I keep doing threads on Mumbai's shocking COVID situation. Another longish one - sorry!
1) Active infection is at eye-watering levels.
2) Everything has happened fast, and things are not yet improving.
3) Modelling can't explain the current surge without new assumptions.
Before any details, a general point about modellers and models. They've got many things wrong during this pandemic. Some embarrassingly wrong. Models keep having to be rewritten. BUT models build and challenge intuition in a way data crunching alone can't - at least for me...
The models didn't find Mumbai's second wave (Sept-Oct 2020) surprising. They said weakening restrictions & uneven spread could cause such a surge. It hit nonslums harder than the first, and slums much less than the first (as expected). It added fewer infections than the first.
But my models did not see the third surge coming. They suggested that full opening up (no more NPIs) would lead to a surge, but limited in scale and fairly minor in the slums. Even despite slow waning of immunity & high slum R0. Only watching Manaus' story unfold was worrying.
Mumbai's surge is anything but slow and limited. Here are some estimates. Right now:

The chances of being infected in a given week are astronomical: >15% if you are a slum dweller and >7% otherwise.

R0 is very high (probably 1.7 or more) and not yet going down.
How much active infection is around? In the last week Mumbai found 63K cases (9K per day). An estimated 16K of these were from the slums. Test +vity was ~20%. Half of tests were rapid - if all had been PCR (as they were early last year) +vity would probably be around 26%.
26% is higher than the average test +vity upto early July, 2020 - over the first wave. During this period only about 0.8% of slum infections and 5.9% of housing society infections were detected. There is evidence detection greatly improved after July when test +vity dropped.
How closely is detection related to test +vity? We don't know. But if we're back at weak first wave detection, then last week saw 2M new slum infections (30% of the slum pop) and 8 lakh new nonslum ones (13% of the nonslum pop). That's hard to swallow, so let's be optimistic...
Suppose detection is 2X higher now despite high +vity (for some reason). Even then, 15% were infected in the slums & 7% in nonslums last week. If infection is "active" for 7 days, it would mean that 10% of the city is actively infected right now. Is this even possible?
Yes. It seems much of the city's rapid testing is random. And RAT +vity is >10%. Remember, 7 of 67 (10%) of vegetable sellers were found +ve at one market using rapid tests? That was March 23 when the situation was less dire.
So there's incredibly high active infection. Which brings us to Point (2): it happened fast and is still not slowing.

Case doubling time is about 9 days right now, and hasn't dropped below 14 days since March 12. That's consistent, rapid growth. But the reality is worse...
...because test +vity has been rising fast, it's very likely that daily infections have been growing faster than cases suggest. There's uncertainty, but my estimates suggest that daily infections have been doubling in less than 7 days for the last 2 weeks. That's R ~ 1.7 or more.
Now, whatever you believe about prior infection in the city (my median estimate: 60%), R ~ 1.7 is shocking. So to point (3) - what does modelling say? I've run several different models of Mumbai over the past year, including a 2 compartment agent-based model (not yet on github).
The 2 compartment model can explain 2020 data. The data was consistent with slums having a much higher R0 (4 to 5) than the nonslum areas (2 to 3). Transmission was slowed by lockdown, but still rapid in the slums. Lockdown largely "uncoupled" the slums and nonslums.
The way slum infections weren't fading away at the end of the year made me introduce (slowly) waning immunity into the model. I believe that understanding transmission at the tail end of 2020 required some waning immunity. But not much, and not rapid. What about this surge?
The model can't reproduce the city's third surge. Decreasing NPIs to zero and allowing low level, steadily waning immunity aren't enough for a surge of this speed and scale. I suspect the data could be reproduced by introducing a much more transmissible or immune escape variant.
I would be curious to know if anyone running any models (deterministic or stochastic, differential equation or agent based...) is able to reproduce the city's data without introducing a more transmissible variant and/or increasing susceptibility (e.g., immune escape).
Finally, what about restrictions and vaccinations? There are now various restrictions in place, and most people are aware of the worrying situation. Fingers crossed, cases & test +vity will start to fall in the next week. Vaccination will reduce deaths in the city...
..but vaccination coverage is just not high enough yet, even amongst the most vulnerable. The huge pool of active infection still has to work its way through households. There are, sadly, a lot of hospitalisations & fatalities still to come.

• • •

Missing some Tweet in this thread? You can try to force a refresh

Keep Current with Murad Banaji

Murad Banaji Profile picture

Stay in touch and get notified when new unrolls are available from this author!

Read all threads

This Thread may be Removed Anytime!


Twitter may remove this content at anytime! Save it as PDF for later use!

Try unrolling a thread yourself!

how to unroll video
  1. Follow @ThreadReaderApp to mention us!

  2. From a Twitter thread mention us with a keyword "unroll"
@threadreaderapp unroll

Practice here first or read more on our help page!

