a couple, both of whom are certain to be carriers for a severe genetic condition. It was not clear where in their DNA the causative mutation was. Simple laws of genetics thus gave their future baby a 1 in 4 chance of being affected. Like a magician in The Lord of the Rings, summoning hobbits, elves, Ents, dwarves, and humans to battle, the folks working with this couple brought together a wealth of expertise from different areas of science and medicine to help.

The DNA of the parents was read, and a candidate region – a stretch of the DNA bearing a suspicious mutation – was identified. Accompanied by the theme music from Jaws, the villain of “coincidental correlation” then entered the room: there are thousands and thousands of DNA differences between people – how does one tell whether a condition is caused by, or simply correlated with, the presence of a particular DNA variant?

The solution was provided by a method called “genome editing” – it allows geneticists to take human cells and then change their DNA sequence in a very precise, targeted way. In this case, when the suspicious DNA variant was made in normal cells, they became “sick” (the quotation marks mean that a cell is not a person, so one has to read out its state of “health” using special tools). The scientists and physicians then agreed – the DNA suspect is the actual culprit. The couple chose to have prenatal testing, and the test came back negative (which, in the world of medicine, means “good”). A healthy baby boy was born.

Not all realms of human endeavor allow a study of such experimental rigor (“if A is hypothesized to cause B, then when you get rid of A, and of A only, then B should stop happening"). That said, as we choose courses of action based on less-than-perfect data, we should always ask ourselves: are we acting under the influenza, i.e. the influence of, correlation vs. causation?

The goal here is not paralysis in the face of imperfect data, but a healthy skepticism (perfectly intelligent people concluded that the appearance of Venus in the sky caused everyone to develop a fever). Also, keep in mind that science moves fast: human genome editing is less than a decade old, and here it is, defeating the dreaded "coincidental correlation" in the real world. (Students who want to learn more about such "science in the real world" examples are invited to take a new class, Biology for Voters, this spring)

The conclusion from all this: dum spiro, spero (as long as one is alive, there is hope).