The Ebola virus that is causing the raging epidemic in West Africa is famously lethal. In previous outbreaks it has killed as many as 90% of the people it infects. That’s why the figures in World Health Organization’s (WHO’s) latest “Situation Report” look like they might be a rare glimmer of good news. Although the rate of infections is picking up speed at an alarming rate, the report says the fatality rate is 53% overall, ranging from 64% in Guinea to just 39% in Sierra Leone.

But there’s a catch: The apparent low proportion of deaths probably depends more on the way health officials are calculating the number than on the deadliness of the virus—or the quality of care patients are receiving. Indeed, the dramatic increase in cases in recent weeks is one of the main reasons the reported death rate appears to be artificially low.

There are several ways to calculate what officials call the “case fatality rate,” or CFR, of a disease outbreak. One of the simplest is to divide the number of deaths by the number of total cases. That is what WHO does in its recent CFR calculations.

But that method doesn’t take into account that many living patients—recently diagnosed and very ill—will not survive. So it underestimates the death rate. And that effect is exaggerated when an outbreak is expanding quickly. The calculation also misses patients who were confirmed as Ebola cases, but then left the hospital before being discharged, says Andrew Rambaut, an evolutionary biologist who studies infectious disease at the University of Edinburgh in the United Kingdom. Many of those patients later died but are not counted in the death statistics.

Another way to calculate the rate is to ignore current patients and count only patients who have officially recovered and been released from treatment or who are known to have died. Those numbers seem to paint a more sobering picture. According to the 7 September update from the Sierra Leone Ministry of Health and Sanitation, 268 patients have been treated and released, and 426 confirmed Ebola cases have died. Those numbers suggest a 61% fatality rate. But that isn’t completely accurate either, notes Marc Lipsitch, an epidemiologist at the Harvard School of Public Health in Boston: Survivors may have longer average hospital stays than patients who die. That would lead to a CFR that is artificially high.

A more accurate way to calculate the rate is to compare the outcomes in patients who were infected around the same time and wait long enough until all have either recovered or died. Rambaut notes that there were 23 survivors among the 77 patients included in a recent paper looking at the evolution of the virus. That’s a CFR of 70%.

Christopher Dye, director of strategy for the WHO, says the organization is moving toward that method and is working to compile data for each patient recorded as a case. “We do need valid estimates,” Dye says. “We want to know if CFR is different in this epidemic from previous ones in central Africa, [and] whether different approaches to patient care in the current epidemic lead to different outcomes.”

Even that method is imperfect. In almost all outbreaks, cases are missed because the patient never seeks care at a health facility, and therefore is not recorded in any statistics. Such missed cases can potentially bias the CFR in either direction, Lipsitch notes. If many cases are relatively mild—in which infected people recover without ever seeing a doctor—then relying on health care records overstates the death rate. (That was the case for the H1N1 pandemic flu in Mexico, and experts suspect it is the case for the MERS virus as well.) A mild case of Ebola is less likely to go unnoticed than a mild case of influenza, Lipsitch says, but given the overall lack of health care in the region, there could be significant numbers of undetected survivors.

On the other hand, researchers already know that many Ebola victims never made it to hospitals and died at home (often infecting family and other caregivers). That means their deaths aren’t counted—reducing the CFR.

Exactly how many unrecorded Ebola deaths have occurred will never be known. Health officials are keeping track of suspected and probable cases, many of which are people who died before they could be tested. Whether to include those numbers in CFR calculations is another source of potential bias. And there are different patterns of testing in different regions: Some places have done more testing on post-mortem cases, for example. “How these biases balance is always the big question,” Lipsitch says.

“We are not naïve about the difficulties of estimating CFR,” Dye writes in an e-mail. “I’m not yet ready to believe … that CFR is much higher in Guinea than Sierra Leone. This is what the data say, taken at face value, but we need to exclude all possibility of ascertainment bias before believing this to be the truth.”

*The Ebola Files: Given the current Ebola outbreak, unprecedented in terms of number of people killed and rapid geographic spread, Science and Science Translational Medicine have made a collection of research and news articles on the viral disease freely available to researchers and the general public.