More from @muradbanaji

5 Apr
Herd immunity. It's surprising people are still being simplistic about this concept, even when the COVID-19 situation worldwide is so messy. In what way messy? Well:
- vaccines
- highly uneven spread (and vaccine coverage),
- movements of people, and
- variants.
So what?...
We've also got much more real world data and experience. We've seen arguments using complicated models go wrong: e.g., "the HIT in practice should be much lower than the theoretical value". Contradicted by high prevalence in many places, even before the arrival of variants.
We've also seen wrong arguments based on simple models: e.g. "there should be no major new outbreaks once the theoretical HIT is reached". Definitely not true when you have uneven spread. Easily shown with a two compartment model. Possibly being seen in practice in some places.
Read 8 tweets
5 Apr
When commenting for this piece by @soutikBBC, here are some things I noticed.
(1/4) Almost all regions are showing rapid exponential growth in cases at the same time.
(2/4) Places badly hit earlier (sometimes more than once) are being badly hit again.
(3/4) At the moment, there is no reason to believe that deaths are rising more slowly than expected from the rise in cases- when you take delays into account. Optimism is good, but not misplaced optimism.
(4/4) In districtwise data from @covid19indiaorg you often see the great majority of districts in a state surging together. This is clearest in log plots of recent data: Andhra Pradesh, Maharashtra, Gujarat and Punjab shown. (4 more districts in the next tweet.)
Read 4 tweets
2 Apr
It's hard to discuss COVID-19 in Mumbai without sounding alarmist. Because the situation is alarming. Calculations suggest:
- more people are being infected daily in housing societies than ever before
- transmission in slums could be close to the huge speeds of April-May 2020.
In the last week Mumbai has recorded 43K cases. But this figure is useless if we don't know how many infections are missed. I estimate:

43K cases = more than 7 lakh infections. That's >5% of the city. In one week.

Let's see how. [Notes on the numbers at the end.]
In the last week ~15% of tests returned +ve. (About half were rapid - if all had been more sensitive RT-PCR tests, positivity would be closer to 20%.) We know that as test positivity (TPR) rises, detection drops - the fraction of missed infections rises. Let's track back...
Read 10 tweets
28 Mar
Mumbai #COVID19 update. There has been a sharp rise in testing in Mumbai over the last 4 days. This has stabilised test positivity (weekly averages), but not yet brought it down. About half of the tests are now rapid tests, up from ~30% earlier. (1/6)
Cases and deaths. Cases are doubling roughly every week at the moment (partly, perhaps, about the rise in testing). In the last two weeks deaths have also been rising - doubling roughly every two weeks (the relatively low numbers mean there's a lot of uncertainty). (2/6)
In the past week, the slum epidemic seems to be growing faster - estimated case doubling times have been lower in the slums. This is a worrying trend - we've seen before how a nonslum surge sparks a faster slum surge. (These are estimates based on examining ward-wise data.) (3/6)
Read 6 tweets
26 Mar
#COVID19 in Punjab. Are we seeing effects of widespread circulation of the more lethal "U.K. variant" (B.1.1.7) in Punjab? (thehindu.com/news/national/…).

Yes, I think so. Apparently this variant could be ~64% more deadly (medicalnewstoday.com/articles/covid…).

#thread (1/6)
Fatalities in Punjab seem to lag cases by about 17 days - that's something of an educated guess since there's a lot of noise in the data. Here's the 17-day delayed CFR in Punjab since the start of the year. It is an astonishing picture. What does it show? (2/6) Image
Through January the delayed CFR hovered between 2 and 3. Since late April it's hovered around 5. A *very clear* change. This could be the effect of increased circulation of the more lethal variant. An alternative explanation also needs to be looked at... (3/6)
Read 6 tweets
22 Mar
Data from Mumbai's latest COVID-19 wave is suggesting that reinfections are important and/or a more transmissible variant is circulating. Here's why. (A slightly technical thread, assumptions and possible objections at the end.) (1/n)
First, the argument in brief: current spread is just too fast. The speed is at odds with levels of prior infection in the city and what we know about R0 - the basic reproduction number in the city - based on the earliest availale data.
The current doubling time for daily cases (weekly average) is ~8 days. With TPR also rising sharply, the true doubling time for infections may be shorter. With standard assumptions, we get R = ~1.57. Estimated cases from slums and nonslums give roughly the same R value in both.
Read 16 tweets

Did Thread Reader help you today?

Support us! We are indie developers!

This site is made by just two indie developers on a laptop doing marketing, support and development! Read more about the story.

Become a Premium Member ($3/month or $30/year) and get exclusive features!

Become Premium

Too expensive? Make a small donation by buying us coffee ($5) or help with server cost ($10)

Donate via Paypal Become our Patreon

Thank you for your support!

Follow Us on Twitter!