LONDON - The next time a headline tells of a sharp fall in measles deaths around the world, or an increase in those on treatment for HIV, or the shifting of the burden of cancer, spare a thought for the number-crunchers behind such far-reaching data.
Above all else, analysing the state of the world's health - be it by looking at obesity rates, cancer cases, malaria deaths, or HIV-free births - requires decent statistics. Billions of dollars are allocated and whole policy shifts made on the basis of figures from United Nations agencies like the World Health Organisation (WHO), UNICEF or the World Bank.
Yet good data are hard to find, as the WHO's statistical analysis team knows. And extrapolating meaningful global figures from sparse raw material can be fraught with danger.
In an interview with Reuters ahead of this week's World Health Statistics report, Ties Boerma, WHO's director of health statistics and information systems, started with a little known but alarming fact: "Two thirds of deaths in the world are not registered. And a third of births are also not registered."
For a team tasked with producing reliable, independent and consistent data, not only on who is dying, but also from what, and where, why and how, that's a tough starting point.
"If you do a global health estimate, you've got to have the data. And getting data is a big problem," Boerma said.
The WHO's annual World Health Statistics reports are billed as a "snapshot of global health" designed to give the most up-to-date picture across all 194 WHO member countries.
The year's report, due on May 16, will give data on everything from rates of measles deaths around the world, to the percentage of women who have no access to contraception, to the number or psychiatrists one country has compared to another.
They are quoted by governments, U.N. agencies advocacy groups and campaigners, and set a standard to compare country with country, disease with disease, and past with present.
But some recent high-profile disputes about some sets of data have focused a spotlight on the way the WHO collects its data and compiles its estimates.
A study by the respected United States Institute for Health Metrics and Evaluation (IHME) published in February found that malaria, for example, kills 1.2 million people a year worldwide - nearly twice as many as WHO figures suggest.
The IHME researchers, whose study was published in The Lancet medical journal, said they thought past studies, including those used by the WHO's statistics team, had overlooked hundreds of thousands of deaths because they wrongly assumed malaria overwhelmingly killed babies and children under five.
Addressing this disparity, Colin Mathers the WHO's coordinator of statistics for mortality and burden of disease, says good assessment tools are essential if health statisticians are going to have even a chance of getting the numbers right.
And those can vary widely from disease to disease. For example with HIV, the human immunodeficiency virus that causes AIDS, there are antibody tests which give clear results, and hence more robust data.
With tuberculosis things are a little trickier. A disease that afflicts mostly the poor and marginalised, it is harder to diagnose and results of tests, if they are done at all, can take days or weeks to confirm.
With malaria, it's harder again, and Mathers says he and other experts have doubts about the strength of IHME's data.
For their study, the IHME said they combined data from national registers with so-called "verbal autopsy" studies, in which researchers interview the relatives of someone who has recently died to identify the cause of death.
"Malariologists would probably argue that verbal autopsy is a pretty poor instrument," Mathers said. "If somebody has a fever and dies, there's a tendency for malaria to be always the first thing that comes to mind."
Then there's the risk that countries might err on one side or the other. There are incentives both for over-estimating - larger numbers of disease cases could attract more funding - and to under-estimating, to demonstrate the effectiveness of a certain policy or programme.
"If you say 20 per cent of deaths are due to unknown symptoms, then of course all your other causes of death are going to be artificially low," Mathers said.
Boerma stresses the WHO uses the best data available and doesn't bend under political pressure. In the end, it publishes its own figures, and the methods by which it reached them, and stands by them even if they are different from data put out by national governments or research institutions.
He is, however, always hungry for academic debate.
"The positive thing about all of this, is that the malaria community is looking at it seriously (the idea that adult death rates may be higher)," he said. "It's not a reason for the WHO to suddenly say its 600,00 or so figure is way too low, but as an academic stimulus to examining the issue, it's a good thing."
As well as the pages of numbers on afflictions ranging from polio to lung cancer, the WHO dedicates a sizeable chunk of its global health report to uncertainty.
Ranges of uncertainty should be clearly published and include not only statistical uncertainty, Boerma says, but also take into account possible systemic biases; for example the extent to which a group of people within a nation may or may not be representative of a nation as a whole.
Hans Rosling, a professor of international health at the Karolinska Institute in Sweden, believes embracing this uncertainty, rather than seeking to banish it, is vital when working with global health data.
"I've spent the last 10 years thinking about nothing else,"he said in a telephone interview. "And yes it's absolutely essential that we continue to collect data and publish it."
Rosling, who trained in statistics and medicine before starting the Gapminder Foundation, a non-profit venture which aims to depict the state of the world in numbers, says we must recognise that some things are easy to measure. He gives women's fertility rates as an example, Others, like malaria deaths, are far more difficult.
"It's not just that good data are hard to find, as that uncertainty is hard to assess," he said.