When Italian authorities confirmed Wednesday that James Gandolfini had just died in Rome of an apparent heart attack, many of the US reports fronted the fact that Gandolfini’s body would be autopsied, “as required by Italian law.” They fronted this news for understandable reasons—an autopsy on someone who died in medical care seemed unusual. In the United States, we usually don’t autopsy people unless the cause of death is mysterious or foul play is suspected. In fact, we autopsy less than 5 percent of all deaths.

So why would you want to do an autopsy on someone when the cause of death seemed pretty clear? Heart attacks are obvious, right?

You would do so because while death and taxes may be the only sure things in life, the actual why and how of any given death is often uncertain, even when doctors think otherwise. The sad fact is that despite medicine’s many modern wonders—the tests, the drugs, the scans that show fibers the size of a hair—the cause of death that American doctors provide for death certificates are dead wrong about a fifth of the time.

How do we know they’re wrong? Because of so-called autopsy studies. In an autopsy study, researchers perform or collect data from post-mortem dissections of large numbers of dead people; determine definitively why they died; and then compare those findings to the causes of death listed on those people’s medical or death records. These studies reliably find something rather shocking: about 15 to 30 percent of the time, the diagnoses at time of death are wrong—and 5 to 10 percent of the time, that diagnostic error probably helped kill the patient.

This is why Italy required Gandolfini’s autopsy: So the doctors and family would know with certainty why and how he died. As it happens, heart attacks and pulmonary embolisms—clots in a lung—present very similarly: Both cause chest pain, shortness of breath, panic, and can kill you within an hour. Doctors thus often think someone died of a heart attack when they actually died of a pulmonary embolism—so often, in fact, that most pulmonary embolisms are not diagnosed unless an autopsy is done. (We know this, of course, only because autopsy studies showed us it was so.)

This isn’t just arbitrary hindsight information. The discovery of an embolism on the autopsy table can provide vital health-risk information for the rest of the family. A victim’s brother, for instance, might have a chest angiography done to look for fluid in his lungs, and if it’s there, start heparin therapy that could save his life.

About 15 to 30 percent of the time, the diagnoses at time of death are wrong—and 5 to 10 percent of the time, that diagnostic error probably helped kill the patient.

Alas, we’re missing most such learning opportunities these days, because the routine autopsy has pretty much died. Fifty years ago, American hospitals autopsied almost half of all deaths. Every doctor had done them, learned from them, and took their presence for granted. It’s hard to overstate how much we learned from making autopsies routine back then. The list of ailments they discovered runs long, from sudden infant death syndrome to Alzheimer’s, Legionnaire’s disease to toxic-shock syndrome.

In the summer of 1999, for instance, four New York City residents died of a lung ailment that the treating doctors diagnosed as St. Louis encephalitis. The New York City medical examiner’s office, an ardent autopsy advocate, insisted on autopsying them anyway—and identified the first known U.S. victims of West Nile virus.

Doctors, meanwhile, learned humility. If you’re a doctor, it’s one thing to know, as an abstract fact, that 10 or 20 percent of patients who seemed to die of heart attack actually died from pulmonary embolism. It is quite another to have a pathologist dissect one of your patients and tell you No, this woman you took care for 25 years, and whose husband you pass on the street each day, did not die of heart attack; she died of a pulmonary embolism, and that’s probably why she complained of shortness of breath when you examined her three months ago. An angiogram might have spotted it.

My father, a surgeon, trained and practiced in the era of high autopsy rates. Once when I was reading The Maltese Falcon, he remarked that surgery was much like detective work: you get a bunch of clues, must decide which to follow, and eventually commit to a decisive line of action. I opined that while that was true, probably no one ever said anything as dramatic as what Sam Spade says at the book’s crucial moral moment. “I bet,” I said,“ that no one ever says, ’I won’t take the fall for you.”

“The hell they won’t,” my father said, shocking me, for he never cursed in my presence. “You screw up a case up good enough, the pathologist will tell you precisely that.”

She died of a pulmonary embolism, and that’s probably why she complained of shortness of breath when you examined her three months ago.

We now autopsy fewer than 1 in 50 deaths in the U.S, missing 49 out of 50 chances to learn something new. The rates have dropped for a number of reasons, but primarily because the government stopped requiring them for hospital accreditation in 1970.

This will almost certainly remain the case. Congress could change this quickly if it made Medicare payments or hospital accreditation depend on autopsy rates of, say, 25 percent. But no one seems to take interest in such regulation. Thus when doctors make diagnostic mistakes in a dying patient, we usually get no chance to learn from it. Instead, we bury it.

But what about the fancy scans? Can’t they tell?

Not really; not always; and sometimes not even with causes of death you’d think would show up easily in a good scan. Doctors take far too much confidence in scan results, feeling they see everything with certainty. They don’t. As a Florida coroner told me a few years ago, “We get this all the time. The doctors get our report and call and say, ‘But there can’t be a lacerated aorta. We did a whole set of scans.’

“We have to remind them that we held the heart in our hands.”

David Dobbs writes on science and culture for The New York Times, National Geographic,  and other publications, and is working his fourth book, The Orchid and the Dandelion (Crown). He keeps his blog and other work at Neuron Culture.