The study, published on Tuesday in the London-based journal Lancet Infectious Diseases, said the toll might have been even higher - as many as 579,000 people.

The original count, compiled by the World Health Organization, put the number at 18,500.

Those were only the deaths confirmed by lab testing, which the WHO itself warned was a gross underestimate because the deaths of people without access to the health system go uncounted, and because the virus is not always detectable after a victim dies.

The new study also shows the pandemic's impact varied widely by region, with 51 percent of swine flu deaths occurring in Africa and southeast Asia, which account for only 38 percent of the world's population.

"This pandemic really did take an enormous toll," said Dr. Fatimah Dawood of the U.S. Centers for Disease Control and Prevention, who led the study. "Our results also suggest how best to deploy resources. If a vaccine were to become available, we need to make sure it reached the areas where the death toll is likely to be highest."

Swine flu, caused by the H1N1 influenza virus, infected its first known victim in central Mexico in March 2009. By April it had reached California, infecting a 10-year-old, and then quickly spread around the world, triggering fears and even panic.

The CDC warned Americans not to travel to Mexico if they could avoid it. Egypt ordered the slaughter of all the country's pigs in a misguided attempt to contain the virus, which was in fact spread from person to person.

The fears reflected the unusual nature of the virus, which contained bits and pieces of bird, swine and human flu viruses, a combination never before detected.

Scientists were unsure how transmissible or deadly this mongrel flu would be, but early signs were ominous: the World Health Organization declared swine flu a pandemic in June 2009, when labs had identified cases in 74 countries.

Such lab-based identification is the gold standard, but every expert acknowledges that it misses more cases than it catches.

One reason is that "some people who contract flu do not have access to health care," said CDC's Dawood, so their illness and even death goes unnoticed by authorities. Another reason is that the virus is not always detectable by the time a victim dies.

LACK OF DATA LOWBALLS FATALITIES

To get around these obstacles, epidemiologists resort to statistical models. They typically take the number of deaths from pneumonia and complications of underlying cardiovascular disease - both caused by influenza - during non-flu periods, count the number during a pandemic, and attribute the excess to the flu.

Unfortunately, "vital statistics data are non-existent or sparse in many lower-resource countries," said Dawood, making this approach infeasible.

Dawood and her colleagues - from Vietnam, Kenya, New Zealand, Denmark and five other countries - tried a different method.

They started with hard data, such as numbers from health workers going door to door in rural villages and asking about flu-like symptoms and testing nasal and throat swab samples, to estimate the proportion of a country's population infected with 2009 H1N1. Such data were available from 13 countries - wealthy, such as Denmark, and poor, like Vietnam.

Then the scientists estimated the fraction of patients who died in each country. They started with solid data on death rates from respiratory illnesses in five wealthy nations.

Since someone with, say, pneumonia has a lower chance of dying if treated in a top hospital in Hong Kong than at a rural clinic in Vietnam, the scientists applied a "multiplier" to the raw data from poor countries.

That is, they assumed that more people with flu-caused pneumonia died in developing nations than developed ones.

These estimates and assumptions can introduce errors, critics note. Newly released mortality data from Mexico, for instance, show that H1N1 killed even more people than the new study estimates, said Lone Simonsen of George Washington University School of Public Health, co-author of a commentary on the study. Estimates of deaths from Japan and Singapore, in contrast, may be too high.

Overall, however, the under- and over-estimates probably even out, said Simonsen, making the global estimate - of 15 times more deaths than those confirmed at the time - about right.

The results paint a picture of a flu virus that did not treat all victims equally.

It killed two to three times as many of its victims in Africa as elsewhere. Overall, the virus infected children most (4 percent to 33 percent), adults moderately (0 to 22 percent of those 18 to 64) and the elderly hardly at all (0 to 4 percent).

Even though the elderly were more likely to die once infected, so few caught the virus that 80 percent of swine flu deaths were of people younger than 65.

In contrast, the elderly account for roughly 80 percent to 90 percent of deaths from seasonal influenza outbreaks. They were probably spared the worst of 2009 H1N1 because the virus resembled one that had circulated before 1957, meaning people alive then had developed some antibodies to it.

The relative youth of the victims meant that H1N1 stole more than three times as many years of life than typical seasonal flu: 9.7 million years of life lost compared to 2.8 million if it had targeted the elderly as seasonal flu does.

H1N1 had begun petering out by November 2009, and the WHO declared the epidemic at an end the following August.

(Reporting by Sharon Begley; Editing by Michele Gershberg and Xavier Briand)

• A Permanent Economic Emergency