Case Fatality Rate for ebolavirus

A case fatality rate (CFR) or case fatality risk is a property of an infectious disease in a particular population which states the risk of fatality due to the disease per case. The first thing to note is that it can't be reliably be calculated for an ongoing epidemic by dividing the reported number of deaths due to a disease by the reported number of cases. There are a number of reasons why this will be a poor estimate: Firstly, this doesn't take into account infections that have yet to run their course (ending in recovery or death). If many new cases are being reported, then this will under estimate the CFR. A more reliable estimate can be made if the number of recoveries is also being reported. Secondly, the estimate will be poor if there is a bias in reporting or diagnosis towards severe cases of the disease. This is the case with MERS-CoV in the Middle East where there are many instances of mild or asymptomatic cases (discovered through contact tracing) but most primary cases are only tested if the patients are hospitalized. This will overestimate the CFR. With ebolavirus, bias may occur if patients are being looked after at home and only being hospitalized or recorded if the disease becomes very severe or if the patient dies. 

Considering the numbers from the latest WHO disease outbreak news:

Total number of suspected and confirmed deaths and cases up to the 26th August 2014.

Country Deaths Cases Ratio
Guinea 430 648 0.66
Liberia 694 1378 0.50
Sierra Leone 422 1026 0.41


This table shows a striking difference in the ratio of reported deaths to reported cases amongst the 3 countries most affected. This might be the result of differences in health care amongst the countries (although there is no specific treatment for ebola virus disease (EVD), general care from an early stage can improve the chances of recovery). However, the other thing to note is that these ratios are in inverse order to the size of the epidemic in number of cases and Liberia and Sierra Leone have had a much higher rate of reporting new cases than Guinea over recent weeks. 

Another way to look at the numbers is to consider only laboratory confirmed cases. This might seem like a good idea as this would remove suspected cases that were caused by other diseases (malaria, Lassa virus, etc) which may be less fatal than EVD. Here is the numbers for only confirmed cases from the same WHO DON:

Total number of EVD confirmed deaths and cases up to the 26th August 2014.

Country Deaths Cases Ratio
Guinea 287 482 0.59
Liberia 225 322 0.70
Sierra Leone 380 935 0.41


This is much more confusing with Guinea dropping to 0.59 and Liberia going up to 0.79. It is difficult to know what is happening here but counties may differ in how they decide which cases to test and it is possible that Liberia is focusing on testing fatal cases or having a greater proportion of cases being reported post-mortem. Plotting these ratios over the last month, see below, shows that Liberia has consistently had a greater proportion of deaths in their confirmed cases suggesting this sort of bias. This ratio has then shot up to 0.9 in early August, dropping to 0.79 in the last couple of reports. 

In July, Guinea had a relatively slow rate of reporting of new cases (1-5 per day), and I suspect that the 0.7-0.75 ratio from that time may be a better reflection of the underlying CFR. As expected, as the numbers of new cases jumped up in mid-August the observed fatality rate dropped. This effect is seen by plotting the rate of new cases per day and the apparent fatality rate on the same graph.

However, until a careful epidemiological analysis is done, one that includes only those cases for which the outcome is known, reports of case fatality rates are unreliable. I would also be very sceptical about claims that this virus is different from previous Zaire-lineage ebolavirus or that it is changing in character over time.

Update 19-Aug-2014

I came across this New York Time article which quotes Medecins Sans Frontieres as reporting "only 61 of the 337 Ebola patients treated at its tent-camp treatment center in nearby Kailahun have survived".  This would estimate a CFR at 0.82 (binomial 95% confidence intervals of 0.77, 0.86). However, this still doesn't say whether this 337 patients number includes patients still being treated. This is in Sierra Leone. A report on the Medecins Sans Frontieres is reporting some different figures. For Sierra Leone it states this:

"In total, MSF treatment centers have admitted 294 patients, of whom 191 were confirmed to have Ebola. Of those, 47 people have recovered and returned home."

Which gives a CFR of 0.84 (0.79, 0.88). However, this suggests this is across all treatment centers and the number is lower than the NYT report about just the one center. 

The MSF report states that "there are currently 50 patients in the center", referring to the Kailahun centre. So perhaps these should be removed from the figures? This would give numbers (taking the NYTs figures) of 61 recovered, 50 still being treated implying 226 fatalities. This would give a CFR of 0.79 (0.74, 0.83). 

For Guinea there is this:

"Since March, MSF treatment centers in Guékédou have admitted 366 patients, of whom 169 were confirmed to have Ebola. Forty-seven patients have recovered and returned home"

The article says that there are 11 patients currently being treated in Guékédou so that would imply 308 fatalities out of 355 resolved cases with an CFR of 0.87 (0.83, 0.90). 

Update 31-Aug-2014

A recent paper in Science, Gire et al "Genomic surveillance elucidates Ebola virus origin and transmission during the 2014 outbreak" contains a table of cases sequenced. See Table S2 in Supplementary Information. Of the 96 cases list, an outcome is known for 77 of them (23 survived and were discharged, 54 died). This gives an estimate for the CFR of 0.70 with 95% confidence intervals of 0.59 to 0.80.

Update 23-Sept-2014

A new paper from the WHO Ebola Reponse Team in New England Journal of Medicine provides the first proper estimate of the CFR from a large sample of cases with known outcomes. It estimates the CFR to be 0.71 (95% CIs 0.69 to 0.73) based on a sample of 1737 patients with definitive outcomes recorded and the estimates are extremely consistent amongst the 3 countries most affected. From the paper:

"Our estimates of case fatality are consistent in Guinea (70.7%), Liberia (72.3%), and Sierra Leone (69.0%) when estimates are derived with data only for patients with recorded definitive clinical outcomes (1737 patients). Estimates for hospitalized patients with recorded definitive clinical outcomes are also consistent across countries but are lower than those for all patients with definitive clinical outcomes. In contrast, simply taking the ratio of reported deaths to reported cases gives estimates that differ among countries (Table 2). These discrepancies perhaps reflect the challenges of clinical follow-up and data capture."

This is extremely similar to the estimate from 77 patients in the Gire et al paper and is consistent with the estimate from Guinea when the number of cases per day was low.