Thread by @KousiksarathyS

#Thread #WuhanVirus #COVIDー19 #StatePerformance
How do you calculate the real fatality during the course of an ongoing epidemic?
Authors- @MohanRaghavan8 @KousiksarathyS @yashaswinimr23 @IITHyderabad
CC: @AyushmanNHA @ICMRDELHI @PrinSciAdvGoI

That’s easy, some say – it is the fraction of the deceased over the infected!
Not so fast, hold your horses! So, it turns out the amount of time it takes for patients to progress from ‘confirmation of infection’ to ‘recovery’ or ‘deceased’ states is not the same!

Say, on a given day, if ‘c’ confirmed cases are reported, ‘r’ recover and ‘d’ are deceased, the confirmed, recovered and deceased all belong to different cohorts, i.e., were confirmed to be infected on different days. Calculating FR with std methods will lead to high fluctuations

Even within recovered or dead, different individuals likely belong to different cohorts i.e., may have been declared positive at different time points. So, there exists a lag between deceased vs recovered cases!

Our modeling studies show that, as the number of recovered and dead increases sufficiently, the above estimate of fatality rate tends to the true fatality rate with sufficient time and number of cases (more details on bioxiv soon)

But let’s cut to the chase.
Can we find this lag? If yes, then we can factor it out and get better estimates of FR.
Assuming that the FR is a fun of the pathogen – host dynamics & pop genetics of the country, variance in true FR in diff states should be minimal.

It turns out the minimal variance exists at lag of 5 days, i.e., the deceased numbers of 5 days ago should be used with today’s recovery numbers.
So we propose a modified FR measure as follows.

The corrected FRs – FR’ gives us more stable estimates of fatality rates. Check out the table below for FR’s nos for 15 states.
Not difficult to observe some have a FR’ <10, some between 10-20 and some above 20.

States with low FR: These states have enjoyed FR below 10% from over -12 to -5 days. While K'taka has reduced its FR from 20%, UP, Ker & Raj have had low FR all through.

Sudden rise:
DL - FR was stable at 15% between 17 and 12 days ago, but crossed 20% over next 5 days
TG- sudden spurt in FR between 7 and 14 days ago, although its FR was close to zero until then.
AP- abrupt increase starting 12 days ago.

Mah-enjoyed FR close to 10% between 23 days ago and 15 days ago, saw a steady increase between -15 and -7 days.
TN decrease in FRs between -15 and -10 days, but over the next 5 days has seen a sudden increase

Two possible reasons for sudden spurt:
1. Possible those pop clusters were exposed to a higher viral load causing higher fatality.
2. FR’ calculated with ratio of deceased to infected => we only have visibility into the infected and detected cases.

But there may also exist infected but undetected who are deceased-generally revealed during a post-mortem. This may lead to an artificial inflation of FR – only workaround is for @MoHFW_INDIA to ask states to release this data as well.

States with high FRs -need urgent attention !
GJ, PB, MP, JK, WB have been at 20% + levels ever since their numbers gained stability. West Bengal has decreased its FR between -15 and -10 days to a little below 20% as per official figures.

Unlike MH, DL or Telangana, these states have never had a lower stable FR.
They have either had large undetected infections or a peculiar cluster with high exposure and fatality.
Urgent aggressive measures are needed in these states !