It’s been about a week since the KSA MOH announced 113 previously unreported MERS cases and 92 deaths, seemingly out of the blue with little comment aside from a decent stacked case count epicurve. Unfortunately, details have yet to be released, leaving those of us who have been maintaining MERS databases with a large chunk of missing information for a very substantial number of cases and deaths… Which has put a little bit of monkey wrench in analytical activities – like comparing case fatality rates across time periods, for instance.
Before the announcement of the 113 previously unreported cases, MERS CFR in Saudi Arabia prior to April 2014 appeared to be substantially higher (~42%) than what we saw from late March through May (~31%). However, with the introduction of the 92 previously unreported deaths, it’s unclear whether or not this still remains the case. Because of this, I thought it might be interesting to see how different distributions of these 92 deaths over time might impact the pre/post-outbreak MERS CFR differential.
To start, I tried my hand at digitizing and recreating the epicurve off of the KSA MOH website using WebPlotDigitizer (HT: David Fisman). Here’s my version, updated through the June 9th MOH media report; the previously unreported cases comprise 16% of the total:
Finally, I developed a simple algorithm to distribute the 92 new deaths proportionally among the 113 new cases. Then, I ran the same algorithm on 14 deaths that I couldn’t match to previously announced cases and distributed them proportionally among all KSA MERS cases. Here’s the death count epicurve that represents this duo of analyses:
With the results of the proportional distribution algorithm in hand, I made two more chartsto visualize case fatality under two scenarios (all cases vs. symptomatic cases only), accounting for both reported and previously unreported cases:
Based off of this proportional distribution, I went on to compare pre/post-outbreak CFRs for both scenarios (all cases vs. symptomatic cases only). I also tried two other distributions – front-loading and back-loading the 92 new deaths, respectively – for both scenarios. Here are the counts, CFRs, and chi-square values across the various distributions, scenarios, and time periods:
It becomes quickly apparent that back-loading and front-loading the 92 new deaths has obvious CFR and chi-square outcomes… But perhaps more interesting is that even if the deaths are proportionally distributed, this year’s outbreak fatality remains significantly lower than the CFR prior to April 2014. Why might this be the case? Can’t say for sure, but what we do know is that the large number of asymptomatic diagnoses these last two months aren’t enough to explain this phenomenon on its own…Needless to say, it’s pretty fascinating. In the coming days, I’ll hopefully be building some Monte Carlo simulations to model more nuanced (read: realistic) distributions of the 92 new deaths. As it stands, it’s still unclear whether or not the CFR from this spring’s outbreak differs substantially from previous months – and if so, in what direction. The reality is that it could go any which way until we have more information… Here’s to hoping that it’ll come soon enough! (At which time, we can see which distribution – if any – actually matches up…)