Sunday, 18 January 2015

Some changes to my Ebola virus disease (EVD) graphs...

To perhaps provide clearer info and to accommodate the changes in the epidemic, namely the reduction in cases and the focus on ridding Guinea, Liberia and Sierra Leone from any and all cases of EVD, I've made some tweaks to my Tableau data visualizations (or dashboards). Briefly...

The dots take their leave.
Was this.

Gone are the dots in my cumulative chart, to be replaced by a third "area under the curve" style graph. 

This brings out the importance of the confirmed cases-more on why that matters later. This week Cedric Moro @Moro_Cedric) asked why we seem to have a relatively large number of suspect and probable cases released in each report World Health Organization situation report (WHO SitRep) or summary (SitSumm). I imagine this is due to the turnaround time once the sample arrives, occasions when results may need to be repeated to confirm strange results, time between seeing a patient and sampling them for Ebola virus testing...but there are probably more obvious reasons. Chime in.


Is now this.

Plot the right data for now.

I'm not an epidemiologist - yes, I know you epidemiologists out there already know that. But I like to play with numbers and pretty colours. So this week I got some information that I didn't have before - the reason why use of cumulative curves was frowned upon by the excellent numbers communicator, Prof Hans Rosling (@HansRosling). 

I had read previously that Prof Rosling was no fan of cumulative curves in graphically explaining progress in ridding west Africa of EVD. But I like them - I've even explained, in my epidemiologically unprofessional opinion - how a flat plateau on a cumulative curve clearly shows the stalling of an outbreak or epidemic. Turns out I either didn't read all of that quote, or the text I read didn't contain the key fact. 

It's not really that cumulative curves are at fault, it's what they are plotting that can mislead. The important thing to plot, especially now that cases are fewer and laboratory capacity is in place, are the confirmed cases, not the total cases which include suspected+probable+confirmed cases.

Confirmed cases are Ebola virus, unconfirmed cases may never be.

In the last WHO SitRep (14-Jan-15) it was noted..
All 54 EVD-affected districts (those that have ever reported a probable or confirmed case) have access to laboratory support within 24 hours of sample collection.
"Access" doesn't mean a result will appear 24 hours after sampling though. But even with this shorter access period, suspected and probable cases are in fact still making up a decent proportion of the total cases reported in even the most recent reports. For example...
  • In Guinea the numbers between 14-Jan-2015 and 15-Jan-2015 saw suspected cases rise by 3, probables stayed the same and confirmed cases lifted by 8; 27% of the total cases reported between this pair of reports were not confirmed to be Ebola virus infections, at the time of reporting. 
  • In Liberia over this period, suspected cases rose by 29, probables by 2 and confirmed case numbers did not change-so none of the 31 cases were laboratory confirmed as an Ebola virus infection. 
  • In Sierra Leone over this period, suspected cases rose by 10, probables remained the same and confirmed cases lifted by 16; 38% of the total cases were not confirmed to be Ebola virus infections.
If we compare those figures to 2 SitReps from well before the WHO had declared the 24 hour laboratory support, dated 24-Sept-2014 and 26-Sept-2014, we find that Guinea only had 8% of its tally unable to be confirmed, Sierra Leone was at 12% not confirmed while 87% of EVD-like cases added to Liberia's tally between reports were not confirmed as due to and Ebola virus infection. 

This may not be a fair comparison of course and it's not one that accounts for every report - just the 2 pairs of reports I arbitrarily chose as being from 'now' and 'back then'. Nonetheless, I expected there to be a bigger and more obvious difference in the proportion of cases that were now being quickly confirmed-I thought that percentage would have gone up as the unconfirmed cases were less frequent. Instead, it seems that the proportion is not that much better. Perhaps this is an indication of the other diseases which mimic EVD early on, that normally emerge at this time of year or have emerged because of the state of healthcare in the countries blasted by the EVD epidemic. As I said above, it may also just be the time it takes to observe, collect a good history and make a clinical decision before a sample is collected. It may also be that laboratory turnaround times (including testing, verifying and reporting) take a bit longer than we naively expect from reading that quote from the WHO above.

More visualizations of confirmed case numbers.

So for the reasons above, I've added the changes I've mentioned and I've also duplicated some of the "total case" graphs by creating versions that only include confirmed cases. 

In the example below I'm showing that it looked like Liberia was experiencing an uptick in cases for 2 consecutive reporting weeks (blue bar graph, right column). I tweeted about this during the week. In fact, those rises were due to unconfirmed cases. The confirmed case plots (green titles in the right-hand column of graphs) show the consistent decline in new EVD cases we had been hearing about. 

Live and learn.

Graphs plotting total EVD cases (including suspected, probable and laboratory confirmed;
brown title bars, left-hand column) 
versus graphs plotting only the laboratory confirmed cases
(green title bars, right-hand column). 

Data are from WHO SitReps and SitSumms
Click on image to enlarge.